Tag Archive | Digital Publishing

Inside the AI-Assisted Newsroom — How the Workflows Come Together

By Cheval John | Vallano Media
This article was created with the assistance of artificial intelligence.
The final version was reviewed, edited and fact-checked by the author


Graphics created with AI-assisted design tools.
Visual concepts, editorial direction, composition, and final creative decisions were developed by Vallano Media.

When a Story Becomes a Newsroom Assignment

One of the most interesting things about building an AI-assisted newsroom is discovering that the technology itself is only one part of the operation.

A newsroom needs more than the ability to generate articles.

It needs research, editorial direction, verification, criticism, organization, and decisions about what deserves publication.

For an independent publisher like Vallano Media, those responsibilities ultimately belong to one person: me.

Artificial intelligence has given me opportunities to organize those responsibilities differently.

Instead of approaching every assignment as one continuous writing exercise, I can divide the work into distinct editorial functions, use AI assistance where appropriate, and evaluate the results before moving forward.

But what does that actually look like?

One of the clearest examples came from an article I developed about Indiana University football in November 2026.

The article was titled “The Twilight Zone of Indiana Football: When Winning Is No Longer Enough.”

At first glance, it was a college football feature examining the transformation of a program whose success had dramatically changed expectations.

Behind the finished article, however, was an editorial process involving research, drafting, verification, criticism, revision, and publication preparation.

That assignment provides an opportunity to explain how my AI-assisted newsroom operates—not as a theoretical model, but through an actual publishing experience.

A Football Story With an Unusual Premise

Indiana football presented an interesting editorial opportunity because the program’s transformation created a striking contrast between its historical identity and its new expectations.

For decades, Indiana was better known nationally for basketball than football.

Yet the football program’s rise under Curt Cignetti had changed the conversation.

The story was no longer simply about whether Indiana could win games.

It was about what happens when extraordinary success becomes the standard against which future performances are judged.

That was where the “Twilight Zone” concept entered the picture.

I wanted the article to capture the strange feeling of watching a historically struggling football program enter an era when winning could somehow seem insufficient.

The premise offered room for humor, historical perspective, and serious analysis.

But a creative premise is not the same thing as a verified article.

The challenge was to preserve the unusual framing while ensuring that the football history, game details, financial information, and conclusions could withstand editorial scrutiny.

That required more than asking AI to write an entertaining story.

It required an organized process.

Research Comes Before the Narrative

Before developing the manuscript, I needed to understand the evidence supporting the article.

The research examined Indiana’s football history, the program’s transformation, Curt Cignetti’s leadership, the university’s financial commitments, and the game developments that made the story timely.

The Minnesota game provided the opening narrative.

Rather than beginning with a general discussion about Indiana’s rise, I wanted to bring readers directly into a football situation that illustrated the program’s changed expectations.

That decision gave the article a recognizable sports journalism structure: begin with a compelling event, establish its significance, and then expand into the larger story.

AI assistance helped organize research questions, examine information, and develop the relationship between the immediate game and the broader argument.

However, research assistance did not eliminate the need to examine sources.

The distinction matters because an AI-generated explanation can sound authoritative even when a statistic, historical claim, or financial figure is incorrect.

For that reason, the research process could not end with a plausible answer.

It had to establish which information was supported, which claims required additional checking, and which interpretations belonged to the analysis rather than the factual record.

The research provided the foundation.

The editorial decisions determined what to build on top of it.

Developing the First Draft Without Losing the Story

Once the research foundation was established, the next challenge was developing the manuscript.

The goal was not simply to summarize Indiana football’s success.

I wanted to explain the changing relationship between achievement and expectation.

The “Twilight Zone” framing helped connect those ideas.

It also created a potential problem.

A creative concept can become so dominant that the reporting begins serving the metaphor instead of the other way around.

That was something the editorial process needed to guard against.

I established a clear direction for the first draft: preserve the Minnesota opening, maintain the unusual tone, and develop the central argument about building success versus sustaining it.

Once that draft existed, I treated it as the working master.

This was important because AI-assisted editing can sometimes introduce changes that weaken an article’s original direction.

An assistant might suggest a more conventional introduction, remove an unusual phrase, or reorganize the story into a familiar explanatory format.

Those suggestions are not automatically improvements.

In this case, the unusual framing was part of the article’s identity.

My responsibility was to determine which revisions strengthened the reporting without sacrificing the story I intended to tell.

That is one of the differences between using AI as a writing assistant and directing an AI-assisted editorial workflow.

The objective is not to accept every improvement the technology proposes.

It is to decide what improvement actually means for the assignment.

Verification Is a Separate Responsibility

The next stage involved examining factual claims.

The Indiana article included different categories of information, each requiring appropriate verification.

Game results and statistics needed to match reliable sports records.

Historical claims needed to reflect the program’s actual development.

Financial information concerning coaching compensation and institutional commitments required particular attention because contractual figures can change and may be reported differently depending on the date or terms involved.

One example of source-based verification involved a 60-yard field goal.

During the reporting process, we consulted the official play-by-play published by Indiana University Athletics to examine the game details and establish the correct description of the kick as a program record.

This demonstrated why a journalist cannot rely exclusively on AI-generated descriptions.

The underlying evidence must support the factual claim.

The distinction between a career-long field goal and a program-record field goal might appear minor, but it changes the meaning of the information.

A career record belongs to an individual athlete.

A program record belongs to the institution’s competitive history.

The example illustrates why verification needs its own purpose.

The drafting process asks whether the story communicates an idea effectively.

Verification asks whether the factual claims are supported by evidence.

Those questions overlap, but they are not interchangeable.

Nor does asking several AI assistants to confirm the same statement automatically establish its accuracy.

Different systems can repeat the same mistake, particularly when they rely on overlapping information.

The supporting evidence remains essential.

In this case, consulting the university’s official play-by-play was part of that process.

The Red Team Challenges the Argument

After factual verification, the article needed another kind of examination.

This was the responsibility of my editorial Red Team process.

The purpose of Red Team review is not to praise the manuscript or produce a more polished version of the same argument.

It is to challenge the work.

For the Indiana football feature, the review was intended to examine whether the article’s central thesis was adequately supported, whether the “Twilight Zone” framing became excessive, and whether the manuscript repeated ideas unnecessarily.

It also meant questioning the relationship between the evidence and the conclusions.

For example, the fact that a football program has experienced extraordinary success does not automatically prove that its supporters have become unreasonable or that every narrow victory represents disappointment.

Those interpretations require careful framing.

The article needed to distinguish observable developments from commentary about their meaning.

Red Team review provided an opportunity to identify weaknesses before publication.

But the process was not an independent editorial authority.

Its criticisms were recommendations for examination.

I still had to decide which concerns were valid, which changes were necessary, and which suggestions would weaken the original story.

A useful Red Team process should make the publisher think more carefully.

It should not replace the publisher’s judgment.

When Different Editorial Functions Work Together

The Indiana assignment illustrates something important about the structure of my AI-assisted newsroom.

Research, drafting, verification, and Red Team review are connected, but they do not perform the same function.

Research helps establish the available information.

Drafting organizes that information into a narrative.

Verification examines factual support.

Red Team review challenges the reasoning and presentation.

Editing addresses clarity, structure, repetition, and voice.

Quality control examines the manuscript before publication.

These responsibilities can be supported by different AI assistants, prompts, or workflow stages.

That does not mean they operate as independent employees.

It also does not mean every stage functions autonomously.

In practice, the process depends on human direction, the information supplied, the questions asked, and the decisions made after reviewing the responses.

The stages are not always completed in a perfectly straight line.

A verification concern may require additional research.

A Red Team criticism may send the manuscript back for revision.

An editing change may introduce a factual ambiguity that needs another check.

The value of organizing the work into separate functions is that each stage has a defined question to answer.

Instead of asking AI whether an article is simply good, I can examine whether it is accurate, complete, logically sound, readable, and appropriate for publication.

That creates a more disciplined editorial conversation.

It does not guarantee a perfect article.

But it gives me a structure for identifying and addressing different kinds of problems.

When Editorial Recommendations Conflict

An organized editorial workflow does not mean every AI-assisted review will reach the same conclusion.

A drafting assistant might recommend preserving a particular sentence because it strengthens the narrative.

A Red Team review might question whether the same sentence goes beyond the available evidence.

A voice review might recommend retaining its original tone.

These recommendations can conflict because the reviews are examining different aspects of the manuscript.

That does not necessarily mean one assistant is malfunctioning.

It means the publisher must evaluate the recommendations according to the article’s purpose and the available evidence.

When the disagreement concerns a factual claim, the underlying source takes priority over an AI-generated interpretation.

When it concerns style or narrative structure, the decision requires editorial judgment.

The objective is not to make every assistant agree.

It is to reach a defensible editorial decision.

Protecting the Writer’s Voice

Another challenge appears when an article has passed factual and analytical review but no longer sounds like the story I intended to write.

This is especially important in an article such as the Indiana football feature.

The “Twilight Zone” concept was not a decorative headline attached to an otherwise conventional football report.

It shaped the tone and the way I wanted readers to understand the program’s transformation.

The story needed to communicate the almost surreal contrast between Indiana’s traditional football reputation and its new competitive reality.

At the same time, it needed to remain grounded in journalism.

That balance required editorial attention.

AI can produce grammatically polished language while removing the personality, qualifications, or thought process that made an article distinctive.

A sentence can become smoother without becoming better.

A paragraph can become shorter while losing an important distinction.

A transition can sound professional while no longer reflecting the writer’s reasoning.

This is why I have incorporated an Authentic Voice and Thought-Flow Review into my broader editorial approach.

The review examines whether the manuscript still communicates the intended argument in a recognizable voice.

For the Indiana article, preserving the original premise mattered as much as improving the language around it.

The final manuscript needed to sound like my journalism, not like an AI system had replaced my editorial perspective with a generic sports analysis.

The Final Decision Belongs to the Publisher

Eventually, the editorial process reaches a point where additional AI responses cannot make the publication decision.

That responsibility remains mine.

For the Indiana football article, I had to determine whether the research was sufficient, whether the verified information supported the narrative, whether the Red Team concerns had been addressed, and whether the manuscript was ready for readers.

I also had to consider whether the article fulfilled its original purpose.

A technically accurate manuscript can still fail to communicate the intended story.

Likewise, a compelling narrative can still contain unsupported claims.

The final editorial decision requires balancing those considerations.

At Vallano Media, AI assistance does not transfer responsibility for publication to a machine.

The business publishes under its own identity.

That means I remain accountable for the material that appears on the website, regardless of how many AI-assisted stages contributed to its development.

The Indiana article was not complete simply because the drafting assistant had finished writing.

It was complete when I determined that the editorial work had reached an acceptable standard for publication.

From Manuscript to Published Article

Even after editorial approval, the assignment was not finished.

The article still needed publication materials.

That included graphics, tags, website formatting, and scheduling.

These activities may appear separate from journalism, but they are part of operating a digital publication.

The headline needs to communicate the story accurately.

The visual presentation should complement the article rather than misrepresent it.

The website structure needs to make the manuscript readable.

The publication process also needs to preserve the approved text rather than introduce unintended changes.

For an independent publisher, these responsibilities do not disappear simply because AI can assist with them.

I still need to organize the material, review the presentation, and ensure that the published version reflects the approved manuscript.

The Indiana football feature also became the foundation for a LinkedIn adaptation, which I published on Thursday, November 12, 2026.

The adaptation allowed me to examine the original subject from a different perspective, connecting the football program’s transformation to broader questions about leadership, expectations, and organizational success.

That was a second editorial product built from the same underlying subject.

It required a different framing because a sports feature intended for general readers and an analysis intended for business professionals do not necessarily answer the same questions.

Repurposing content responsibly means adapting the interpretation while preserving the underlying evidence.

The existence of that second publication does not, by itself, demonstrate greater audience engagement or improved editorial quality.

It demonstrates how one researched subject can support more than one editorial product when the adaptation remains subject to appropriate review.

What the Workflow Cannot Guarantee

Organizing editorial responsibilities does not eliminate the limitations of artificial intelligence.

Different AI assistants may repeat the same factual error, particularly when they rely on similar information.

A Red Team review can overlook a weakness.

A verification process can fail to identify an important missing source.

Even an extensive review process can create the appearance of rigor without demonstrating that the resulting journalism is consistently more accurate.

There is also the possibility of excessive review.

If an article moves through too many stages without a clear editorial purpose, the process can become unnecessarily complicated.

Additional AI responses may introduce new problems rather than resolve existing ones.

These limitations are important because the Indiana football assignment represents one publishing experience.

It demonstrates how I organized the work and applied editorial judgment.

It does not establish that the same process will produce equally strong results for every assignment.

That question requires continued observation, documentation, and evaluation.

An AI-assisted newsroom should be willing to examine its own weaknesses with the same skepticism it applies to the stories it publishes.

What the Indiana Article Taught Me

Looking back at the Indiana football assignment, the most important lesson was not how quickly the article could be produced.

That is a separate question, and one I intend to examine more closely as I accumulate additional production records.

The more immediate lesson concerned organization.

The article began with a creative editorial idea.

It developed through research and analysis.

The manuscript then passed through factual examination, critical review, editing, and final approval before publication preparation.

Each stage addressed a different responsibility.

And throughout the process, I remained the person determining what the article should accomplish.

That experience helped demonstrate the practical distinction between having access to AI tools and operating an AI-assisted newsroom.

A collection of tools can produce responses.

A newsroom requires an editorial process that determines how those responses are evaluated, revised, accepted, or rejected.

The Indiana assignment also revealed that this system depends on continued human involvement.

The workflows did not eliminate my responsibilities.

They changed how I organized and carried them out.

That distinction is essential to understanding the operation I am building.

Building a Newsroom Around Decisions

When I first began experimenting with artificial intelligence, much of the attraction came from discovering what the technology could do.

Over time, the questions changed.

Instead of simply asking whether AI could write an article, I began asking how different forms of assistance could support the responsibilities involved in producing journalism.

The Indiana football feature offers one example of that transition.

It demonstrates how research, verification, criticism, editing, and publication preparation can become connected parts of an organized workflow.

It also demonstrates why the publisher cannot surrender editorial judgment simply because the process contains multiple AI-assisted stages.

The newsroom does not become stronger merely because it has more assistants.

It becomes more organized when each function has a purpose, each important claim can be examined, and every publication decision remains accountable.

For me, that is the next stage of building an AI-assisted newsroom.

The technology helps carry out the work.

The editorial process connects the responsibilities.

But the decisions that define the journalism remain human.


Graphics created with AI-assisted design tools.
Visual concepts, editorial direction, composition, and final creative decisions were developed by Vallano Media.

The AI Newsroom Needs an Editor, Not Just More Tools

By Cheval John | Vallano Media
This article was created with the assistance of artificial intelligence.
The final version was reviewed, edited and fact-checked by the author


Graphics created with AI-assisted design tools.
Visual concepts, editorial direction, composition, and final creative decisions were developed by Vallano Media.

When the Experiment Starts Growing

There comes a point when experimenting with artificial intelligence creates a new problem.

You have more tools, more workflows, and more possibilities than you originally anticipated.

At first, that sounds like progress.

But having more capabilities does not necessarily mean having a better system.

That is something I began considering as I developed AI-assisted workflows for Vallano Media.

What started as learning how to communicate with ChatGPT gradually evolved into something more structured.

I began assigning responsibilities to different workflows, experimenting with research methods, developing verification procedures, and finding ways to repurpose content for multiple publishing platforms.

Eventually, I was no longer thinking only about what artificial intelligence could do.

I was thinking about how to manage what I had created.

That is an important distinction.

Using AI is one activity.

Managing an AI-assisted operation is another.

For an independent media company, that difference can determine whether technology becomes a productive resource or another source of unnecessary complexity.

The Managing Director’s Responsibility

At Vallano Media, my responsibilities extend beyond writing articles.

As managing director, I oversee editorial direction, business development, content production, publishing decisions, and the company’s broader strategy.

Artificial intelligence can assist with many of those responsibilities.

But assistance should not be confused with authority.

For example, I can use an AI workflow to organize research material, examine a draft for weaknesses, or suggest ways to present an article to a different audience.

Those activities can support my decision-making.

They do not make the decisions for me.

I still determine which stories deserve attention, which sources are credible, which arguments are supported, and whether an article meets the company’s editorial standards.

The same principle applies to business operations.

An AI assistant can help evaluate an idea, organize information, or identify potential risks.

But it cannot assume responsibility for financial decisions, client relationships, or the long-term direction of Vallano Media.

That responsibility remains with the person managing the business.

For me, the challenge is learning how to delegate tasks without surrendering accountability.

The Difference Between an Assistant and an Employee

As my collection of workflows expanded, I began describing some of them in terms of newsroom roles.

A research assistant might help organize sources.

A verification workflow might examine factual claims.

A red-team process might challenge the reasoning behind an article.

Thinking in terms of roles helped me organize the operation.

However, there is an important limitation to that comparison.

These are not human employees.

They do not independently develop professional judgment through reporting experience.

They do not establish personal relationships with sources.

They cannot be held professionally accountable in the same way a journalist or editor can.

And they should not be treated as independent authorities simply because they produce confident answers.

Not every workflow is an autonomous agent, either.

Some involve structured conversations with language models.

Others may support more automated functions, such as monitoring information or issuing notifications.

The distinction matters because the level of human involvement can vary from one process to another.

A newsroom role is useful only when its responsibilities are clearly defined.

A research workflow should not be treated as a source of verified facts merely because it produces a detailed summary.

A red-team review should not be treated as a final editorial decision.

And an AI-generated quality assessment should not be confused with independent evidence that an article is accurate.

Each workflow needs boundaries.

Without those boundaries, a collection of AI assistants can create the appearance of organizational structure without providing the accountability that a real newsroom requires.

Why More Workflows Are Not Always Better

One of the attractions of artificial intelligence is that it makes experimentation relatively easy.

If I identify a problem, I can explore whether a new workflow might help solve it.

That flexibility has been valuable.

But it also creates a temptation to develop a new system for every task.

As my collection of workflows expanded, I began considering whether some of them might eventually overlap or create unnecessary complexity.

That possibility does not mean the workflows have failed.

It means managing them requires the same discipline as developing them.

Several workflows might perform similar functions.

Instructions may overlap.

One process may introduce information that another accepts without sufficient scrutiny.

Eventually, coordinating the tools could consume some of the time they were supposed to save.

That raises an important question.

Does every workflow need to exist?

The answer should depend on whether it serves a useful purpose.

If two workflows accomplish essentially the same task, combining them may make more sense than maintaining both.

If a process repeatedly produces unreliable results, it may need to be redesigned or abandoned.

And if a task can be completed more effectively through direct human judgment, there is little reason to complicate it with artificial intelligence.

The objective should not be to create the largest possible collection of workflows.

It should be to build a system that performs necessary work reliably.

Making Every Workflow Justify Its Place

Creating a workflow is relatively easy compared with determining whether it consistently delivers value.

For me, that evaluation should begin with five questions.

What specific problem is the workflow designed to solve?

Does it produce reliable results when checked against appropriate evidence?

How much time does it require, including the time spent reviewing and correcting its output?

Does it introduce risks or additional responsibilities that outweigh its benefits?

And would the task be completed more effectively without artificial intelligence?

These questions do not require every workflow to produce an immediate financial return.

Some exist to strengthen editorial review, organize information, or identify potential errors.

Their value may be measured differently.

But every workflow should have a clear purpose.

If it creates more confusion than clarity, it needs to be revised.

If it duplicates another process without adding meaningful value, it may need to be combined with that process or discontinued.

This is an evaluation approach I want to continue developing at Vallano Media, not a claim that every existing workflow has already undergone a formal performance audit.

Every experiment should teach me something.

But only the workflows that demonstrate continuing value should become permanent parts of the operation.

Editorial Control Is More Than Final Approval

One lesson from my experience is that editorial responsibility cannot be reduced to clicking an approval button at the end of the process.

A journalist must remain involved throughout the work.

That includes evaluating the original idea, reviewing source material, questioning assumptions, and determining whether the final article accurately represents the available evidence.

This is particularly important when multiple AI-assisted stages are involved.

Imagine that an AI research summary contains an incorrect claim.

The drafting workflow incorporates that claim into an article.

A subsequent editing workflow improves the language without recognizing the underlying factual problem.

The result may be a polished article built around inaccurate information.

Adding another AI review does not automatically solve the problem.

The journalist must return to the evidence.

A language model can help identify factual claims requiring investigation, but it cannot establish their accuracy simply by producing a convincing explanation or citing a source.

The journalist must examine the underlying evidence, determine whether the source actually supports the claim, and consider whether important context has been omitted.

Verification requires more than finding information.

It requires evaluating what that information establishes.

The Limits of One-Person Oversight

Artificial intelligence can help an independent journalist generate ideas, organize research, and prepare drafts more quickly.

But faster production creates another challenge.

The journalist must still have enough time to evaluate the resulting work.

If AI-assisted workflows produce more material than I can responsibly review, the operation has not necessarily become more efficient.

It may simply have created a larger editorial backlog.

That is why managing an AI-assisted newsroom also requires knowing how much work to accept.

Some stories may need to be delayed.

Others may require additional research.

Certain assignments may be unsuitable for the available resources.

The ability to generate content does not create an obligation to publish it.

As managing director, my responsibility is not to maximize the number of articles produced.

It is to determine how much credible journalism the operation can realistically support.

That also means recognizing that editorial capacity is not the same as technical capacity.

A language model may generate several drafts in the time it takes a journalist to verify one complicated story.

The additional drafts are not necessarily additional completed journalism.

They remain unfinished work until the necessary reporting, verification, and editorial decisions have been completed.

For an independent publisher, knowing those limits is part of responsible management.

Accountability Does Not End at Publication

Even a carefully reviewed article can contain mistakes.

The presence of multiple AI-assisted editorial stages does not guarantee that every error will be identified before publication.

If a significant mistake reaches readers, the responsibility to address it remains with the publisher.

That means examining the evidence, correcting the information when necessary, and being transparent about material changes.

It also means reviewing the process that allowed the mistake to occur.

Was the original source misunderstood?

Did an AI-generated summary introduce unsupported information?

Did the verification stage fail to examine the underlying evidence?

Or did the final editorial review overlook a warning?

The purpose of asking those questions is not simply to assign blame to a tool.

It is to improve the process.

An accountable newsroom must be prepared to correct its work, including work produced with artificial intelligence.

For Vallano Media, this is an important standard to continue developing as the publishing operation grows.

A correction should not be treated merely as an inconvenience.

It is part of maintaining a trustworthy relationship with readers.

Managing the Workflow Like a Business Operation

An independent newsroom also needs to consider the resources required to produce its work.

Time matters.

So do operating expenses, software costs, and the effort required to maintain editorial standards.

For Vallano Media, AI-assisted publishing is connected to a broader business strategy.

Sports journalism remains the company’s flagship editorial product, while AI workflow integration for small businesses is being developed as its primary revenue opportunity.

That means the workflows I use internally serve an additional purpose.

They provide an environment where I can experiment with AI-assisted operations before considering how similar approaches might serve other small businesses.

However, an internal workflow that supports my publishing operation is not automatically a validated commercial service.

Client needs, implementation costs, reliability, training requirements, and measurable results must be evaluated separately.

Different businesses have different needs, resources, risks, and operating procedures.

A workflow designed for publishing a sports article may not be appropriate for managing customer inquiries or organizing administrative information.

The underlying principles may transfer.

The implementation must still be evaluated.

My company’s experience can provide a starting point.

It cannot substitute for evidence of what works in another business.

That is why I believe the first question should be what problem a business needs to solve, rather than which AI tool it should purchase.

Technology should follow the operational need.

Not the other way around.

Measuring What Actually Works

One of the next challenges for Vallano Media is developing better ways to evaluate the workflows themselves.

It is relatively easy to count how many workflows have been created.

That number says very little about their effectiveness.

A more useful evaluation would consider total production time, the amount of human correction required, the reliability of factual claims, and the value delivered to readers or clients.

Those measurements need to account for the complexity of the work.

A short article based on straightforward source material is not equivalent to an investigative report requiring interviews, extensive documentation, and multiple rounds of verification.

The same performance expectations should not automatically apply to both.

The approximately 43-minute production window I recorded for the opening installment of Volume One on November 1st, offers one example.

That benchmark included drafting, truth review, red teaming, editing, quality control, graphic preparation, and publication.

It excluded research time.

The result showed that I could complete several editorial production stages within a relatively short period.

But it did not establish that every future article could be produced at the same speed or that the process would be equally effective across different subjects.

A more complete evaluation would consider total working time, including research, alongside accuracy, corrections, and the complexity of each assignment.

The objective is to understand what the system actually accomplishes.

Not simply what it appears capable of doing.

Knowing When to Stop

There is another responsibility that comes with managing an AI-assisted operation.

Knowing when not to use AI.

Not every task requires automation.

Not every problem needs another workflow.

And not every technological capability deserves a place in the business.

Sometimes the best decision is to complete a task manually.

Sometimes the best decision is to revise an existing process rather than create a new one.

And sometimes the most valuable contribution artificial intelligence can make is helping identify a problem that still requires human investigation.

Those decisions are part of management.

They require understanding both the capabilities and the limitations of the technology.

For an independent journalist, that understanding is especially important because the same person may be responsible for reporting, editing, publishing, and managing the business.

There is no large organizational structure to absorb every mistake.

The responsibility remains concentrated.

The Newsroom Needs Leadership

As I continue developing Vallano Media, I am beginning to see artificial intelligence less as a collection of individual tools and more as part of an operating system that requires direction.

That does not mean every activity should be automated.

It means the workflows should serve a defined editorial or business purpose.

They should be evaluated, revised, and sometimes discontinued when they no longer provide meaningful value.

The technology may help expand what an independent journalist can accomplish.

But expansion without discipline can create as many problems as it solves.

A newsroom needs standards, evidence, accountability, and leadership.

Those responsibilities do not disappear because artificial intelligence becomes more capable.

They become more important.

Building an AI-assisted newsroom has taught me that leadership is not measured by the number of tools I can deploy.

It is measured by the decisions I make about how those tools are used.

The goal is not to manage more AI assistants.

The goal is to build a better media operation.


Graphics created with AI-assisted design tools.
Visual concepts, editorial direction, composition, and final creative decisions were developed by Vallano Media.

Beyond the Experiment — From Writing a Book to Building an AI-Assisted Newsroom

By Cheval John | Vallano Media
This article was created with the assistance of artificial intelligence.
The final version was reviewed, edited and fact-checked by the author


Graphics created with AI-assisted design tools.
Visual concepts, editorial direction, composition, and final creative decisions were developed by Vallano Media.

The End of One Volume, the Beginning of Another

Fifteen chapters.

Fifteen days.

One writing challenge that became something larger.

When I began the November 2026 nonfiction writing challenge, I wanted to revisit an experience from 2014, when I participated in a similar challenge without the assistance of generative artificial intelligence.

Back then, writing required a different process.

Research, outlining, drafting, editing, and revising were largely separate activities.

Each demanded time, concentration, and personal discipline.

Twelve years later, I approached the challenge with a different set of tools and a different understanding of how technology could support my work.

Artificial intelligence had become part of my daily routine.

Instead of treating it as a machine that simply generated paragraphs, I had begun developing structured workflows to assist with research, writing, verification, editing, and publishing.

The first fifteen chapters of “Journalist and AI” gave me an opportunity to put those methods into practice over a sustained period.

But something else happened along the way.

I began recognizing the connection between producing a book with AI assistance and building an independent media operation that could use many of the same methods.

The lessons were no longer limited to writing.

They were becoming lessons in newsroom management.

What Volume One Taught Me

One of the most important lessons from Volume One was that artificial intelligence does not eliminate the need for editorial discipline.

In many ways, it makes that discipline more important.

A language model can produce a draft quickly.

It can suggest a structure, identify potential weaknesses in an argument, and offer alternative ways to explain a complicated subject.

But speed does not establish accuracy.

A well-written paragraph can still contain an incorrect date, an unsupported claim, or a misleading interpretation.

That is why my writing process developed into several distinct stages.

The first stage involves drafting.

The next examines factual claims.

Another challenges the reasoning and identifies weaknesses.

Subsequent stages refine the writing, evaluate its quality, and prepare the chapter for final approval.

The process is designed to prevent a polished AI-generated response from being mistaken for finished journalism.

I also learned that working with AI requires knowing when to disagree with it.

There were moments when the system’s suggestions needed clarification, correction, or rejection.

Sometimes an explanation was too broad.

Sometimes a proposed sentence did not accurately reflect my experience.

Other times, the writing sounded polished but lacked the perspective I wanted to communicate.

Those moments were reminders that the journalist must remain engaged.

The value of AI assistance is not measured by how often the journalist accepts its suggestions.

It is measured partly by whether the process helps the journalist produce stronger, more accurate work.

And that cannot be determined by speed alone.

Forty-Three Minutes: A Different Writing Experience

On November 1, 2026, I completed the production process for the opening installment of “Journalist and AI” in approximately 43 minutes, from the initial draft through publication on Vallano Media.

That figure did not include research time, which I tracked separately.

The production process included drafting, editorial verification, red-team review, revisions, quality control, graphic preparation, and publication.

For me, the experience demonstrated how much my working methods had changed since participating in a nonfiction writing challenge in 2014.

I was no longer approaching each editorial task as an entirely separate activity.

Instead, I had a structured process that allowed me to move from one stage to another while using AI to support specific functions.

However, the 43-minute result needs to be understood in context.

I did not conduct a controlled comparison between my 2014 and 2026 writing processes.

I cannot claim that AI made me a particular percentage more productive.

Nor does completing an article quickly prove that the finished work is accurate or valuable.

The experience demonstrated something narrower but still meaningful: “a structured AI-assisted process helped me complete several editorial tasks within a recorded production window while retaining responsibility for the finished work.”

The time was measurable.

The quality still required judgment.

That distinction would become increasingly important as I considered how the same approach might support Vallano Media’s broader operations.

The Difference Between Writing and Managing a Newsroom

Writing a book and operating a newsroom may appear to be different activities.

In practice, they share several responsibilities.

Both require planning, research, organization, editing, fact-checking, and decisions about what deserves publication.

The difference is that a newsroom must repeat those activities across multiple stories, subjects, and deadlines.

That distinction became increasingly relevant as I developed Vallano Media.

My company began as an independent media venture, and its development has required me to think beyond individual articles.

I have had to consider how stories are selected, how information is gathered, how claims are verified, how content is distributed, and how the operation can become financially sustainable.

Artificial intelligence offers opportunities to support those activities, particularly for a small business with limited personnel and resources.

But it also introduces new responsibilities.

If an AI system summarizes a source incorrectly, who catches the mistake?

If it introduces a fabricated quotation, who prevents that quotation from reaching publication?

If several AI-assisted stages repeat the same unsupported assumption, how does the editorial process identify the problem?

The answer cannot simply be to add more AI tools.

A newsroom needs a process that defines what each tool is supposed to do, what evidence it must use, and when human intervention is required.

And even with those procedures, mistakes remain possible.

Building a Newsroom Without a Traditional Staff

Traditional newsrooms divide responsibilities among reporters, editors, researchers, producers, photographers, and other specialists.

An independent journalist often performs several of those functions personally.

That creates an operational challenge.

The journalist must produce stories while also managing the business that makes publication possible.

For Vallano Media, I began experimenting with AI-assisted workflows that could support different editorial functions.

Some were designed to help organize research.

Others examined factual claims, challenged arguments, improved drafts, or evaluated whether a piece was ready for publication.

Over time, I began thinking about these workflows as functional newsroom roles.

That does not mean artificial intelligence has become a staff of independent journalists.

An AI system does not possess the professional accountability, firsthand reporting experience, or independent judgment of a human reporter.

These systems are tools operating within a process established and supervised by a human editor.

They can assist with defined tasks, but they do not assume responsibility for the resulting journalism.

As the managing director of Vallano Media, I remain responsible for editorial decisions and final publication approval.

The final editorial decision remains mine.

There is also an important distinction between developing AI-assisted newsroom capabilities and operating a fully established newsroom.

A functioning media organization needs more than software and editorial workflows.

It needs reporting relationships, reliable sources, corrections procedures, publishing standards, audience trust, and a business structure capable of supporting the work.

AI can contribute to those activities.

It cannot replace the need to build them.

From Individual Prompts to Repeatable Systems

Early experimentation with generative AI often begins with a simple request.

A user asks a question, requests a summary, or asks the system to write something.

That can be useful, but it is not necessarily a reliable workflow.

A workflow requires more structure.

It begins with a defined purpose and continues through a sequence of actions that can be repeated, reviewed, and improved.

For example, an AI-assisted editorial process might begin by gathering source material.

The next stage identifies claims requiring verification.

Another examines the argument for weaknesses.

A later stage improves the presentation without changing the underlying facts.

Finally, the journalist reviews the completed work and decides whether it is suitable for publication.

Each stage serves a different purpose.

This separation matters because asking one language model to write, verify, critique, and approve its own work in a single interaction creates obvious risks.

The system may overlook its original mistakes or reinforce assumptions introduced earlier in the process.

But separating tasks does not automatically eliminate those problems.

Several AI systems can repeat the same incorrect information, especially when they rely on similar sources or assumptions.

A language model agreeing with another language model is not independent verification.

Factual claims still need to be checked against reliable evidence, including original documents, official records, interviews, and other appropriate sources.

The objective is not to create a system that never makes mistakes.

It is to establish procedures that make mistakes more visible and easier to correct before publication.

When the Workflow Becomes the Problem

There is another lesson I have had to consider.

More steps do not always mean better results.

An editorial process can become so complicated that managing the workflow consumes time that could otherwise be spent reporting, researching, or developing a story.

AI can also produce convincing material that requires substantial correction.

A red-team review may identify a weakness without providing a reliable solution.

A factual review may flag a questionable claim but still require the journalist to locate the original evidence.

Sometimes the most useful decision is to reject an AI-generated suggestion and return to the source material.

That is not a failure of the process.

It is part of exercising editorial judgment.

The challenge is to determine which steps provide meaningful value and which introduce unnecessary complexity.

A workflow should earn its place in the newsroom.

It should not be preserved simply because it has been created.

The Business Reality Behind the Technology

There is another dimension to building an AI-assisted newsroom that receives less attention than the technology itself.

The business must survive.

Producing articles, developing editorial systems, and experimenting with new technology all require time and resources.

An independent media operation must also consider expenses, revenue, audience development, and the value it provides to readers or clients.

For Vallano Media, that means developing two connected areas of work.

One is sports journalism and digital publishing, which provides an opportunity to demonstrate the company’s editorial methods through actual reporting.

The other is helping small businesses understand and implement practical AI workflows.

The AI workflow services are intended to become the company’s primary revenue source, while sports journalism remains its flagship editorial product.

That is a business strategy, not proof of commercial success.

Faster content production does not automatically create revenue.

A sophisticated workflow does not guarantee that a business will attract clients.

And producing more articles does not necessarily mean that readers will find them more useful.

The company must demonstrate value beyond the technology itself.

For me, that means continuing to develop editorial work that readers can trust while building services that address real operational needs.

These activities serve different purposes, but they share an underlying principle: technology should solve a real problem rather than become an end in itself.

What Fifteen Days Actually Demonstrated

Completing fifteen chapters demonstrated that I could maintain a structured AI-assisted writing process over a sustained period.

It did not establish that the process was more accurate than traditional editorial methods.

Nor did it prove that the same approach would succeed in every reporting situation.

Those questions require additional evidence.

A workflow should be evaluated not only by how quickly it produces a finished article, but also by the accuracy of the published information, the corrections required, the resources consumed, and the value delivered to readers.

For Vallano Media, the challenge is to move beyond demonstrating that AI-assisted production is possible and begin evaluating whether the methods are consistently reliable and economically practical.

That is the difference between experimenting with technology and building an accountable media operation.

The first fifteen chapters represented an important stage in that process.

They provided practical experience, revealed limitations, and reinforced the importance of maintaining editorial control.

But they were not the finish line.

They were the foundation for the next set of questions.

The Next Phase

Volume One examined the development of my relationship with artificial intelligence and the lessons that emerged as I began using it more systematically.

Volume Two will move further into the practical consequences of that relationship.

What does it mean to operate an AI-assisted newsroom as an independent publisher?

How should editorial workflows be evaluated?

When does automation improve journalism, and when does it introduce unnecessary risk?

How can a small media company use these technologies without sacrificing accuracy, credibility, or its own identity?

And perhaps most importantly, how does a journalist maintain professional independence while relying on systems developed and operated by technology companies?

These are not questions that can be answered by generating a few impressive demonstrations.

They require sustained practice, careful evaluation, and a willingness to change methods when the evidence demands it.

The first fifteen chapters represented one stage of that process.

The next stage begins with a different understanding of what I am building.

I am not simply experimenting with artificial intelligence to see whether it can help me write faster.

I am examining whether carefully designed AI-assisted workflows can support a credible, independent, and sustainable media business.

The experiment showed me what was possible.

The next challenge is determining what is reliable, sustainable, and worth building.

That is where Volume Two begins.


Graphics created with AI-assisted design tools.
Visual concepts, editorial direction, composition, and final creative decisions were developed by Vallano Media.

Scaling Without Losing Control

By Cheval John | Vallano Media
This article was created with the assistance of artificial intelligence.
The final version was reviewed, edited and fact-checked by the author


Graphics created with AI-assisted design tools.
Visual concepts, editorial direction, composition, and final creative decisions were developed by Vallano Media.

Eventually, the question changes.

At first, I was trying to figure out how to use AI responsibly.

Then I was trying to build a workflow I could trust.

Then I wanted to make that workflow repeatable.

Then measurable.

But once a system starts working, another question appears:

“What happens when I try to do more with it?”

More stories.

More platforms.

More research.

More graphics.

More automation.

More clients.

More coverage.

More people.

More responsibility.

That is where scaling begins.

And that is where control can start to slip.

Growth Creates a New Kind of Risk

A workflow that works for one article does not automatically work for ten.

A process that works when I am personally watching every step may not work the same way when tasks are automated, delegated, or running at the same time.

Growth changes the system.

That matters.

The danger is not only that the work becomes larger.

The danger is that the distance between the decision and the result becomes larger too.

When I am doing everything myself, I can see almost every handoff.

When the system grows, some of those handoffs become harder to see.

That is where mistakes can hide.

Scale Is Not the Goal

There is an assumption underneath a lot of technology conversations that bigger is better.

More output.

More automation.

More users.

More clients.

More reach.

But not everything worth doing needs to be scaled.

Some work is valuable precisely because it remains slow, deliberate, or highly personal.

Some reporting deserves hours of attention.

Some investigations should resist automation.

Some projects may never become efficient in the traditional sense.

That does not make them failures.

So I do not want to scale for the sake of scaling.

The goal is “responsible capacity”.

The ability to do more where doing more actually creates value.

Scaling Is Not Just Doing More

It is easy to think scaling means increasing output.

More articles.

More posts.

More clients.

More revenue.

More automation.

Those things may be part of it.

But real scaling means something more difficult:

“Increasing capacity without losing the standards that made the system valuable in the first place.”

If I double the output and double the errors, I have not scaled well.

If I automate more tasks but lose visibility into how decisions are being made, I have not improved the system.

If I grow faster than I can verify, review, or understand the work, I have created a larger problem.

Scale should expand capacity.

Not expand confusion.

And scale should never become the excuse for producing more work that matters less.

The Workflow Has to Grow With the Work

A small workflow can survive informality.

I can remember what I did.

I can keep notes in my head.

I can recognize which version is final.

I can remember which source I trusted.

That becomes harder as the volume grows.

So the workflow has to become more explicit.

Not necessarily more complicated.

More explicit.

What happens first?

What gets verified?

What requires human approval?

What can be automated?

What cannot?

Where does the work stop if something goes wrong?

Who owns the final decision?

Those questions become more important as the system grows.

More AI Means More Oversight, Not Less

This may sound backwards.

AI can reduce the amount of manual work I have to do.

But greater AI integration can increase the importance of oversight.

Why?

Because more tasks can happen faster.

More output can move through the system.

More mistakes can move with it.

A bad assumption inside one manual task may affect one piece of work.

A bad assumption inside an automated process can be repeated across many outputs.

That means the cost of a weak checkpoint can increase as the workflow expands.

Automation changes the scale of both success and failure.

The Human Role Changes Again

Earlier in this book, I wrote about the human in the workflow.

At small scale, that might mean reviewing a draft.

Checking a source.

Approving a graphic.

Choosing a headline.

At larger scale, the human role becomes more about designing and supervising the system itself.

What tasks should AI handle?

What should remain manual?

Which decisions require approval?

What warning signs should stop the process?

What happens when something fails?

How much work can I realistically review?

That is a different kind of responsibility.

The human is no longer only editing the output.

The human is managing the conditions under which the output is produced.

But there is an important limit.

“Human oversight does not scale simply by adding more work to the same human.”

Eventually, the system has to account for the limits of human attention too.

Control Has to Be Designed In

I do not want control to depend on me noticing that something feels wrong.

That works sometimes.

It does not scale.

Control has to be part of the workflow.

That means checkpoints.

Verification.

Clear approval stages.

Version control.

Documented standards.

Fallback procedures.

Visible automation.

The ability to stop the process.

But there is another trap here.

A system can look controlled on paper and still fail in practice.

A dashboard is not control.

A checklist is not control.

An approval button is not control.

Those things matter only if they change what happens when something goes wrong.

The goal is not the appearance of oversight.

It is functioning oversight.

Automation Should Have Boundaries

One of the easiest mistakes is to automate something simply because it can be automated.

That is not enough.

The better question is:

“Should this task be automated?”

Some tasks are good candidates.

Repetitive monitoring.

Formatting.

Organizing information.

Routine reminders.

Basic transformations.

Some tasks require more caution.

Source evaluation.

Interpretation.

Final publication decisions.

High-risk factual claims.

Anything where context matters more than speed.

The boundary should not be based only on technical capability.

It should be based on consequence.

Agents Make This Even More Important

Depending on how they are designed, AI agents can monitor information, trigger actions, move information between steps, prepare material, and perform recurring tasks.

That can create real capacity.

But agents can also create distance.

A task may run without me actively initiating every step.

That makes visibility important.

I need to know:

What triggered the action?

What information did it use?

What did it produce?

What happens if it is wrong?

Can I stop it?

Can I review what happened?

Can I recover?

An agent should not become a black box simply because it saves time.

There is also another risk:

“autonomy creep.”

A system might begin with an agent monitoring information.

Then it starts summarizing.

Then recommending.

Then triggering another action.

Each expansion can seem small by itself.

Eventually, the agent may be doing far more than the original workflow intended.

That is why boundaries need review.

Authority can expand gradually even when nobody explicitly decided to expand it.

Scale Can Hide Weaknesses

Small systems often reveal their problems quickly.

When I personally touch every part of the workflow, I notice friction.

When the system expands, some problems become less visible.

Maybe an automation works most of the time.

Maybe a prompt slowly starts producing weaker results.

Maybe a source changes its format.

Maybe a workflow depends too heavily on one tool.

Maybe a shortcut becomes normal.

The system can keep producing output while quality quietly declines.

That is why measurement matters.

Chapter 14 was not separate from scaling.

It was preparation for it.

If I do not know how the workflow performs at small scale, I will have a harder time understanding what happens when it grows.

And averages are not enough.

A system can perform well most of the time and still produce one failure serious enough to matter.

Average performance can hide individual failures.

Standardization Helps — Until It Does Not

Scaling usually requires some standardization.

Templates.

Checklists.

Naming systems.

Reusable prompts.

Defined workflows.

Those things make repetition easier.

But there is a limit.

Journalism is not manufacturing.

Stories are different.

Sources are different.

Risks are different.

Some assignments need more reporting.

Some need more verification.

Some need legal review.

Some need speed.

Some need patience.

The workflow should create consistency without forcing every story into the same shape.

A standardized process should make judgment easier to apply.

It should not replace judgment with procedure.

The System Needs Exceptions

A mature workflow needs rules.

It also needs a way to break them deliberately.

Not casually.

Deliberately.

If a story requires more verification, the workflow should allow it.

If an automation fails, I should be able to switch to a manual process.

If a source cannot be confirmed, publication should stop.

If a project does not fit the standard template, I should be able to adapt.

Exceptions are not always signs that a system failed.

Sometimes they are evidence that judgment is still present.

Growth Can Create False Confidence

This is another danger.

A system works repeatedly.

I become comfortable with it.

The output looks good.

The process feels familiar.

That familiarity can turn into confidence.

Then confidence can turn into assumption.

That is when I may stop asking questions.

The workflow worked yesterday.

The source was reliable last time.

The automation usually works.

The AI normally gets this right.

That is how oversight can weaken.

Past performance can justify confidence.

It should not eliminate verification.

Scale Magnifies More Than Errors

A small error repeated once is small.

A small error repeated 100 times is not.

That is one of the biggest differences between manual work and automated work.

A formatting mistake can multiply.

A bad prompt can multiply.

A wrong assumption can multiply.

A weak source can multiply.

A misleading headline pattern can multiply.

But factual mistakes are not the only things that repeat.

Framing can repeat.

Source preferences can repeat.

Assumptions can repeat.

Blind spots can repeat.

Bias can repeat.

The more repeatable the system becomes, the more important it is to make sure the thing being repeated is worth repeating.

Efficiency multiplies whatever is already inside the system.

Good or bad.

Capacity Has a Limit

AI can expand capacity.

It does not eliminate limits.

I still have only so much attention.

So much energy.

So much time.

So much ability to review.

That means there is a point where additional output stops being useful.

If I can generate 20 drafts but only responsibly review five, then 20 drafts are not capacity.

They are backlog.

Scaling has to account for the human bottleneck too.

That may be the most important bottleneck in the entire system.

The Review Capacity Has to Scale Too

If production grows, review capacity has to grow with it.

That could mean:

better tools,

better organization,

stronger checkpoints,

clearer priorities,

fewer unnecessary outputs,

or eventually more people.

But there has to be enough review capacity to match the amount of work flowing through the system.

Otherwise AI simply creates more material than I can responsibly supervise.

That is not scale.

That is overload.

Adding People Changes the Workflow

Eventually, scaling may involve other people.

Editors.

Writers.

Researchers.

Designers.

Clients.

Collaborators.

Adding people can increase capacity.

It also creates more handoffs.

And every handoff creates another place where assumptions can develop.

Who verified this?

Who approved it?

Which version is final?

Who owns the decision?

Was this checked already, or did everyone assume someone else checked it?

Once more people enter the system, the workflow has to become understandable outside my own head.

Other people need to know:

What the standard is.

What AI can be used for.

What needs verification.

What requires approval.

What should be documented.

What happens when there is uncertainty.

What happens when something goes wrong.

A workflow that only makes sense to the person who created it is difficult to scale.

Documentation Becomes Infrastructure

At small scale, documentation can feel optional.

At larger scale, it becomes infrastructure.

Not because every process needs a giant manual.

It does not.

Too much documentation can become its own form of friction.

The goal is to preserve important knowledge.

Not bury the workflow under paperwork.

Useful documentation can preserve:

standards,

decisions,

workflows,

sources,

prompts,

exceptions,

lessons from failures.

It turns individual knowledge into organizational knowledge.

That matters whether the organization has two people or 200.

A Fallback Has to Actually Work

I can say I have a fallback.

That does not mean I do.

If an automation breaks, can I really continue manually?

If a tool disappears, do I know another way to perform the task?

If a source feed stops working, do I know where the original information comes from?

A fallback that nobody remembers how to use is not really a fallback.

Resilience has to be practiced enough that it remains real.

Tool Dependence Becomes a Business Risk

The more useful a tool becomes, the easier it is to build around it.

Heavy dependence on a single tool can become a business risk.

What happens if the price changes?

The features change?

The model changes?

The company changes its policies?

The service goes down?

The tool disappears?

And using several tools does not necessarily mean the system is truly independent.

Different products can still rely on the same underlying provider, infrastructure, or model family.

A workflow can look diversified while still containing a shared point of failure.

The workflow should belong to the work.

Not to the vendor.

Tools can change.

The principles should survive.

Information Risk Grows Too

As workflows become larger, more information moves through them.

Documents.

Sources.

Client information.

Internal notes.

Drafts.

Credentials.

Potentially sensitive material.

That means scaling is not only about productivity.

I also have to think about what information enters the system, who can access it, where it moves, and whether it belongs there in the first place.

More capability creates more responsibility for how information is handled.

Scaling Costs Money Too

There is another side of growth that productivity discussions sometimes ignore.

Cost.

More tools.

More subscriptions.

More storage.

More automation.

More people.

More training.

More maintenance.

More time managing the system.

AI can lower some costs.

It can create new ones too.

A workflow is not scalable simply because it produces more.

The economics have to make sense.

If the cost of maintaining the system grows faster than the value it produces, something needs to change.

Complexity Is a Tax

Every tool adds something.

It may also add:

another login,

another interface,

another failure point,

another subscription,

another set of settings,

another thing to monitor.

Complexity accumulates quietly.

At first, every addition seems useful.

Then the system becomes difficult to understand.

That is why simplicity matters even more as the workflow grows.

The goal should not be to use every tool available.

The goal should be to use the smallest system that can reliably do the work.

Scale What Works

I do not need to scale everything.

That is important.

Some experiments should remain experiments.

Some workflows should be retired.

Some automations should stay small.

Some tasks should remain manual.

The things worth scaling are the things that have demonstrated value.

That means I should know:

Does this improve the work?

Does it save meaningful time?

Does it reduce friction?

Does it protect quality?

Can I maintain it?

Can I explain it?

Can I recover when it fails?

If the answer is no, making it bigger will not fix it.

Slow Growth Can Be Good Growth

There is pressure in technology to move quickly.

Adopt faster.

Automate faster.

Scale faster.

That does not always make sense.

Sometimes slow growth is the responsible choice.

Build one workflow.

Test it.

Measure it.

Fix it.

Then expand.

That may look less impressive than building a giant AI system immediately.

But it creates something more valuable.

Understanding.

I want to know why the system works before I depend on it.

Scaling Vallano Media

For me, this is not theoretical.

I am already thinking about what it means to expand Vallano Media.

More sports coverage.

More publishing.

More AI-assisted workflows.

Potential clients.

Potential collaborators.

More automation.

The system I use today cannot simply be stretched forever.

It has to evolve.

That means this is where I have to apply what I have argued throughout this book.

The lessons cannot exist only on the page.

If I believe verification matters, it has to matter when production increases.

If I believe human authority matters, it has to survive automation.

If I believe workflows need measurement, then growth has to be measured too.

I want technology to increase what Vallano Media can do.

But I do not want growth to weaken the standards behind the work.

That is the challenge.

The Standard Has to Scale Too

This may be the most important point in the chapter.

When output grows, standards have to grow with it.

Verification cannot become optional because there is more work.

Human judgment cannot disappear because automation is convenient.

Editorial control cannot weaken because production is faster.

Transparency cannot become harder because the system is more complicated.

If anything, those things become more important.

Scale should increase capability.

It should not dilute responsibility.

The Journalism Still Has to Matter

There is one final danger.

The workflow could scale successfully.

The systems could become faster.

The automation could work.

The metrics could improve.

And the journalism could become less distinctive.

Less local.

Less ambitious.

Less human.

Less worth reading.

That would not be success.

The workflow exists to support the journalism.

The journalism does not exist to feed the workflow.

If scale makes the work more ordinary, I need to question what I am scaling.

The Final Question

This book began with a simple idea:

The journalist is still responsible.

Everything since then has built on that.

Research.

Drafting.

Verification.

Red-teaming.

Editing.

Human oversight.

Publishing.

Measurement.

Sustainability.

Scaling.

The technology became more capable.

The workflow became more complex.

The potential became larger.

But the central question never really changed.

“Who is responsible for what gets published?”

AI can assist.

Automation can accelerate.

Agents can monitor.

Systems can scale.

But responsibility cannot disappear into the workflow.

Someone still has to decide.

Someone still has to verify.

Someone still has to stop the process when something is wrong.

Someone still has to stand behind the result.

Scaling Without Losing Control

I do not want to build the biggest AI workflow.

I want to build one I understand.

One I can explain.

One I can challenge.

One I can improve.

One I can stop.

One that lets me do more without lowering the standard.

The goal is not total control.

No system gives me that.

The goal is enough visibility to understand what is happening.

Enough authority to intervene.

Enough resilience to recover when something fails.

And enough judgment to know when the system should not continue.

That is what scaling means to me.

Not replacing judgment.

Not removing responsibility.

Not automating everything that can be automated.

It means increasing capacity while protecting the things that matter.

Because the real test of an AI-assisted workflow is not how much it can produce.

It is whether I can still stand behind the work when the system gets bigger.

If I can do more while keeping the standards, the judgment, and the responsibility intact, then the workflow has grown with me.

If I cannot, then I have not really scaled.

I have only lost control.


Graphics created with AI-assisted design tools.
Visual concepts, editorial direction, composition, and final creative decisions were developed by Vallano Media.

Measuring the Workflow

By Cheval John | Vallano Media
This article was created with the assistance of artificial intelligence.
The final version was reviewed, edited and fact-checked by the author


Graphics created with AI-assisted design tools.
Visual concepts, editorial direction, composition, and final creative decisions were developed by Vallano Media.

A workflow can feel productive without actually being productive.

That is one of the traps.

I can move faster.

Generate more drafts.

Create more graphics.

Publish more often.

Use more tools.

Build more automations.

And still end up with a system that is not improving the work.

Activity is easy to see.

Effectiveness is harder.

That is why I need to measure the workflow itself.

Not just how much it produces.

But whether it is helping me produce better work with less unnecessary friction.

The question is not:

“Am I busy?”

It is:

“Is the system actually working?”

Chapter 11 Measured the Change

Earlier in this book, I looked at what changed after I began integrating AI into my work.

Time.

Output.

Capacity.

Repeatability.

Attention.

That gave me useful benchmarks.

But those numbers only tell part of the story.

A workflow can become faster and still become worse.

It can produce more and still create more errors.

It can feel efficient while quietly increasing rework.

It can save time in drafting and lose all of that time in correction.

So measuring the impact of AI is one question.

Measuring the health of the workflow is another.

That is what I want to focus on here.

I Need More Than One Metric

There is no single number that tells me whether the workflow is good.

Time is useful.

But time alone is incomplete.

Output is useful.

But output alone can reward volume.

Traffic may matter.

But traffic does not automatically tell me whether the journalism is accurate or useful.

So I need several signals.

For my workflow, I keep coming back to three broad areas:

Quality.

Time.

Impact.

Those three together tell me much more than any one measurement by itself.

But even that needs restraint.

If measuring the workflow takes more effort than improving the workflow, the measurement system is too heavy.

The goal is not to build a second workflow just to watch the first one.

Quality Comes First

If the workflow is faster but the work gets worse, I have not improved anything.

So quality has to come first.

That means asking questions like:

Did I catch factual problems before publication?

Did the central claims survive verification?

Did the red-team pass expose anything important?

Did the final article preserve the meaning of the evidence?

Did the graphic accurately represent the story?

Did I publish the correct version?

Did I have to issue a correction?

Did I overlook something that the workflow should have caught?

These are not always easy to reduce to a number.

That is fine.

Not every useful measurement needs to become a score.

Sometimes the most useful record is simply:

What went wrong?

Where did it happen?

Why did the workflow miss it?

Corrections Are Useful Data

Nobody wants to make an error.

A correction is a failure in the published work.

But it can still produce useful information about the workflow.

Where did the mistake enter?

Research?

Drafting?

Verification?

Editing?

Graphic production?

Scheduling?

Was the source wrong?

Did I misread the source?

Did AI introduce the error?

Did I approve something too quickly?

Did the wrong version get published?

The correction itself matters.

But so does understanding how it got through.

That is how a workflow improves.

And no corrections does not automatically mean no errors.

A problem may simply have gone unnoticed.

That is another reason not to treat one metric as proof that the system is healthy.

Near-Misses Matter Too

Not every important problem reaches publication.

Sometimes the workflow catches it first.

That is a success.

A bad statistic caught during verification.

A misleading headline caught during red-teaming.

A strange graphic caught during QC.

A sentence that changes meaning during editing.

A weak source discovered before publication.

Those near-misses are worth recording.

They tell me which safeguards are actually doing useful work.

But their numbers can be deceptive.

More detected problems can sometimes mean the safeguard improved, not that the journalism declined.

If the truth pass suddenly starts finding more weak claims, that may mean I have a bigger problem.

Or it may mean the truth pass has become better at finding them.

The number needs interpretation.

And if a check rarely catches anything, I should not automatically remove it.

It may be unnecessary.

Or it may protect against a low-frequency, high-consequence problem.

I need to know which.

Time Still Matters

Quality comes first.

But time still matters.

A workflow that produces excellent work but takes twice as long as necessary may not be sustainable.

So I want to know:

How long does research take?

How long does drafting take?

How long does verification take?

How much time goes into rework?

How many graphic revisions are needed?

How long does the full publication package take?

Where does the process slow down?

I do not need to time every article forever.

But periodic timing gives me useful benchmarks.

The goal is not to race the clock.

The goal is to understand where the time goes.

And there is one caution I need to remember:

“Precision in the number does not create precision in the conclusion.”

I can know that something took exactly 36 minutes and 54 seconds.

That does not mean I can make an equally exact claim about why it took that long.

Measure Rework, Not Just First-Pass Speed

This has become especially important.

A fast first draft can look impressive.

But if I spend another hour fixing it, the first-draft time does not tell me much.

The same applies to graphics.

A graphic generated in seconds is not efficient if it takes six attempts to get one usable version.

So I need to think about rework.

How many times did I have to go back?

What caused the revision?

Was the problem predictable?

Could the instruction have been better?

Could the workflow have caught the issue earlier?

A system that gets the work closer to right the first time may be more valuable than one that produces the first version fastest.

Bottlenecks Tell Me Where to Look

Every workflow develops bottlenecks.

The slowest point may not stay the same forever.

Early on, drafting may be the problem.

Later, verification may become the bottleneck.

Then graphics.

Then version control.

Then publishing.

A recurring bottleneck is one place I should investigate for possible improvement.

That does not mean the slowest stage is automatically broken.

Verification may be slow because the reporting requires care.

The answer may not be to shorten it.

Sometimes the answer is better organization.

Better instructions.

Better source access.

Better documentation.

A better tool.

Or a simpler process.

Measurement tells me where to look.

Judgment tells me whether something actually needs to change.

Output Needs Context

I can count articles.

Posts.

Graphics.

Chapters.

Social updates.

That is useful.

But output needs context.

More finished work may indicate better capacity.

Or it may indicate that I am publishing too much.

So I need to ask:

Did the additional output maintain the standard?

Was it necessary?

Was it useful?

Did it create more correction work?

Did it reduce the time available for higher-value reporting?

The number itself is not the answer.

It is a signal.

Impact Is Harder to Measure

Impact is probably the hardest part.

What did the work actually do?

Did people read it?

Did they spend time with it?

Did they share it?

Did it lead to another conversation?

Did a source respond?

Did someone contact me because of it?

Did the article help establish credibility?

Did it support the larger business?

Did it make the next piece easier to produce?

Not every article needs the same kind of impact.

A sports recap may serve a different purpose than an AI journalism essay.

A book chapter may be building long-term authority rather than immediate traffic.

A business article may support a service offering.

So impact has to be measured against the purpose of the work.

“Impact should be interpreted, not simply counted.”

Traffic Is Not the Same as Value

This matters to me.

A high-traffic article may be valuable.

A low-traffic article may also be valuable.

Traffic is one signal.

It is not the entire score.

A piece may reach a small audience but the right audience.

It may create a business lead.

It may strengthen expertise.

It may become part of a larger book.

It may be cited later.

It may perform slowly over time.

So I want to be careful about letting page views decide what deserves to exist.

The purpose comes first.

Then I measure whether the work served that purpose.

Some Measures Happen Before Publication. Others Come After.

There are really two sides to measurement.

Some signals tell me what is happening inside the workflow.

Verification problems.

Rework.

Headline changes.

Graphic errors.

Weak sources.

Those are visible before publication.

Other signals arrive later.

Corrections.

Reader response.

Traffic.

Business inquiries.

Long-term usefulness.

Those tell me what happened after the work reached the audience.

I need both.

Internal measurements tell me how the process behaved.

External measurements help tell me what the work accomplished.

The Workflow Should Reduce Avoidable Errors

One useful long-term question is simple:

Am I making fewer preventable mistakes?

Not zero mistakes.

That is not realistic.

But fewer mistakes caused by:

missing a basic fact,

using the wrong version,

skipping verification,

forgetting a disclosure,

publishing the wrong graphic,

or accepting an AI-generated claim without checking it.

If those problems decrease, the workflow may be doing something useful.

If they keep happening, the system needs attention.

I Can Measure Consistency

Another question:

Do I follow the process consistently?

Not perfectly.

But consistently enough that I know what standard the work went through.

That includes:

verification,

red-teaming,

voice editing,

QC,

final approval.

If some articles get the full process and others skip important safeguards without a reason, the workflow is not really stable.

Consistency does not mean identical treatment.

It means the appropriate standard is applied intentionally.

An unusual story should not be treated as a workflow failure simply because it does not fit the normal benchmark.

Some work deserves a different path.

I Can Measure Friction

This is less obvious.

What part of the workflow annoys me?

Where do I hesitate?

What do I keep avoiding?

What feels harder than it should?

Those feelings can reveal design problems.

If I keep avoiding documentation, maybe it is too complicated.

If I keep skipping a check, maybe the check needs to be redesigned.

If a tool constantly interrupts the process, maybe it does not belong.

Friction is not always bad.

Verification is sometimes slow because verification needs to be slow.

The goal is to remove unnecessary friction.

Not necessary care.

I Can Measure Cognitive Load

This is harder to quantify, but I can still observe it.

Do I feel like I am juggling too many things?

Am I constantly wondering what comes next?

Am I keeping too many unfinished tasks in my head?

Do I lose track of which version is final?

Does the workflow make the work feel more organized or more chaotic?

A strong system should reduce unnecessary mental clutter.

If automation creates more monitoring than it removes, that matters.

If five tools require five different kinds of attention, that matters too.

The workflow should give attention back.

Not consume all of it.

I Can Measure Reliability

Does the system work when I need it?

That includes the technology.

But it also includes the process.

Can I still publish if one tool is unavailable?

Can I verify a claim manually?

Can I recover if an automation breaks?

Can I identify the final version?

Can I explain what happened?

Reliability should not mean every story follows the same path.

It should mean the workflow responds predictably when something goes wrong.

A fast system that constantly breaks is not efficient.

It is fragile.

I Should Measure the System, Not Perform for the Metric

There is a danger in measurement.

Once I start tracking something, I can start optimizing for it.

If I measure speed, I may rush.

If I measure output, I may publish more than I should.

If I measure traffic, I may chase clicks.

If I measure correction rate, I may become too cautious.

So the metrics have to serve the work.

The work should not serve the metrics.

And the moment a metric becomes something I am trying to win, I need to question whether it is still helping me.

Measurement should help me ask better questions.

Not create a game I am trying to beat.

Metrics Need Maintenance Too

A useful metric today may become less useful later.

The workflow changes.

The assignments change.

My experience changes.

The technology changes.

So the measurements need review too.

Does this number still tell me what I think it tells me?

Is this benchmark still comparable?

Am I measuring something simply because I have always measured it?

A metric should earn its place just like a workflow step should.

A Simple Workflow Scorecard

I do not need a complicated dashboard.

A simple review can tell me a lot.

After a project, I can ask:

“Quality”
– Did the work meet the standard?
– What errors or near-misses occurred?
– What did the workflow catch?

“Time”
– How long did the major stages take?
– Where did I lose time?
– How much rework was required?

“Impact”
– Did the work achieve its purpose?
– What happened after publication?
– Was the result worth the effort?

Then one more question:

“What should I change next time?”

That last question turns measurement into improvement.

The Workflow Should Improve Over Time

If I collect data but never change anything, the measurement does not do much.

The point is not to create reports about the workflow.

The point is to make the workflow better.

Maybe I discover that graphic revisions are taking too long.

I improve the design specification.

Maybe verification repeatedly finds missing dates.

I add a date check earlier.

Maybe social copy keeps overstating the headline.

I add a comparison against the locked manuscript.

Maybe one automation rarely saves time.

I remove it.

Measurement should lead to decisions.

Remove What Does Not Help

This is just as important as adding improvements.

Workflows tend to grow.

They rarely shrink on their own.

A new problem creates a new step.

A new tool creates a new process.

A new idea creates another check.

Eventually I need to ask:

What can go?

If a step no longer protects the work, improves the work, or saves meaningful effort, it may not deserve to remain.

Measurement gives me permission to simplify.

That is part of sustainability too.

Look for Trends, Not One-Off Results

One article does not prove much.

Neither does one perfect graphic.

Or one terrible day.

I want patterns.

Are production times gradually falling?

Is rework decreasing?

Are the same errors recurring?

Is verification catching problems earlier?

Are graphics requiring fewer revisions?

Is the workflow handling more work without becoming chaotic?

That is more useful than reacting to every individual result.

The goal is not perfection.

The goal is direction.

The Numbers Need Notes

A number without context can mislead me.

Thirty minutes may be fast.

Or it may be slow.

It depends on the assignment.

So when I record a benchmark, I also need to know what happened.

Was the research already complete?

Was the topic familiar?

Did an interview delay the process?

Did a tool fail?

Was the story unusually simple?

Did I have to redo the graphic?

Context keeps the number honest.

I Am Measuring My Own System

There is another limitation I need to remember.

I am both using the workflow and evaluating it.

That means I can be biased toward believing the system I built is working.

I may notice the successes more readily than the failures.

I may explain away a weak result because I know how much work went into the process.

So I have to be alert to the possibility that I am seeing what I want to see.

That does not make self-measurement useless.

It means I should document enough evidence that I can challenge my own interpretation.

Measurement Should Include What AI Gets Wrong

If I am measuring an AI-assisted workflow, I should track AI failures too.

Not every typo.

Not every awkward sentence.

Not every useless suggestion.

That would become its own job.

I care more about recurring or consequential failures.

Fabricated details.

Bad source suggestions.

Overconfident summaries.

Voice drift.

Weak headline recommendations.

Graphic text errors.

Misleading visual choices.

Repeated failures tell me where AI needs tighter supervision.

And an AI mistake is not automatically a workflow failure if the workflow catches it before publication.

Sometimes catching the mistake is evidence that the safeguard worked.

Measurement Should Include What AI Does Well

The opposite matters too.

Where is AI consistently useful?

Organizing notes?

Finding contradictions?

Creating alternative structures?

Cleaning up grammar?

Generating first-pass graphics?

Repeated success in a specific task may justify greater operational confidence in that task.

But success in one task should not automatically expand trust into another.

A system that organizes notes well has not proved that it should decide which source is credible.

A tool that produces useful headline options has not earned authority over publication.

Trust has to remain specific.

Trust Should Be Specific

I do not want to say:

“I trust AI.”

That is too broad.

I may trust one workflow for one type of task under one set of conditions.

That is different.

I may trust AI to organize my notes.

But not to establish the facts.

I may trust it to suggest headlines.

But not to decide which headline gets published.

I may trust it to flag a possible contradiction.

But not to resolve the contradiction without checking the sources.

Measurement helps make that confidence more specific.

And the required checks remain even when a task has performed well repeatedly.

The Journalism Cannot Become More Ordinary

There is one more thing I do not want measurement to erase.

Ambition.

Originality.

Work that does not fit the normal template.

A workflow can become cleaner and more efficient while the journalism becomes more ordinary.

That would not be success.

Some stories deserve more time.

Some projects deserve more experimentation.

Some reporting will never fit neatly inside the average benchmark.

I do not want an unusual story to look like a failure simply because it took longer.

The workflow should support ambitious journalism.

Not train me to avoid it.

The Workflow Has to Produce More Than Activity

This is what I keep coming back to.

A busy workflow can look impressive.

Multiple agents.

Automations.

Dashboards.

Prompts.

Checklists.

Constant output.

But none of that matters if the work does not improve.

I want a system that produces:

better decisions,

fewer avoidable mistakes,

less unnecessary rework,

more usable capacity,

and journalism I am willing to defend.

That is what I am measuring.

Measuring the Workflow

So how do I know whether my AI-assisted workflow is actually working?

I look beyond speed.

I look beyond output.

I look beyond whether the technology did something impressive.

I ask:

Is the quality holding?

Is the process becoming easier to repeat?

Is rework decreasing?

Are problems being caught earlier?

Am I using my time better?

Is the work achieving its purpose?

Is the system still under control?

Those questions tell me more than any productivity number by itself.

Because a workflow is not successful simply because it keeps me moving.

It is successful when it helps me move in the right direction.

And measurement is useful only when it helps me see whether I still am.


Graphics created with AI-assisted design tools.
Visual concepts, editorial direction, composition, and final creative decisions were developed by Vallano Media.