After three years leading her city's AI transformation, Deputy CIO Renata Voss had a problem most leaders would envy and most never solve. She had learned an enormous amount: what worked deploying AI in a unionized workforce, why two procurements failed, how she rebuilt public trust after a chatbot fiasco. All of it lived in her head and a few internal slide decks. When a peer from another city asked how she had done it, she realized her hardest-won knowledge would leave the building when she did. A colleague suggested she write it up. Her first reaction was the one almost every public servant has: "I'm an operator, not a writer. Who would publish me?" Eighteen months later her field-tested framework for AI procurement had been cited in two state policies and adopted by four agencies.

At the most senior level of a career, your contribution shifts from running programs to shaping the field. Publishing is how you do that at scale. A well-placed article reaches people you will never meet and changes practices you will never see. This lesson is a practical guide to turning your operational experience into published work that actually moves government AI forward, and to navigating the constraints of publishing as a public servant, which are real, which are not the same as the constraints a private author faces, and which are the reason most of the best material in this field never gets written down.

The missing voice in the government AI conversation

The government AI conversation is dominated by two voices: vendors selling tools and academics studying them from outside. The voice that is almost missing is the practitioner who actually deployed something in a real agency, with real constraints, and can say what happened. Vendors cannot publish the failures because failures are not a sales asset. Academics rarely have access to the internal decisions that determined the outcome. The person who was in the room when the procurement went wrong is the only one who can describe why, and that person is usually too busy to write.

Notice what this scarcity does to the field. When practitioner accounts are missing, the published record of government AI consists of capability claims and external critique, with almost nothing in between about how the work is actually done. New leaders arriving in the job read that record, find no operational guidance in it, and repeat mistakes that were solved years ago in another city. The cost is not borne by the writer who stayed silent. It is borne by the next agency, and by the residents of whatever jurisdiction it serves.

The barrier is never lack of substance. It is three myths, and it is worth naming them precisely because each one dissolves under a specific counter-argument rather than under general encouragement. Encouragement fails here because the myths are not confidence problems. They are category errors about what practitioner writing is, who it is for, and what it costs, and a leader who has quietly accepted all three will decline the invitation politely every time it is offered without ever noticing that the reasoning does not hold.

• "I'm not a writer." You are not writing literature. You are documenting what you learned so the next leader does not relearn it at full cost. Clarity beats eloquence, and the strongest practitioner writing tends to read like a well-run meeting: a clear problem, what was tried, what happened, what you would do differently.
• "It's not original enough." What is obvious to you after three years is a revelation to someone in year one. Lived detail is the originality. Nobody else can write your specific account of why a procurement failed, because nobody else saw the sequence of decisions that produced it.
• "I don't have time." One sharp 1,500-word article has more reach than a hundred meetings. It is leverage, not overhead. The time question is really a sequencing question: the article competes with meetings for this month and beats all of them combined over the years that follow.

Renata's own version of the third myth is instructive. She did not find spare time. She recognized that she was already explaining the same procurement lessons to peers one conversation at a time, and that writing them down once was the cheaper version of a conversation she was going to keep having anyway. That is usually where a practitioner's first article is hiding: in the explanation you have already given several times.

Choosing which experience to write about

Renata came to the page with three candidates, and they were not equally publishable. She had learned what worked deploying AI in a unionized workforce, why two procurements failed, and how she rebuilt public trust after a chatbot fiasco. Any of the three would have made an article. Only one of them was ready. Working out which of your experiences to write first is a triage problem, and getting it wrong is the most common reason a promising first piece stalls halfway through a draft.

Three tests do most of the sorting. First, can you state the lesson in one sentence that a peer would find useful rather than obvious? A subject that resists that sentence is usually still being processed rather than understood. Second, is there an artifact at the end of it, something a reader could apply directly? Experience that has not yet condensed into a framework, checklist, or template will produce reflection rather than adoption. Third, can the account survive the removal of everything you cannot publish?

That third test is where the unionized-workforce material and the chatbot recovery were harder than the procurement work. Both involved named parties, live relationships, and detail that would have to come out, and what remained was thinner than it looked. The procurement lessons passed all three tests: a clear one-sentence claim, a framework a procurement officer could pick up, and a story that worked without naming a vendor or quoting a scoring sheet. Write the one that passes first. The others usually become writable later, once the programs close and the relationships settle.

Matching the message to the venue

Different outlets reach different audiences and demand different things. Pick the venue for the change you want to create, not for prestige. Prestige is a poor guide here, because the most cited venue in your field may be read by nobody who can act on what you know, while a trade publication your peers actually read can move practice quickly. Decide first who has to behave differently for your effort to have been worth it, then work backward to where that person reads.

• Practitioner and policy publications, which include government technology outlets, professional association journals, and the trade press. These reach peers and policymakers who can act. They are the best home for frameworks, lessons learned, and how-to material, and they offer the fastest path to the widest practical impact. Start here.
• Mainstream media op-eds. These reach the public and elected officials, and they are the right choice for shaping the broad debate about trust, transparency, and public value. They are short, opinionated, and built around one clear argument, which makes them the hardest form for an expert who wants to include every caveat.
• Research and academic papers. These reach the evidence base and the long-term record, and they suit rigorous results that are worth formal scrutiny. They are slow, demanding, and durable, and they are often best done with an academic co-author who knows the conventions of the field you are publishing into.

The venue choice also determines the review you will need, which is the part practitioners underestimate. A framework in a trade journal, an opinion piece under your own name, and a co-authored study using agency data raise different questions for your agency's reviewers, and the third will raise questions the first two never do. Choosing the venue early lets you start the clearance conversation with something concrete, which is a far better conversation than asking in the abstract whether you are allowed to write.

The article skeleton that works

Renata's procurement article followed a structure you can reuse for almost any practitioner piece. It turns experience into something a reader can act on, and it survives the transition between venues with only the length changing. The reason it works is that it matches the order in which a busy reader decides whether to keep going: first whether this situation resembles theirs, then whether there is a single takeaway worth the time, then whether the author is credible, then whether there is anything to take away and use.

1. Open on the problem, concretely. A real moment: the failed procurement, the chatbot crisis. Specifics earn attention, and the specific moment is also what stops a reader deciding in the first paragraph that your situation was nothing like theirs.
2. Name the lesson in one sentence. The single thing you want the reader to remember, placed early. If you cannot write that sentence, you are not ready to write the article, and no amount of drafting around the gap will hide it.
3. Show the evidence. What you tried, what happened, real numbers and timelines, what failed. Honesty about failure is what makes practitioner writing trusted; a piece in which everything worked reads like a vendor case study and is discounted accordingly.
4. Give them the artifact. A framework, checklist, or template the reader can use Monday morning. This is what gets cited and adopted, and it is the difference between a piece people admire and a piece people use.
5. Close with the broader implication. Why this matters beyond your agency, stated in a way that a reader in a different jurisdiction can map onto their own situation.

The fourth step is the one most first drafts skip, and it is the one that did the work in Renata's case. Her framework was adopted by four agencies not because her narrative was compelling but because there was something at the end of it that a procurement officer could put into a document the following week. If your draft ends on reflection rather than on an artifact, you have written an essay about the work rather than a contribution to it.

Evidence you can actually publish

Here is where the practitioner's advantage collides with the public servant's constraint. The evidence that makes your account credible is often exactly the material you cannot release: the vendor scoring sheet, the incident timeline, the numbers behind a decision that is still under procurement. The instinct is either to publish it anyway, which is a serious mistake, or to abandon the piece, which is the more common and more wasteful one. There is a third option, and it is usually the right one.

Publish the method rather than the record. You can almost always describe how you evaluated something, what the decision turned on, and what you would test differently, without releasing the underlying documents. A reader who learns the questions that exposed a weak vendor claim gets more usable value than one who reads your scoring sheet. Where a number is genuinely central and genuinely publishable, use it. Where it is not, say plainly that the figure is not public and describe the shape of the finding instead. Readers respect the boundary when you name it; they distrust vagueness that is never explained.

Two related disciplines protect you and your reader. Verify every number you do publish against its source before submission, because a figure you half-remember from a slide will be the one a critic checks. And describe failures fairly, including your own share of them, without making a named colleague the villain of your article. The person who owned the failed procurement may still be in the job. A practitioner account that costs somebody else their standing will be the last one your agency lets anyone write.

The public servant's publishing checklist

Publishing as a government employee carries obligations a private writer does not have. Clear these before you submit, every time. This is the artifact to keep taped to your monitor, and the ordering matters as much as the content: clearance sits first because it is the item that kills articles late when it is skipped. Every other item can be fixed in a revision. Clearance, discovered at the end, can cost you the piece, the placement, and the editor who had made room for it, which is why experienced government authors treat it as the first task rather than the last hurdle.

• Clearance. Have you followed your agency's review and approval process for public writing? Ask what that process is before you have a draft, not after, and get the answer from whoever actually owns it rather than from a colleague's recollection.
• Capacity. Are you writing in a personal capacity or representing the agency? State it explicitly, usually with a disclaimer that the views are your own.
• Sensitive information. No nonpublic data, security details, procurement-sensitive specifics, or anything that could embarrass a named individual.
• Ethics rules. No appearance of endorsing a specific vendor and no use of your office for private gain. Describe categories of tools rather than favored brands.
• Accuracy and fairness. Numbers verified, failures described fairly, credit shared with your team.
• Plain language and accessibility. Written so a non-specialist can follow, with acronyms unpacked on first use.

Treat this list for what it is: a set of prompts that tells you which questions to raise, not a determination that you have satisfied them. Your agency's review process is the authority on what you may publish, and a checklist completed at your desk has never once substituted for it. The items about capacity and vendor endorsement in particular turn on rules your ethics official interprets, and the honest answer to most of them is that you do not know until you ask.

Ask early, and ask in writing. A reviewer asked for permission to publish an unspecified article gives a cautious answer; a reviewer given a venue, a subject, a length, and an outline gives a usable one. When the answer is no, find out which item drove it, because a no is often narrower than it sounds. Removing a procurement-sensitive detail, waiting until a program closes, or reframing a case as a category rather than a named system converts a large share of refusals into approvals with conditions.

None of this is meant to silence you. It is meant to let you publish repeatedly with less friction each time, which is the actual goal. The leaders who shape the field are the ones who work out their agency's process once, build a relationship with the person who runs it, and then keep writing, rather than treating every article as a fresh negotiation.

Writing in your own name, and giving credit

The capacity question deserves more care than the one line it usually gets. A disclaimer that the views are your own tells the reader in what capacity you intend to write. It does not by itself convert official-capacity writing into personal writing, and it does not stand in for whatever review, outside-activity approval, or ethics consultation your agency requires. If you were invited because of your position, if you are writing about systems you administer, or if any part of the work touches official time or resources, the disclaimer is a labeling decision made after the substantive question has been answered, not instead of it.

The substantive question is one for your agency's ethics official, and the useful move is to arrive with the specifics rather than the abstraction: this venue, this subject, this relationship to my duties, this compensation or none. Practitioners who ask that way generally get a clear answer. Practitioners who ask whether they are "allowed to write about AI" generally get told to be careful, which is not an answer and cannot be relied on later.

Credit is the other half of writing in your own name. Almost nothing you will write about was achieved by you alone, and the account will be both more accurate and better received if the team appears in it. Name the people whose work you are describing where they consent to be named, offer co-authorship where someone's contribution is genuinely co-equal, and let colleagues read the draft before an editor does. That last step catches factual errors, and it also prevents the quiet damage of a colleague learning from a published article how you characterized their project.

From one article to a body of work

The first article is the hardest and the least important. Its real value is that it opens doors: a speaking invitation, a peer who adopts your framework, an editor who asks for more. Treat publishing as a practice rather than an event. Renata's procurement piece led to a conference talk, the talk led to a state working group, the working group led to the policy citation. The article did not create the influence directly. It made her legible to people who were looking for someone with exactly her experience and had no way to find her.

A practice needs a cadence you can actually sustain alongside the job, and a sustainable one usually means fewer, better pieces rather than a publication schedule. It also means reusing material honestly. The framework in an article becomes the spine of a conference talk, the talk surfaces the objections you did not anticipate, and those objections become the second article. Each pass through the loop is cheaper than the last because the clearance path is already established and the underlying thinking is already done.

It is worth deciding in advance how you will know whether any of this worked, because the obvious measures are the wrong ones. Views and shares tell you almost nothing about whether a procurement officer in another jurisdiction changed a document. The signals that matter are the ones Renata got: a peer who writes to say they used the framework, an invitation to a room where the decision is being made, a citation in something that binds. Those arrive slowly and unpredictably, which is a reason to judge a practice over years rather than to judge an article over weeks.

Knowledge that would have left with Renata instead became part of the institution, and then part of several institutions that had never employed her. That is the legacy work of a senior leader: not the systems you ran, which will be replaced, but the practice you left behind for the people who come after you. The systems have a service life measured in years. A framework that four agencies adopted, and that the next generation of procurement officers learned from, outlasts every one of them.

Anti-Patterns to Avoid

• Treating the disclaimer as clearance. "Views are my own" is a labeling decision, not an authorization. It does not convert official-capacity writing into personal writing and it does not satisfy an agency review process. If you have not asked, the disclaimer protects nobody, least of all you.
• Treating a generic checklist as your agency's answer. The list in this lesson tells you which questions to raise. Your agency's reviewers and ethics officials decide the answers, and their answer is the one that governs. A checklist completed at your desk has never made an unapproved publication approved.
• Seeking clearance last. Writing the whole piece, placing it with an editor, and then discovering that one detail is procurement-sensitive wastes the work and burns the editor relationship. Ask before you draft, with a venue and an outline in hand.
• Publishing the sanitized success story. An account in which everything worked reads like marketing and is discounted as marketing. The failures are the part only you can contribute, and removing them removes the reason anyone would trust the piece.
• Writing the essay without the artifact. A reflective piece with no framework, checklist, or template at the end gets read and forgotten. Adoption follows the thing a reader can use on Monday, not the narrative that led you to it.
• Making a colleague the villain. Honest failure analysis is not the same as naming who lost the argument. A piece that damages someone still in the job will be the last one your agency approves from anyone.
• Chasing the prestigious venue. Publishing where the audience cannot act on what you know is effort spent on the wrong readers. Decide who has to behave differently, then find out where that person reads.

Practice Prompts

• Write down the explanation you have already given peers more than once. That is your first article. Draft its one-sentence lesson before you write anything else, and stop if you cannot.
• Take a project you led and list the evidence that made you confident in your conclusion. Mark each item as publishable, publishable in summary, or not publishable. Then draft the paragraph that conveys the finding using only the first two categories.
• Find out, this week, exactly who owns pre-publication review in your organization and what they need from you. Write down the process in the order the steps actually occur. You now have an asset your whole team can use.
• Choose the change you want your writing to produce and name the specific role that would have to act on it. Identify two venues that role reads. Compare them on audience, length, and the review your agency would require.
• Build the artifact first. Write the checklist, framework, or template you would want handed to you on your first day in the job, then write the article that explains where it came from.
• Draft the capacity question you would put to your ethics official: venue, subject, relationship to your duties, and whether any compensation is involved. Notice how much clearer the question is than "am I allowed to write about this".

Reflection

Think about the hardest thing you learned in this work, the lesson that cost real money or real trust to acquire. Where does it currently live? If the honest answer is your own memory and a slide deck on a shared drive, then it will leave when you do, and the next person in your seat will pay for it again. That is not a personal failing. It is the default outcome of a system that gives senior public servants no time and no incentive to write, and it repeats in every jurisdiction unless someone decides to interrupt it.

Then ask what has actually been stopping you. For most people it is not clearance, which is usually navigable once asked, and it is not ability, which is usually adequate. It is the belief that what you know is too obvious to be worth writing. Test that belief against a specific person: someone one year into the job you have held far longer. Write down what you would tell them in a single conversation. If that conversation would be valuable to them, the article is worth writing, and the only remaining question is when you start.

Glossary

• Pre-publication review. Your agency's process for reviewing and approving material an employee intends to publish. The process, its owner, and its requirements vary by agency, and it is the authority on what you may publish.
• Personal capacity. Writing as yourself rather than on behalf of your agency. Whether a given piece qualifies turns on rules your ethics official interprets, not on the presence of a disclaimer.
• Practitioner publication. A government technology outlet, professional association journal, or trade publication whose readers are peers and policymakers positioned to act on operational guidance.
• Op-ed. A short opinion piece in general-audience media, built around a single argument and aimed at the public and elected officials rather than at practitioners.
• The artifact. The framework, checklist, or template a reader can apply directly. It is the element of a practitioner article that drives citation and adoption.
• Procurement-sensitive information. Material relating to an acquisition whose release could affect the competition or the process. A common and legitimate reason for a reviewer to require changes or a delay.

Related Lessons

• Speaking and Presenting on Government AI covers the platform your first article most often opens, and the material transfers directly.
• Writing Policy Papers and Legislative Proposals takes the same discipline into documents intended to change law rather than practice.
• Building Institutional Knowledge addresses the internal half of the problem this lesson solves externally.
• Mentoring Next-Generation Leaders is the person-to-person channel for the same hard-won judgment.
• Public Reporting and Algorithmic Transparency deals with what the agency itself must publish, as distinct from what you publish about your practice.
• Academic and Research Collaboration is the route into the research and academic venue described here, usually with a co-author.
• Contributing to Standards Bodies (NIST, ISO, IEEE, OECD) is another way practitioner experience enters the durable record.

Closing

Renata did not become a writer. She became a public servant who wrote, which is a different and much more achievable thing. The framework she published was not more sophisticated than what many of her peers were carrying around in their heads. Its advantage was simply that it existed outside one person's memory, in a form somebody else could pick up and use, with the failures left in so that readers could tell it was real.

That is the whole proposition. Your operational experience is the scarcest input in the government AI conversation, and it is currently sitting in slide decks that will be deleted when your account is closed. Clear the review process once, write the thing you have already explained several times, put an artifact at the end of it, and let it travel to the agencies you will never visit. The systems you built will be replaced. What you wrote down about how to build them will not be.

Key Takeaways

• Publishing is leadership at scale. At the senior level your job shifts from running programs to shaping the field, and writing reaches people and practices you never could in person.
• The practitioner voice is rare and valuable. The debate is dominated by vendors and academics; the leader who actually shipped something has the missing perspective, and its absence makes the next agency repeat solved mistakes.
• The barriers are myths. You do not need to be a writer, your obvious lessons are someone else's revelation, and one sharp article out-leverages a hundred meetings.
• Match venue to intended change. Practitioner and policy outlets for frameworks, op-eds for public debate, research papers for the durable evidence base. Decide who must act, then find where they read.
• Use a proven skeleton. Concrete problem, one-sentence lesson, honest evidence including failures, a usable artifact, and the broader implication. The artifact is what gets adopted.
• Publish the method when you cannot publish the record. Describe how you evaluated something rather than releasing the underlying documents, and name the boundary instead of being vague about it.
• Clearance comes first, not last. Ask before you draft, with a venue and an outline, and find out which item drove any refusal, because a no is often narrower than it sounds.
• A disclaimer is a label, not an authorization. "Views are my own" does not convert official-capacity writing into personal writing or replace agency review and ethics consultation.
• Share the credit and check the draft with the team. Naming colleagues and offering co-authorship where it is earned catches errors and protects the relationships that make the next article possible.
• Treat it as a practice, not an event. The first article matters mainly because it opens the doors that turn your hard-won knowledge into institutional legacy.

Frequently Asked Questions

How long does agency clearance take?

It varies by agency, by the subject, and by how much of the material touches active programs, and anyone who quotes you a single number for government generally is guessing. The practical move is to find out your own organization's answer before you have a deadline: ask who owns the review, what they need from you, and what turnaround they can commit to. Then build that answer into your conversation with an editor rather than discovering it afterward.

Can I write about a system my agency is still running?

Often yes, with conditions, and the conditions are the point of asking. Live systems raise questions about nonpublic data, security detail, and procurement sensitivity that a closed program does not. Bring your reviewer a specific proposal rather than a general request, and expect the useful outcome to be a narrower piece: the method rather than the numbers, the category rather than the named vendor, or the same article after a procurement closes.

Do I have to run everything past the agency if I write in a personal capacity?

Ask your ethics official rather than assuming. Whether a piece is genuinely personal turns on factors including why you were invited, whether the subject is within your official duties, and whether official time or resources are involved, and those are interpretations you should not make on your own behalf. Adding a disclaimer records how you are presenting the work; it does not answer the underlying question, and it will not answer it later either.

What if my agency's leadership does not want the failures published?

That objection is worth taking seriously and negotiating rather than either abandoning or overriding. Much of the value of a failure account survives the removal of identifying detail: the sequence of decisions, the signal that was missed, and the control that would have caught it are all publishable in a form that does not embarrass anyone. If the answer is still no, the piece may simply have to wait, and it is worth writing the internal version now so that the material still exists when the constraint lifts.

Should I write alone or with a co-author?

It depends on the venue and on who did the work. Research and academic papers usually go better with an academic co-author who knows the conventions of that literature. Practitioner pieces are often stronger single-authored because the first-person account is the asset. Regardless of the byline, let the colleagues whose work you describe read the draft, and offer co-authorship where a contribution was genuinely co-equal rather than treating acknowledgment as sufficient.

How do I write about tools without endorsing a vendor?

Describe categories rather than favored brands, and describe capabilities and limitations rather than rankings. The concern behind the rule is the appearance that a public official is steering business, and that appearance attaches to comparative praise more than to plain description. When a specific product is genuinely unavoidable to the account, raise it with your reviewer rather than deciding for yourself where the line sits.

Nobody in my agency has ever published. Where do I even start?

Start by establishing the path rather than by writing. Find who would own review if someone asked, ask them what the process would be, and write down what they tell you. That document is useful to your whole organization and it costs you very little. Then draft the explanation you have already given peers several times, keep it short, put a usable artifact at the end, and take it to a practitioner publication whose readers are people in your job.