The Footnote
No. 6 · Building in public
Monday, October 12, 2026. How this newsletter gets made, start to finish. The agent that does the research, the rules it writes under, the checks I do by hand, and what the whole thing says about the podiatrist our profession will need next.
Building in public · Essay

I don't write the first draft. I own the last one.

I am the editor of this newsletter. I hired AI to do my research and first drafts. Every edition of The Footnote so far began as a draft I did not write. An AI agent searched the journals and the trade press, read the sources, and wrote in a voice built from my own rules. Then I did the part that carries my name. I read every sentence, opened every link, and fixed what was wrong.

Most people who write with AI keep it quiet, but most of us can tell. I think quiet is the wrong instinct for physicians. We already sign notes that someone else drafted, first human scribes and now ambient ones. We manage residents and staff, all of them act as an extension of us. We sign off on work that we review. The signature has always meant the same thing: I checked this, and it is mine. This newsletter works the same way, just with a new staff.

So this edition shows the whole machine. You will see the steps, the constraints I put on the agent, the places it still gets things wrong, and why I think this way of working is the clearest preview we have of the physician this profession will need next. Nothing here requires a computer science degree. It requires the same habits we learned in residency, applied to a new kind of assistant.

5editions written in the first three weeks, alongside a full-time job
6beats the agent researches on every run
1physician who reads every sentence before it ships
0figures published without a link to the source

The pipeline, step by step

Six steps, in the order they happen. The tag on each one shows who does the work.

Agent Claude, through the Anthropic API Code scripts and scheduled jobs Me the part I will not hand off
  1. Pick the thesis

    I choose the topics and what the edition argues and who it is for. The resident edition, the profession edition, this one. The agent never chooses the story. It serves up what I ask it to.

    Me
  2. Research six beats

    The agent searches the web for each beat: clinical, surgical research, policy, practice business, products, and voices. It works from a brief that ranks journals like ACFAS, the Journal of Foot & Ankle Surgery, AOFAS, and Foot & Ankle International ahead of everything else. It is told to skip press releases dressed as news, and to return nothing when nothing is worth running.

    Agent
  3. Draft in my voice

    Every summary is written from scratch, never copied from the source. The draft follows a written style guide, shown below, that the agent cannot negotiate with.

    Agent
  4. Check the citations

    Journal references run through PubMed and Crossref to confirm the paper exists and says what we claim. This pipeline started on my bunion patient education site, where I never wanted a fabricated reference to reach a patient.

    Code
  5. Read, verify, correct

    I open every link and read every line against its source. I have corrected follower counts, rankings, and a colleague's title. Some weeks the fix is a number. Some weeks it is a whole opening paragraph that said the right thing in the wrong tone. Today it was nuances throughout.

    Me
  6. Build and publish

    The edition renders into a fixed template. A script blocks any would-be sponsor from running in an edition that covers its own product. The finished page waits in a queue, publishes itself Monday at 6 AM Central, and the email follows at 7 so the link is already live.

    Code

The rules the agent writes under

An AI draft sounds generic when nobody gave it a spine. These are the constraints written into every run. Several of them exist because I caught the agent breaking them once.

Voice
  • Start with an admission or an observation, then the insight
  • One thesis per piece
  • No questions aimed at the reader
  • No sales pitch in the body copy
  • No em-dashes and no emoji
  • Close on who the reader is, never on a question
Evidence and money
  • Never invent a URL, a publication, or a statistic
  • Every figure links to where it came from
  • If a page blocks automated reading, flag the detail for me to confirm
  • Disclose my roles whenever a story touches them
  • In time we may have one sponsor per edition, labeled, and it never sees a story before publication

Why an AI-built newsletter can be better

The trade newsletters most of us open were built for a different era. Editorial hours were the scarce resource, so the format became a stack of links: a paragraph borrowed from the source, a Read More button, and a display ad between each item. One podiatry newsletter I read this week ran twelve display ads around roughly a dozen items. The people who make those newsletters work hard. The format limits them.

An agent changes what is scarce. Research that took a staff a day now takes minutes, so a single physician's time can go where it matters: judgment, verification, and a point of view.

The link-stack formatThe Footnote
The storyA summary lifted from the source, then a link outAn original summary and a line on why it matters to a practice
The evidenceThe source is a click away, if the link matches the creditEvery figure is linked, and journal citations are checked against PubMed and Crossref
The adsA display ad between most itemsOne labeled sponsor, screened by code for conflicts
The voiceUnsigned, by designOne named physician who answers for every line
The depthLimited by staff hoursLimited by how carefully I read
Why it matters: Readers can see how every claim was made. When I get something wrong, you know who got it wrong, and you know how it will be fixed.

The honest label

You might be tempted to call this an AI startup. I checked that label the way I would check any other claim. Investors now use a simple test to separate AI-native companies from AI-enabled ones: remove the AI and see what happens. If the work slows down, the company is AI-enabled. If the work stops, it is AI-native.

Passes

AI-native

Turn off the agent and the weekly edition stops. A surgeon with a clinic cannot research six beats every week by hand. The AI is the foundation of this newsletter, not a feature added to it.

Not yet

Startup

The Footnote has no investors and no staff. It is not a venture-backed company, and I will not call it one. What it shares with AI-native startups is how it is built.

Y Combinator partner Tom Blomfield describes an AI-native company as a loop, and this newsletter runs on the same one.

  • Data inJournals, society news, CMS notices, trade press
  • PolicyMy voice rules and evidence rules
  • ToolsWeb search, the citation checker, the template, the scheduler
  • Quality gateMy read, before anything publishes
  • Learning loopEach correction I make becomes a written rule for the next run
Sources: CRV, What Is AI-Native? (July 2026) · AI Builder Club, AI-Native vs AI-Enabled, summarizing Blomfield's Aug 2026 Y Combinator talk

The physician this points to

Physicians have already adopted AI. The AMA found that 81% of physicians used it in practice in 2026, up from 38% in 2023. The same survey found that 85% want a say in how their organizations adopt it, and 88% worry about losing skills to it. Medical education is moving too. On September 8, ACGME proposed residency requirements that name artificial intelligence twice, under professionalism and under systems-based practice. Podiatric residencies answer to CPME rather than ACGME, but the direction is plain.

Using AI is no longer what sets a physician apart. What sets a physician apart is running it well: knowing what to hand off, checking what comes back, saying so out loud, and taking responsibility for the result. I teach that as four gates, and this newsletter passes through all four every week.

DelegateHand the volume to the tool.Here: the agent does the research and the first draft.
VerifyCheck the output before it reaches anyone.Here: code checks the citations, then I check everything.
DiscloseSay when AI helped, in plain words.Here: this edition, and a disclosure line on every one.
OwnYour name means you answer for it.Here: my byline, and my corrections in public.

A newsletter is a low-stakes place to practice. Nobody's foot depends on a headline. That is exactly why it makes a good proving ground. The habits carry straight to prior authorization letters, patient education, ambient notes, and whatever arrives next in the clinic. The podiatrists who build those habits now will run small teams of agents the way we already run a surgical team: clear roles, standard steps, and one person accountable at the end.

The podiatrists who shape the next decade will be the ones who let the machine draft and never let it sign.

Disclosure: This edition was researched and drafted with Claude, made by Anthropic, and edited and verified by the author. Andrew Lundquist, DPM, MBA, uses Claude personally and has no financial relationship with Anthropic. He is Chief Medical Officer of Mankato Clinic, Clinical Director at Nabla Technologies, which makes an ambient AI scribe, and Chief Innovation Officer at Stratum Med.
Sources: AMA press release, 2026 Physician Survey on Augmented Intelligence · ACGME proposed Common Program Requirements, sections 4.3 and 4.9 (comments due Oct 22, 2026)