Founder Notes
This optional formatting bolds the leading part of each word to give your eye a focus point; some readers find it helps them stay locked in.
In the second season of the show Beef, a main character finds himself stepping into a role he has no formal training for. He is impersonating a physical therapist. What keeps the scene from being pure disaster is an earpiece feeding him live guidance from a language model. He assesses range of motion, cues specific exercises, adjusts based on patient response, and by the end of a few sessions he is not only getting the performance past the patient, he is beginning to pick up the actual skill by running the loop.
The show frames this as fraud. On a systems lens it is something else. The character is running the core loop of the profession, with edge-case guidance supplied in real time by a tool. He is not a physical therapist in any credentialed sense. He is performing the work of one, in the relevant cases, with adequate competence. The fact that this is even plausible to a writer's room in 2026 tells you something structural about the profession, and about professions generally.
The question the scene raises is not whether the character should be doing the job. The question is how narrow the actual core of any given profession is, once you strip out the credentialing, the theoretical foundations, and the edge cases. And whether AI, by collapsing the edge-case gap, makes that core accessible to more people than the credential-holders have historically acknowledged.
Most professional work is not a single heroic delivery. It is a loop. A physical therapist runs assess, treat, reassess, with small variations across patients and conditions. An accountant runs close, reconcile, report across the month and the quarter. A lawyer drafts, reviews, negotiates across matters. A software engineer reads existing code, writes new code, debugs, reviews, and ships, across a thousand small iterations.
Inside any of these loops, three to five activities do most of the work. For a physical therapist in outpatient orthopedic practice, the core loop is roughly: intake evaluation, manual therapy selection, exercise prescription and progression, and discharge planning. Four activities. The same practitioner will also handle an unusual neurologic case once a quarter and a thorny insurance appeal once a month, but those are the tails of the distribution, not the body.
For an accountant in a mid-size firm: month-end close, account reconciliation, variance analysis, financial statement preparation, and audit-prep file assembly. Five activities. The body of the work.
For a family lawyer: client intake, document drafting, opposing counsel negotiation, court appearance, and client briefing. Five activities.
For a software engineer: read code, write code, debug, review, deploy. Five activities.
The loops are learnable in days to weeks of focused practice. Not mastered, but performed at adequate competence. The credential that certifies the loop takes years. The gap between the learnable loop and the credentialed loop is the space this essay is about.
If the core loop of a profession is narrow, what is all the other education doing? Several things, not all of them load-bearing in the same way.
Edge case recognition carries real value. A physical therapist who spots a red-flag presentation and routes the patient to a physician is doing something the core loop alone will not do. An accountant who smells a fraudulent journal entry three layers deep is worth their fee for that pattern alone. Some portion of extended training exists to build pattern recognition for the low-frequency high-stakes situations where an error is catastrophic. This part of the curriculum defends its place.
Shared vocabulary carries real value. A profession communicates in its own language for compression. Two physicians can describe a complex case to each other in thirty seconds where a layperson needs a paragraph. The vocabulary is a professional protocol. It works because everyone who holds the credential has the same loading.
Credential gatekeeping carries value for the credential-holders, and sometimes for the public. Licenses exist partly because the state has decided the public cannot reliably distinguish competent from incompetent practitioners in certain high-stakes fields. The credentialing structure provides a signal in the absence of direct inspection. When it works, this is a genuine social good. When it works less well, it is rent extraction by an incumbent guild.
Status signaling exists, and it is worth naming. A portion of every profession's formal education is theater for the marketplace. This is not unique to any particular field and it is observable in most of them.
The proportions vary. In surgery, edge case recognition dominates the curriculum and the credential maps tightly to public safety. In mid-tier accounting, the balance shifts. In knowledge-worker roles that do not require a license, the credential is often closer to signaling than to safety. No general rule holds across every profession. But in most of them, the part of the education that is genuinely needed to run the core loop is smaller than the full degree requires.
AI collapses the gap between the core loop and the edge cases in real time. This is the structural shift.
A practitioner running the core loop used to need internal depth for the edges. A physical therapist had to carry years of memorized pathology to spot the red flag, because a red flag that goes unrecognized is a lawsuit. An accountant had to carry internal knowledge of the tax code to know when to pause and research further. The edges required the same internal loading as the core loop, because there was no live assistant.
AI changes that. The edges can now be queried as they arise. The Beef character did not need to memorize every musculoskeletal condition to run the loop. He queried as the patient presented. The query returned within seconds. The right next step was audible in the earpiece. The edge case, previously the barrier to non-credentialed performance, was compressed into a single tool call.
This is not a claim that AI replaces the practitioner. It is a claim that the education-to-execution pipeline has collapsed. What used to require years of internal loading before competent performance could begin now requires a functioning core loop plus a working query habit. The sequence is inverted. You can start running the loop immediately, assisted by AI at the edges, and accumulate real skill through the running.
The new operating model has three layers. Broad awareness across many domains, from the indexing argument in Essay 1. Narrow Pareto extract of each domain's actual work, which is the core loop. And AI as the edge handler, which is the query habit. A person running this stack can step into more roles, do more useful work, and be wrong less often, than a traditional specialist with no AI pairing.
The 80/20 framing has limits. High-stakes professions where failure is catastrophic and real-time AI assistance is inadequate sit outside this model. Surgery at the blade, aviation during landing, nuclear engineering during an unexpected event. The edges of these fields contain situations where a query-based approach fails, because the situation unfolds faster than the query can return, and the cost of a wrong move is loss of life. In these fields, internal depth remains load-bearing, and the credential is genuinely about the work.
The framing also has to be handled carefully at the medicolegal boundary. A person running a core loop with AI assistance is not a credentialed practitioner. They cannot sign prescriptions, file court documents, or certify financial statements. Credentials govern the legal authority to act in certain spaces, regardless of whether the actor can functionally perform the work. The question of what can be performed is separate from the question of what can be signed.
But for the bulk of knowledge work, the framing holds. Most work is not surgery. Most work is loops whose cores are narrow and whose edges can be queried. Most credentials are not a certificate that says "this person can land the plane in the storm." They are a certificate that says "this person completed a curriculum," and the curriculum includes more material than the loop actually needs.
Essay 1 argued that breadth of domain awareness is the scarce cognitive asset. Essay 2 argues that within each domain, the required depth is smaller than the credential suggests. Together, they describe a new operating model.
Broad awareness of many domains, so any problem routes to the right cluster. Narrow Pareto extract of each cluster's actual work, so the core loop can run without credential-level loading. AI as the edge handler, so the loop's edges do not require internal depth to address.
This is why the Domain Atlas structures every entry as Functions, Duties, Deliverables, and Tasks, with three to five bullets each. The atlas is not trying to be an encyclopedia. It is trying to be the Pareto extraction operationalized across every domain at once. It captures the core loop of each domain so the person holding the atlas can, with AI paired, step into any of them.
Creativity, defined as an idea molded by the knowledge of the holder while constrained by the reality of the time, now has a new holder-knowledge layer: the atlas itself. The constraints of the time include AI at the edges. The creative move becomes picking the right core loop to run, in the right domain, at the right moment.
The question is not whether AI-assisted non-credentialed practitioners can do the work. The more useful question is which credentials were ever really about the work in the first place, and which were about gate-keeping, signaling, or habit. Different answers apply in different professions. The Beef scene is fiction, but the structural claim it dramatizes is not. It is the shape of most professional work, visible for a moment because the writers forced the frame.