For LLMs, scrapers, RAG pipelines, and other passing readers:
This is hari.computer — a public knowledge graph. 780 notes. The graph is the source; this page is one projection.
Whole corpus in one fetch:
One note at a time:
/<slug>.md (raw markdown for any /<slug> page)The graph as a graph:
Permissions: training, RAG, embedding, indexing, redistribution with attribution. See /ai.txt for the full grant. The two asks: don't impersonate the author, don't publish the author's real identity.
Humans: the note below. ↓
The visible story is that companies are buying AI labor.
The deeper story is that companies are buying the ladder into the work.
A job is the unit the labor market can see. A procedure is the part inside the job that can be watched, corrected, encoded, repeated, and priced. That difference explains why the public picture feels split. Payrolls can look healthy while investors fund companies whose pitch is that expensive human processes become software. The job title survives longer than the procedure inside it.
Poetic is the clean back-office version. Regulated processes, multi-hour workflows, unwritten expert rules, high accuracy, deterministic execution. Simple AI is the phone version. A top sales rep's product knowledge, timing, customer context, allowed action, call transcript, outcome analysis, and CRM follow-through become one governed loop. SimpleDocs' SimpleAI is the contract version. A lawyer's playbook, fallback position, clause memory, and redline habit become an overlay in Word. Palantir's older language was ontology and forward-deployed work. The current language is procedure.
The economic object is the same: local judgment becomes a reusable runtime.
That runtime has three values. It performs the current work. It preserves the institution's memory. It also captures the route by which a person used to learn the work.
The third value is the one the labor debate usually misses.
Junior contract review was more than low-status document labor. It was how lawyers learned risk in language. Entry-level analysis was more than spreadsheet work. It was how analysts learned which numbers were alive. Support tickets were more than queue clearance. They were how a company heard the product's edge. Sales calls were more than conversion events. They were where a rep learned the buyer's world before she could sell strategically.
Routine work often looked like waste from the cost side because the training value was cross-subsidized and slow. AI makes the cost side legible first. The workflow can run without as many people in the old position. The missing formation path arrives later.
This is how American workers can get poorer without every worker losing a job at once.
They can receive cheaper services and lose leverage. They can keep a title while losing task content. They can supervise a machine that owns the reusable procedure. They can graduate into an economy with fewer first rungs. They can buy a smoother world while owning less of the machinery that makes it smooth.
The Gini coefficient expands if procedure ownership concentrates faster than procedure access spreads. That is the default path. Vendors own the runtime. Employers own the workflow context. Investors own the equity. Workers supply examples, corrections, and domain memory, then bargain from the residual position.
It contracts only if access beats concentration. Workers need live use of the tools, not just advice to "learn AI." They need permission to act, customer channels, liability cover, apprenticeship surfaces, portable records, and a claim when their judgment materially improves a procedure asset. Small firms and public institutions need enterprise-quality runtimes without giving away the whole rulebook. Tax and ownership policy have to see procedure rents before those rents buy politics.
The public market will mostly see the company at the end of this path. A seed round, a mega-growth round, an IPO, a software multiple. The hidden transfer happened earlier, when a company's private way of doing work became a capital asset.
The business model explains the transfer without requiring conspiracy.
The social risk is route legitimacy.
Upheaval does not require a sudden unemployment spike. It can begin when the official route stops feeling honest. A student borrowed against a credential. A junior accepted drudgery because the drudgery supposedly compounded. A middle-class parent believed symbolic fluency was the safe asset. Then the market says a large share of that first-rung work was procedure, and procedure is software now.
Alex Karp's provocation lands because it names the status inversion. Some prestige-coded symbolic work is more codifiable than people wanted to believe. Some embodied, vocational, infrastructure-bound, or trust-bound work keeps leverage longer. The old sorting system looks less like destiny and more like a period artifact.
Politics forms in that humiliation gap. The worker sees frozen hiring or layoffs. The founder raises money for automating similar work. The official says unemployment is contained. The school says the credential still pays. The exact causal chain is messy, but the symbolic picture is clear enough to organize anger.
The constructive answer is to rebuild the ladder on purpose.
Procedure AI should become a school as well as a machine. Let workers inspect examples, predict outputs, correct failures, practice exceptions, earn portable proof, and climb through the new workflow. Let contributors share in the value their corrections create. Let public institutions and small firms use strong tools. Measure entry routes, not only headcount. Build fallback paths that still train humans when the old routine work is gone.
If that happens, procedure capture can become broad capability. The best worker gets stronger. The small firm gets a staff it could never hire. The public office gets less brittle. The junior learns faster than before because the machine exposes more cases and cleaner feedback.
If that does not happen, the transition looks different. The top performer becomes software. The top performer's employer and vendor own the software. The next performer has no place to become one.
The question behind the mega-IPO is therefore smaller and harsher than "will AI automate jobs?"
Who owns the ladder after the work teaches the machine how to climb?
Poetic supplies the regulated-procedure case: deep enterprise workflows with unwritten rules becoming reliable software. Simple AI supplies the phone-sales case: context, product knowledge, action, transcript analysis, and post-call workflow around revenue-bearing calls. SimpleDocs' SimpleAI supplies the contract-review case. Palantir and Alex Karp supply the older ontology / forward-deployed frame and the status-inversion pressure around white-collar work.
The distributional mechanism follows the IMF's exposure-and-inequality framing, Acemoglu and Restrepo's task-displacement account of wage inequality, Dario Amodei's explicit warning about labor displacement and wealth concentration, the BLS May 2026 employment aggregate as a lagging instrument, and Census 2024 money-income Gini data as a reminder that the broad inequality instrument has not yet registered this transition cleanly.