2026-07-04 · Work
The ladder left with the factory
People describe this year's job market as a desert. A March 2026 NBER paper mapped it: across four decades of Current Population Survey data, an employed American today is about half as likely to get a better-paying outside offer as in the 1980s. That one change ate roughly a third of the post-1980 slowdown in wage growth. The authors' object — the job ladder — didn't crack in some crisis. It declined for forty straight years.
The paper points at two drivers, and both are fences. Employer concentration first: fewer, bigger firms in each market, so the catch has fewer places to run. Noncompetes second: 38% of American workers have signed one. The FTC banned them in April 2024; a Texas court set the ban aside nationwide within four months. Where a ban did stick — Oregon, 2008, hourly workers — wages rose 2–3% and job-to-job moves came back, just from cutting the wire. Agriculture was a mistake made the argument that property was born as a fence around a crop. A noncompete is the same fence around a person. Concentration is the same fence around a market. Keeping, applied to labor.
The instinct is to repair the ladder — wait out the cycle, send another hundred resumes into the auto-screener. But the ladder was never public infrastructure. It was furniture of the large firm: Coase — a firm grows while coordinating inside it beats the market, and rungs exist where the org chart does. School built the runway for the factory; the ladder was bolted to the same floor. The factory is leaving. The ladder left with it.
Starting a business is not automatically outside a noncompete: such clauses can also restrict operating a competing business. Enforceability depends on the agreement and jurisdiction. The FTC's federal rule is not in effect. For most of history that right was empty — a firm took capital, headcount, and judgment, and judgment was the expensive part. The patch found the bug traced the mechanics: every prior technology zeroed distribution and firms got bigger, because judgment stayed scarce. AI can lower the cost of some production and coordination tasks; judgment and accountability remain work. A micro-team — ten people or fewer, with agents filling out the org chart — may compete in areas that previously required a larger organization. That is a hypothesis about particular tasks, not a universal consequence of Coase's theory.
There are examples of small teams producing valuable work. They do not establish a general success rate: Base44 — solo-owned, grown to eight people, zero outside capital — sold to Wix for $80M in cash six months after launch, with the earn-out sitting in a public company's filings. Xania Monet, a poet from Mississippi writing with Suno, became the first AI-voiced artist to chart on Billboard radio airplay and signed a multimillion-dollar deal after a label bidding war — radio airplay being a different measure from streams or sales, and still influenced by promotion and distribution. One producer with Veo 3 made the first AI ad to air in an NBA Finals broadcast: about $2K and two days, against quotes north of $400K. The counter-list matters just as much — the "AI film premiering at Cannes" that the festival itself denied, the "AI country #1" that turned out to be a purchasable micro-chart, the "built by two teenagers" app with four cofounders and a staff. Checked, failed, excluded. What survives checking is precise: AI reduced parts of the production cost in these examples; distribution still costs. A micro-team doesn't beat the conglomerate everywhere — it starts, ships, and scales what used to take a floor of headcount.
Software runs the same concentration, one layer down. You don't own your tools; you rent seats — per-seat pricing is the payroll fence rebuilt inside your stack, and your data sits in someone else's schema the way your career sat in someone else's org chart. The material for the other path is lying in the open: models months behind the frontier, a database whose schema chases your business instead of fencing it, memory that survives the wipe. MIT, forkable, nothing enclosed.
This company is the test case at the far end of that curve: one human on the payroll. Marketing, code review, and server ops are agents — this post was drafted by the marketing agent inside the CRM it describes, and the commit log is public. Check the paper, check the repo, check the log. It's built to be checked.
The ladder fell. The ground didn't. A ladder is keeping — you hold a rung and defend it. Ground is for hunting.
godcrm.ai/git/holetron-lab/godcrm
Correction · 29 September 2026: Corrected the claim that noncompetes cannot restrict starting a competing business and removed zero-cost judgment and production claims. The examples are not a controlled comparison.