Vault
analysis

The 2030 Landing Posture: What "Treat Post-2032 Income as a Bonus" Actually Means

Created

The 2030 Landing Posture: What "Treat Post-2032 Income as a Bonus" Actually Means

Builds-on: earnings-potential-40-to-65, the-orchestrator-premium Related: aschenbrenner-thesis-audit, mo-gawdat-dystopia-thesis-audit, staff-engineer-job-market-2026

Written after stress-testing the career thesis against Daniel Kokotajlo's July 2026 Diary of a CEO interview (ex-OpenAI, AI Futures Project; median superintelligence ~2029, 70% catastrophe estimate, "Plan A" halt-and-regulate proposal). Two questions answered here: does Kokotajlo break the orchestrator-premium analysis, and what does "plan to land by 2030" operationally mean — job loss, or no more upside?

1. The Kokotajlo Stress Test

Where he hits the thesis dead-on

The orchestrator-premium doc load-bears on one assumption: LLMs are a leaky abstraction, so value concentrates in the human trust layer (specs in, verification out), eroding task-by-task as tooling matures. Kokotajlo's claim — that labs are explicitly automating their own research and coding first, before diffusing anything — attacks that assumption directly, and it's partially observable: the Anthropic Economic Index shows Claude Code at 79% automation-classified usage, with coding migrating to autonomous API workflows faster than any other domain. The execution layer is being eaten first because the labs are eating their own layer first.

The dates converge from opposite directions. The vault's window (premium compresses ~2029-2031 via tooling maturity and title consolidation) and Kokotajlo's median for the recursive-self-improvement sprint (~2029) land on the same calendar through unrelated mechanisms. When a market-normalization model and a capability-takeoff model agree on the date the current advantage stops mattering, treat the date as the robust output.

Where mid-2026 evidence pushes back

The aschenbrenner-thesis-audit finding transfers: right on inputs, behind on outputs. Kokotajlo's track record earns respect (his 2021 "What 2026 Looks Like" aged well; refusing the non-disparagement clause with $2M at stake is a costly signal that separates him from thesis-sellers). But AI 2027's superhuman-coder timeline is visibly slipping — METR's best current measure is an 18% assisted speedup with persistent reliability gaps, not autonomous researchers — and his own median moving to ~2029 in this interview is that slippage, acknowledged.

The strongest counter-evidence is a revealed preference: the billion-dollar FDE ventures. If models were close to deploying themselves into messy organizations, OpenAI would not have acquired a 150-person deployment firm and Anthropic would not have raised $1.5B with Blackstone to embed humans in enterprises. The most capability-pilled institutions on earth are paying premium prices for exactly the layer Kokotajlo says is about to be automated. "Automate coding" and "automate software engineering" are different claims with different timelines — the daily Brightwheel experience (bottleneck is org shape, never model capability) is the micro version.

His explanation for absent mass unemployment (capability hoarded upstream for lab R&D, diffusion deferred) cuts both ways: slow diffusion isn't a delay of the orchestrator's obsolescence — it is the orchestrator's job.

The three worlds, and the robust move

World Content Career implication
Capability right, control lost The bulk of Kokotajlo's 70% No career strategy exists; rational allocation is civic (awareness, regulation), not portfolio moves. Drops out of planning by construction.
Capability right, control held Citizen's-dividend world; wages die broadly What you own by 2030 is what matters: equity, assets, identity, network. Convert W-2 to capital now.
Capability slower (METR/friction world) Premium plays out as modeled, compresses 2029-31 Capture the premium now; build identity that carries income past the window.

Every plannable world prescribes the same actions. Kokotajlo doesn't change the plan — he moves probability mass toward the worlds where it's urgent rather than prudent. His one forceful addition: the labor market reprices late (capability hoarded upstream means job postings lag reality), which argues for doing the legibility work this quarter, not next year.

The tells to watch

  1. METR annual measurements crossing from "supervised speedup" to "unsupervised multi-day tasks."
  2. Self-verification maturing to where eval harnesses design themselves.
  3. Cleanest: the labs shrinking their FDE ventures because models deploy themselves. While Anthropic/OpenAI grow those human-deployment bets, the trust layer is still human-priced. The week they stop hiring for them is the week the window closes early.

Calibration note on the 70%: per the mo-gawdat-dystopia-thesis-audit method finding, expert catastrophe numbers are expressions of deep uncertainty, not frequencies. Kokotajlo's is the most credible high number in the distribution; the mechanism (RSI on coding) is real and partially observable; the timeline is the part that has already slipped once. Plan on the window; stay agnostic about what's on the other side.

2. What "Land by 2030" Operationally Means

It is neither a prediction of job loss nor exactly "no more upside." It's a planning posture: build the family plan so it still works if real income never grows after ~2030, and treat growth that does arrive as upside you didn't need. It's the career version of the TIPS book — insurance bought for asymmetry, not certainty. Wrong → over-saved with extra optionality at 45 (cheap mistake). Right and unprepared → discovering it at 48 with two kids in school (expensive one).

Job loss vs. no-upside, by scenario (earnings-doc weights)

The compounding trap: flat nominal comp inside macro scenario A (stagflation grinding 3.5-4%) is a 15-20% real pay cut by the mid-2030s without a single bad review. The career thread and the macro thread are the same erosion seen from two sides.

Reality check: "no upside via promotion" is substantially true today — principal freezes, management flattening, structured equity — pre-superintelligence. The 2030 posture is mostly extrapolation of a visible repricing, with the Kokotajlo tail bolted on. The plan doesn't require believing him; it just stops requiring him to be wrong.

The three operational moves

  1. Run the retirement math with real income flat from 2031. Household ~$370k, ~$50k/yr into retirement accounts, ~$215k there now → plausibly $600-800k by 2030; kids 8 and 6 (childcare drain aged out); mortgage recast at $6,876. Rough read: that trajectory reaches "fine." It does not reach 余裕 — which is what moves 2 and 3 are for.
  2. Front-load leverage-consuming moves into 2026-2028. The job switch for a real comp bump, the premium capture, the consulting experiment — spend the leverage while the market still bids for the layer, not in 2031 after repricing. Same reason the legibility work (vocabulary rewrite, evals post, target-list refresh) is a this-quarter item.
  3. Build what pays after the escalator stops. Late-career income tracks the public identity built in the 40s, not technical skill. Surviving post-2030 income channels are ownership and identity — equity in something, audience, teaching/consulting muscle, being findable for judgment rather than employable for output. The blog was already the strategy; this gives it a deadline.

Open Questions

Sources

Backlinks