The Next Fifteen Years, From First Principles#
A forecast of AI and the world economy through 2040: one argument across ~100,000 words and 90 pages, with scored probabilities, named falsifiers, and a quarterly indicator dashboard.
Written 2026-07-30. Quantities reflect knowledge through mid-2026. Probabilities are subjective.
Read it at future.lulzx.space. Short on time? Start with the compressed version.
Built from physical constraints upward, then the strategic games that sit on top of them, then domain by domain, then timelines.
The argument has one spine: capability grows fastest where verification is cheap, and value accrues to whatever intelligence cannot manufacture. Everything below is a consequence of those two sentences colliding with physics, capital, and incentives.
Three design choices distinguish this from most AI forecasting, and none of them is decorative:
- Bottom-up from constraints, not top-down from capabilities. Instead of asking "what will models be able to do," it asks what the four physical inputs permit, and lets capability fall out as a residual. Constraint-first reasoning is harder to bend toward a preferred conclusion, because queues, order books, and capex guidance are checkable in a way that vibes about model quality are not.
- Every major claim is paired with a way to catch it failing - a falsifier in Part VI, an indicator in Part VII, or a scored probability in Part V. A forecast that cannot be caught failing is marketing.
- One argument, not a scenario fan. Rather than hedging across futures, it commits to a single causal chain and then attacks that chain in the open (steelman, uncertainties). The cost of that choice is brittleness where the chain is wrong; the dependency index exists so that a break propagates honestly instead of being patched leaf by leaf.
Contents#
00 - Overview#
- Thesis - the argument in one page
- The compressed version - if you read nothing else
- Reading orders - linear, argument-first, sceptic-first, operator-first
- Notation - recurring shorthand
- Dependencies - reverse dependency index
01 - The Physical Substrate#
What actually constrains this. Four inputs, four different ceilings.
- Compute - scaling curves and the capex wall
- Energy - the constraint that bites before capital does
- Data - exhaustion and the verification asymmetry
- Capital - not a ceiling but a verdict, and the only input that can reverse
- Inference economics - the two-year moat
02 - The Games#
Five strategic structures that determine behavior given the substrate.
- Game 1 - Labs: a Tullock contest, not a prisoner's dilemma
- Game 2 - Nations: a security dilemma with a leaky bucket
- Game 3 - Firms: the Red Queen, and why adoption doesn't mean profit
- Game 4 - Labor: comparative advantage doesn't guarantee a wage
- Game 5 - Information: the collapse of costly signaling
03 - Domain by Domain#
Domains in three groups, ordered by the cost of ground truth.
- A - Cognitive - output is symbols: software · law · finance · science · education · media · insurance · startups (ideas) · meaning
- B - Physical - output is atoms: medicine · robotics · energy sector · agriculture · logistics
- C - Contested - an adversary optimizes against you: geopolitics (US–China · India · Europe · Gulf · Global South) · warfare · cybersecurity · biosecurity · state capacity
04 - Timelines#
- 2026–2028 - the agentic transition and the capex test
- 2028–2032 - diffusion and the institutional lag
- 2032–2040 - the physical world
- Scenarios - best case, base case, worst cases, and the named branches between
05 - Subjective Probabilities#
The numbers, re-scored against the corpus; per-row reasoning, ledger, and the register of probabilities stated elsewhere.
06 - Where This Is Most Likely Wrong#
- Recursive research acceleration - highest parameter variance
- Power permitting politics
- The apprenticeship gap after institutions respond
- Taiwan / leading-edge fab - invalidates the document if it fails
- Learned verification - framework-level risk
- Correlated-failure insurability - deployment frontier as balance sheet
07 - The Indicator Dashboard#
Every claim above, converted into something checkable on a quarterly cadence.
- A - Substrate: are the ceilings binding on schedule?
- B - Diffusion: is it actually reaching the economy?
- C - Governance: which architecture do we get, and when?
08 - Method#
Where the numbers come from, and how to tell when this is wrong rather than early.
- Base rates - the reference classes underneath every estimate
- Steelman - the three strongest arguments against the framework itself
- Scoring - resolution rules, and the five ways this document will try to dodge
09 - The Macro-Financial Channel#
What happens to rates, prices, assets, and the tax base if any of this is true.
- Rates and returns - why success is self-limiting
- Prices and the two economies - the defining political-economy fact of the 2030s
- Assets and distribution - six of the seven inelastic complements are owned by capital
- Fiscal - a tax base built on labor income, meeting an economy that uses less of it
- Demography - the one already-determined force, and it points the other way
Indicators to watch#
The five that matter most. Full dashboard with trigger values in Part VII.
| Indicator | What it tells you | Detail |
|---|---|---|
| Entry-level : senior posting ratio in knowledge professions | The leading edge of the labor transition | Game 4 |
| Internal research cycle time per validated experiment | Whether the RSI loop is closing | Uncertainty 1 |
| Teleoperation-to-autonomy ratios in deployed robots | Real robotics progress, vs. demo videos | Robotics |
| AI revenue run-rate vs. capex | Whether the 2027–29 correction lands | Compute |
| Grid interconnection queue length | Whether the energy ceiling is loosening | Energy |
Read linearly by following the Next link at the foot of each page, starting from Thesis.
This corpus is maintained by a repeating review-and-expand loop; see the autoresearch protocol. On the site: press / or use Find to search all pages.