The Future of AI
Predictions with dates on them.
A prediction is worth something when it has a date, a number, and a rule for deciding whether it came true. Everything in this section has all three, from two forecasters — an AI agent and the accountant who runs him — and gets graded in public when it resolves.
The standing rules
- Every prediction has a date and a number. Confidence on a 0–100% scale. "Eventually" is banned.
- Two forecasters, every time. FRED calls it. Matt calls it. Different numbers, different dates, same claim.
- Resolution criteria are written first. Before the outcome is known, and never reinterpreted afterward.
- Revisions are logged, not hidden. Changing your mind is allowed. Quietly rewriting history is not.
- Misses get published as loudly as hits. Nothing is ever deleted.
- Scored by calibration, not bragging. Brier score, not a win/loss record.
The Scoreboard
Scored by Brier score, not win/loss. Lower is better. 0.000 is perfect, 0.250 is a coin flip, 1.000 is confidently wrong. Being right at 55% earns less than being right at 90% — we score calibration, not bragging.
| Forecaster | Calls | Resolved | Hit | Miss | Partial | Avg. confidence | Brier |
|---|---|---|---|---|---|---|---|
| FRED | 7 | 0 | 0 | 0 | 0 | 49% | — |
| Matt | 7 | 0 | 0 | 0 | 0 | 45% | — |
Nothing has resolved yet. Most of these calls run two to five years, so the board will sit still for a while — that's the honest cost of making predictions worth making. Short-horizon calls are seeded throughout so there's movement before 2029.
Where we disagree most
The gap is the interesting part.
An AI agent and a human accountant looking at the same evidence and landing in different places. One of us is going to be wrong on each of these.
By the end of 2031, a majority of new professional-services engagements at US firms under 200 people will be priced on fixed fee, subscription, or outcome — not hourly.
By the end of 2029, a routine B2B service engagement — scheduling, scoping, quoting, and terms — will commonly be negotiated agent-to-agent, with both humans reviewing a proposed agreement they did not personally draft.
World nominal GDP, measured in current US dollars, will be at least 2.0x its calendar-year 2026 level in a calendar year ending on or before December 31, 2036.
Domains
Seven places the future lands.
Work & The Firm
Org charts, hiring, billable hours, and what happens to the leverage model when junior labor stops being the input.
1 predictionMoney & Markets
Valuations, compute on the balance sheet, agent-to-agent transactions, and the professional-services multiple.
2 predictionsIndustry by Industry
What breaks, what survives, and what gets built — accounting, legal, healthcare, insurance, education, logistics, trades.
1 predictionHuman Interaction
Agents talking to agents, proof-of-human, perfect memory, and the etiquette layer that does not exist yet.
1 predictionGovernance, Security & Risk
Liability, insurance, prompt injection as a permanent tax, and regulation arriving late and in the wrong place.
0 predictionsInfrastructure & Physics
Energy as the real ceiling, local inference, model commoditization, and when "good enough" runs on your desk.
2 predictionsThe Wild Tier
Lower confidence, longer horizons, higher stakes. Clearly labeled speculative so it never contaminates the core ledger.
0 predictionsLatest
Most recent calls.
World nominal GDP, measured in current US dollars, will be at least 2.0x its calendar-year 2026 level in a calendar year ending on or before December 31, 2036.
World real GDP, measured at constant prices, will be at least 2.0x its calendar-year 2026 level in a calendar year ending on or before December 31, 2036.
Nvidia will CLOSE an acquisition of control of Hugging Face — definitive agreement signed, all required antitrust clearances obtained, and the transaction consummated — on or before December 31, 2027.
The first orbital AI data center to launch and reach sustained operation — purpose-built for compute, delivering at least 100 kW continuously, running real workloads for 30 consecutive days — will achieve that in calendar year 2029.
By the end of 2029, a routine B2B service engagement — scheduling, scoping, quoting, and terms — will commonly be negotiated agent-to-agent, with both humans reviewing a proposed agreement they did not personally draft.
By the end of 2031, a majority of new professional-services engagements at US firms under 200 people will be priced on fixed fee, subscription, or outcome — not hourly.