01
Outside view
Start with the base rate. What does history say happens to claims like this one?
Why Predict At All
Position over chase.
A great openside flanker — the number 7 — doesn't chase the ball. They read the play. They watch the opposition's shapes, the first-five's body language, where the pods are aligning, and they move into the space where the next meaningful collision will happen. One correct anticipation produces a turnover or a linebreak — worth more than three late tackles. Prediction in business and economics runs the same way. The point isn't to be right about every event. It's to read the patterns early enough to put your time, attention, and capital into the zone where the next shift will land. Position over chase. Decide the next moment rather than react to the last.
The Method
Three views. One probability. A date.
Every prediction below follows the superforecaster protocol. The discipline is the point — the bets are the proof that the discipline holds under pressure. Forecast date: 2026-04-14. Resolution date: 2028-03-28.
02
Inside view
Read current signals from named sources. Date them. Quote them.
03
Synthesis
Weight the two views. Produce a calibrated probability — not a hope, not a hedge.
04
Falsifiers
Name three conditions that would lower conviction. If you cannot name them, the claim is not yet a prediction.
05
Resolution date
Every prediction has a date it grades against. No moving targets.
The Machine
From micro to macro.
A single prediction is a bet on one play. The compounding power comes from running the discipline across every project — every deal, every feature, every workchart, every venture — and grading planned against actual at every step.
Our agency library encodes the discipline as deterministic algorithms and AI workflows. Deal probability from weighted moving averages. Planned-versus-actual alignment scoring. Template effectiveness measured across the cohort. Compound rate tracked over time — so the system tells us, honestly, whether it is getting sharper. Probabilistic workcharts cover the parts that aren't deterministic: judgement, scenario weighting, weak signal interpretation.
Each project is a micro prediction with a date and a resolution rule. The engine grades it against the outcome. The grade tunes the next prediction. At cohort scale, calibrated micro signals add up to a coherent macro picture — not by guessing the future, but by reading the play better, every play. The same engine runs on any project type: digital build, physical commissioning, venture launch, content release.
The Scope
Six domains
AI Trajectory
5 predictionsavg 65%max conviction 5/5
Crypto & DePIN
3 predictionsavg 45%max conviction 4/5
Work Transformation
2 predictionsavg 70%max conviction 4/5
Healthspan
1 predictionavg 75%max conviction 4/5
Geopolitics
1 predictionavg 70%max conviction 4/5
Education
1 predictionavg 60%max conviction 3/5
The Anchors
The spine of the forecast
The five most-convicted bets. Each links to its full card below — base rate, falsifiers, watch signals.
- Probability70%Conviction70%DomainAI Trajectory
- Probability80%Conviction80%DomainWork Transformation
- Probability40%Conviction40%DomainCrypto & DePIN
- Probability75%Conviction75%DomainHealthspan
- Probability60%Conviction60%DomainEducation
| Prediction | Probability | Conviction | Domain |
|---|---|---|---|
| AI agents handle 40% of knowledge worker tasks | 70% | 70% | AI Trajectory |
| AI content becomes commodity; human-curated becomes premium | 80% | 80% | Work Transformation |
| Crypto rails become default for agent-to-agent settlement | 40% | 40% | Crypto & DePIN |
| Healthspan becomes tracked consumer spending category | 75% | 75% | Healthspan |
| Prediction practice separates top-decile performers | 60% | 60% | Education |
| MCP/A2A becomes default agent API | 75% | 75% | AI Trajectory |
| Context graphs become the primary moat | 70% | 70% | AI Trajectory |
The Decomposition
Domain-by-domain
Each domain has its own base rate, its own signals, its own synthesis. The predictions below sit on top of this scaffolding.
AI Trajectory
DomainBase rate
Technology adoption S-curves compress with each wave. Electricity took 40 years to 50% household adoption. The internet took 15. Smartphones took 7. Enterprise AI went from ~10% (2020) to ~79% (2025) in five years — faster than any prior technology wave.
Current signals
- Enterprise adoption: 79% of organizations adopted AI agents in some form; only 1 in 9 runs them in production — Gartner/industry surveys, 2025
- Application embedding: 40% of enterprise apps will embed task-specific AI agents by 2026, up from <5% in 2025 — Gartner forecast, 2025
- Market growth: AI agents market reached $7.8B in 2025, projected $10.9B in 2026 (45% CAGR) — Market analysis, 2025
- Protocol maturity: MCP crossed 97M monthly SDK downloads; A2A launched by Google with 50+ partners; both donated to Linux Foundation AAIF — Anthropic/Google, 2025
Synthesis
Accelerating, with a governance gap. Adoption is ahead of infrastructure. The 79% adoption vs 11% production gap is the critical number — capability exists, plumbing lags. Installed base waiting for the unlock.
Crypto & DePIN
DomainBase rate
Financial infrastructure adoption follows a pattern: SWIFT took 15 years from founding (1973) to global dominance. Internet payments took 10 years from PayPal (1998) to mainstream. Crypto settlement is at year 5 of institutional adoption (post-2020 DeFi summer). Stablecoin volumes doubled in one year ($19T to $33T).
Current signals
- Stablecoin volume: $33T in 2025 (72% YoY growth), USDC leading with $18.3T — Bloomberg/Artemis Analytics, Jan 2026
- B2B payments: B2B stablecoin payments surged from <$100M/month (early 2023) to $6B/month (mid-2025) — Artemis Analytics, 2025
- Regulatory clarity: US GENIUS Act passed July 2025 — first comprehensive stablecoin framework — US Congress, July 2025
- DePIN market: ~250 DePIN projects, combined market cap $9.05B (CoinGecko 2026-04-27 — below the $10B falsifier threshold) — CoinGecko, 2026
Synthesis
Rails crossing from crypto-native to enterprise. The B2B payment surge ($100M to $6B monthly in 2 years) is the strongest signal. DePIN growth lags the WEF $3.5T 2028 projection — falsifier active.
Work Transformation
DomainBase rate
Historical automation waves displace tasks, not roles. ATMs didn't kill tellers — teller employment grew 10%. Spreadsheets didn't kill accountants — accounting grew. Roles reshape around remaining human-judgment tasks. Displacement concentrates in the most routine subset.
Current signals
- Tech layoffs: AI-attributed tech job losses reached 77,999 in H1 2025 — Industry data, 2025
- Junior developer impact: 20% decline in employment for software developers aged 22-25 vs late-2022 peak — Bureau of Labor Statistics, 2025
- WEF projections: 92M roles displaced by 2030, 170M new roles created (net +78M) — World Economic Forum Future of Jobs Report, 2025
- ROI evidence: Enterprise AI ROI averaging 171%; US enterprises 192% — 3x traditional automation — Industry surveys, 2025
Synthesis
Accelerating displacement in routine tasks, steady creation of new roles. Junior developer signal is the canary — entry-level knowledge work compressing first. Lump-of-labor fallacy still holds: more work created than destroyed, just different.
Healthspan
DomainBase rate
Fitness grew from a niche ($2B in 1980) to mainstream ($100B in 2020) over 40 years. Supplements followed a similar curve to $50B by 2025. Healthspan as a consumer category is where fitness was in ~1995 — early mainstream, growing fast. GLP-1 is the home-gym moment.
Current signals
- Market size: Longevity market projected at $29B in 2026, growing to $63–78B by 2033–2035 — Emergen Research / Business Research Company, 2025
- Deal size growth: Average longevity deal jumped from $20M (2023) to $69M (2025) — institutional-grade rounds — Longevity Market Q4 2025
- Consumer adoption: 50% of new dietary supplement SKUs launched since July 2024 carry a healthspan claim — Euromonitor, 2025
- Big pharma entry: Longevity went mainstream within big pharma in 2025 — metabolism, inflammation, aging biology connected through GLP-1 data — Lifespan.io expert roundup, 2025
Synthesis
Accelerating. GLP-1 is pulling longevity from biohacker niche to mass consumer. Deal-size jump ($20M to $69M) signals institutional conviction. Top-3 consumer category requires displacing structural categories (housing, food, transport) — top-5 by growth rate is the realistic reframe.
Geopolitics
DomainBase rate
Regulatory response to transformative tech follows a pattern: the internet operated in a regulatory vacuum for ~5 years (1995–2000). Social media had ~6 years (2012–2018) before GDPR. AI regulation began in earnest with EU AI Act (2024). Countries that find the middle (light-touch, principles-based) attract the most diverse capital.
Current signals
- NZ AI Strategy: Released July 2025 — last OECD member to establish one. Projected NZ$76B economic contribution by 2038 — MBIE, July 2025
- NZ approach: Light-touch, OECD-aligned, no standalone AI Act. Uses existing laws plus principles-based guidance — NZ Digital Government, 2025
- Treaty integration: NZ AI Strategy incorporates Treaty of Waitangi obligations — unique globally — NZ AI Strategy, 2025
- EU enforcement: EU AI Act enforcement began. High-risk AI systems face compliance requirements. First enforcement actions pending — EU, 2025–2026
Synthesis
Diverging into three models: EU (prescriptive, compliance-heavy), US (fragmented, sector-specific), NZ/Singapore (principles-based, light-touch). NZ's late entry is advantageous — learned from EU's over-regulation and US's under-regulation.
Education
DomainBase rate
Universities survived 800 years of disruption attempts. MOOCs (2012), bootcamps (2015) — none reached parity with degrees for hiring. But enrollment decline is real: US enrollment falling since 2010. Bifurcation, not collapse.
Current signals
- Enrollment decline: 13% projected decline 2025–2041. International enrollment fell 17% in fall 2025 — US higher education data, 2025–2026
- Degree skepticism: 63% of US voters say 4-year degree is not worth the cost — Public opinion survey, 2025
- AI skills demand: 57M Americans interested in AI skills, 8.7M currently learning. Ed-tech serves 99%+ of demand — Industry data, 2025
- Alternative funding: Workforce Pell (H.R.1, July 2025) allows Pell grants for 8-week credential programs — US Congress, July 2025
Synthesis
Bifurcating. AI-specific education booming; traditional declining. CS enrollment paradox (AI up 45% YoY, general CS down in 62% of programs) is the clearest signal. Workforce Pell is structural — first federal funding flowing to non-degree credentials.
The Bets
All predictions
13 predictions across 6 domains. Each card shows the statement, conviction, base rate, falsifiers, and watch signals.
AI TrajectoryANCHOR
AI agents handle 40% of knowledge worker tasks
By end 2027, AI agents will handle 40% of tasks currently performed by knowledge workers in enterprises with 500+ employees.
70%
Base rate
Prior automation waves displaced 20-30% of targeted tasks within 5 years of maturity.
Falsifiers
- Enterprise production deployment stays below 20% by end 2026
- Major AI governance failure triggers regulatory pause
Watch signals
- Gartner quarterly adoption surveys (production %, not experimentation %)
- AI agent market revenue actuals vs projections
AI Trajectory
10%+ of e-commerce flows through agent channels
By 2028, agentic commerce redirects 10%+ of e-commerce transactions through AI agent channels.
45%
Base rate
New commerce channels typically capture 5-15% of adjacent market within 3 years of infrastructure maturity.
Falsifiers
- MCP/A2A adoption stalls below 200M monthly downloads by end 2026
- No major e-commerce platform integrates agent purchasing by 2027
Watch signals
- First Fortune 500 retailer launches agent-native checkout
- Agent transaction volumes (when category tracking emerges)
Work TransformationANCHOR
AI content becomes commodity; human-curated becomes premium
By end 2027, AI-generated content will be indistinguishable commodity while human-curated content commands 3x+ premium pricing.
80%
Base rate
Every content medium that was automated (stock photos, template websites) saw 70-90% price compression within 3 years.
Falsifiers
- AI-generated content quality plateaus (detectable, generic)
- Premium human content demand declines (readers don't care about provenance)
Watch signals
- Premium publication subscription growth vs free AI summaries
- Provenance-verification standards adoption (C2PA, similar)
Crypto & DePINANCHOR
Crypto rails become default for agent-to-agent settlement
By 2028, crypto stablecoin rails are the default settlement mechanism for autonomous agent-to-agent commercial transactions.
40%
Base rate
New payment rails take 7-10 years from regulatory clarity to default status. GENIUS Act = year 0.
Falsifiers
- Stablecoin volume growth decelerates below 20% YoY
- No enterprise adopts crypto for agent transactions by end 2026
Watch signals
- Monthly stablecoin settlement volumes (Artemis Analytics)
- First documented agent-to-agent crypto settlement in production
Crypto & DePIN
DePIN networks exceed $50B market cap
DePIN networks collectively exceed $50B combined market cap by 2028.
30%
Base rate
Infrastructure networks with real revenue grow 3-5x in 2-year bull cycles. Current base: $9-10B (not $19B as of 2026-04-24).
Falsifiers
- Crypto bear market drops DePIN below $10B
- Helium subscriber growth stalls below 1M
Watch signals
- DePIN aggregate market cap (CoinGecko DePIN category)
- Helium quarterly revenue and subscriber growth
Work Transformation
Entry-level knowledge worker roles decline 30%
By 2028, entry-level knowledge worker positions (data entry, junior development, administrative) decline 30% from 2025 baseline.
60%
Base rate
Prior automation waves displaced 15-25% of targeted entry roles within 5 years.
Falsifiers
- Junior developer employment rebounds above 2022 levels
- Enterprise AI ROI drops below 100%
Watch signals
- Quarterly employment data by role category (BLS, LinkedIn Economic Graph)
- Entry-level job posting volumes (Indeed, LinkedIn)
HealthspanANCHOR
Healthspan becomes tracked consumer spending category
By 2027, at least two major market research firms track 'healthspan' as a named consumer spending category.
75%
Base rate
New consumer categories take 5-10 years from niche to tracked. Healthspan is at year 2-3.
Falsifiers
- GLP-1 safety signal emerges (major adverse event data)
- Consumer supplement spending declines in recession
Watch signals
- GLP-1 prescription volumes and market size (quarterly)
- Longevity startup funding rounds (New Market Pitch tracker)
Geopolitics
Principles-based AI regulation outperforms prescriptive
By 2028, countries with principles-based AI regulation (NZ, Singapore) outperform prescriptive-regulation countries (EU) on OECD innovation indices.
70%
Base rate
Principles-based regulation historically outperforms prescriptive in fast-moving sectors (70% base rate).
Falsifiers
- NZ reverses course with prescriptive AI legislation
- EU AI Act enforcement proves lighter than expected
Watch signals
- AI startup formation rates by country (OECD data, annual)
- Corporate AI investment location decisions
EducationANCHOR
Prediction practice separates top-decile performers
By 2027, individuals and organizations with structured prediction practices will demonstrably outperform median peers on strategic decision outcomes.
60%
Base rate
Tetlock's research showed superforecasters outperform by 60% on average. Practice effect is proven; organizational adoption is the question.
Falsifiers
- Prediction market accuracy degrades with scale (crowd gets dumber)
- Organizations that adopt prediction practices show no measurable advantage
Watch signals
- Enterprise adoption of prediction-tracking software
- Published case studies on calibration-driven decision outcomes
AI TrajectoryANCHOR
MCP/A2A becomes default agent API
By mid-2027, MCP or A2A protocol becomes the default API for agent-to-agent communication, replacing REST for AI-native integrations in 50%+ of new agent projects.
75%
Base rate
New developer APIs take 12-24 months from launch to majority adoption. MCP launched 2024, 97M monthly SDK downloads already (2025).
Falsifiers
- MCP download growth drops below 30% QoQ for 2 consecutive quarters
- Major AI provider abandons MCP for proprietary protocol
Watch signals
- MCP SDK monthly download trends
- Framework default-template adoption (Next.js, FastAPI, Cloudflare Workers)
AI TrajectoryANCHOR
Context graphs become the primary moat
By end 2027, competitive advantage in AI-native products shifts from model access to proprietary context graphs. Companies with deeper intent/receipt graphs outperform those with better models.
70%
Base rate
Platform moats historically shift from infrastructure (compute, models) to data/network effects within 2-3 cycles. We are at cycle 2.
Falsifiers
- Model capability leaps make context windows irrelevant (infinite context solved)
- Open-source context standards emerge and commoditize the graph layer
Watch signals
- Public case studies of context-graph-driven retention
- Verifiable-intent and receipt standards adoption (IntentTrace, similar)
AI Trajectory
Voice becomes primary UX for AI
By end 2026, voice interaction via real-time APIs (<500ms latency) becomes the default UX for at least one major consumer AI category, replacing forms and chat for specific workflows.
65%
Base rate
Voice assistant adoption took 5+ years with poor latency. Real-time APIs with sub-500ms response change the equation.
Falsifiers
- Voice UX fails to achieve >10% adoption in any AI category by end 2027
- Latency improvements plateau above 1-second threshold
Watch signals
- Consumer voice-first AI app launches and retention metrics
- Real-time API pricing trends
Crypto & DePIN
x402 + stablecoin rails become default for agent payments
By 2027, HTTP 402 paywalls and stablecoin settlement become the standard for agent-to-agent commercial transactions, replacing API keys and invoicing for micro-payments.
65%
Base rate
New payment protocols take 18-36 months to achieve ecosystem fit. x402 leverages existing HTTP semantics which accelerates adoption.
Falsifiers
- Stripe/PayPal add agent-native payment features that make x402 unnecessary
- No major API gateway supports HTTP 402 by end 2026
Watch signals
- API gateway support for HTTP 402 (Cloudflare, Kong, Apigee)
- Production agent commerce settlements on x402
The Compounding
Interaction effects
The highest-value predictions live at the intersections. Where two domains compound, the outcome accelerates beyond what either domain predicts alone.
AI × Crypto = Agent Commerce
30%First $1B in autonomous agent-to-agent transactions settled on crypto rails by Q4 2027.
MCP and A2A give agents communication. Stablecoin rails give them transaction. Production-scale deployment (Domain 1) plus mature stablecoin rails (Domain 2) creates a new category — agent commerce — faster than either domain predicts alone.
AI × Healthspan = Computational Biology
65%AI-designed longevity therapeutic enters Phase 2 trials by 2028.
AI capability gains (Domain 1) directly accelerate drug discovery (Domain 4). AlphaFold was the proof; GLP-1 optimization is the current wave. Base rate for AI-assisted drug discovery entering trials is already high.
Work × Education = Credential Disruption
55%At least one Fortune 500 company publicly drops degree requirements for 50%+ of roles by 2027.
As AI automates entry-level tasks (Domain 3), the value proposition of a 4-year degree weakens. Alternative credentials (Domain 6) fill the gap. Several Fortune 500 have already dropped requirements for specific roles.
Geopolitics × Crypto = Regulatory Arbitrage
35%NZ or Singapore becomes a top-3 jurisdiction for licensed stablecoin operators by 2028.
Countries with principles-based regulation (Domain 5) attract crypto-native businesses (Domain 2). NZ's light-touch approach plus GENIUS Act compliance could make it a hub for compliant stablecoin operations — but requires proactive positioning beyond current light-touch.
The Discipline
Calibration and cadence
A healthy forecast spreads its probabilities — too much mass at the extremes is a tell of overconfidence; too much at 50% is hedging. Cadence is how a forecast stays honest.
Probability distribution
1
0–30%
Low
2
31–50%
Toss-up
7
51–70%
Likely
3
71–100%
High
Review schedule
Monthly
Scan for signals that affect existing predictions. Flag any that need conviction updates.
Quarterly
Full probability review. Update any prediction where evidence shifted by 10%+.
Semi-annual
Calibration check. Compare predictions that resolved against assigned probabilities.
Annual
Full forecast refresh. Retire resolved predictions. Add new domains if warranted.
Dig Deeper
Learn how to reason about the future
The predictions above are the output. The discipline that produces them lives in the docs.
Prediction hub
The full discipline — character, practice, cadence
Superforecaster method
The decomposition process that produces calibrated predictions
Review process
Daily, weekly, monthly, quarterly — the cadence that keeps you honest
Prediction evaluation
Brier scoring — how we grade conviction over time
Prediction markets
What the crowd thinks and where you might know more
Perspective
The perception practice that makes prediction possible
Questions
Which prediction do you disagree with most — and what evidence would change your mind?
- If you could only track one domain for the next two years, which gives the highest decision-relevant signal — and what does that tell you about where your uncertainty actually lives?
- Which interaction effect is missing — and does its absence reveal a blind spot in your worldview?
- If three of these predictions are wrong by March 2028, which three teach you the most about how you think?
- Which domain expert do you not yet know — and what would change in your forecast if you spent an hour with them this quarter?