Models get smarter. Costs fall. The natural trajectory: AI agents graduate from dumb intern → smart junior → manager → advisor → cofounder. As they climb, they do higher-leverage work. The highest-leverage work humans do is decision-making — and decisions are the most scalable output in existence. One decision can redirect billions of dollars or save thousands of lives.1
A decision’s quality is a function of four inputs:
As models commoditize, the first three inputs converge. Everyone has access to similar IQ. Taste remains personal. Problem context is inherently private. That leaves available information as the marginal variable — the one where adding more of it directly improves the decision.
The thesis: there will be massive demand for on-demand access to the vast knowledge, perspectives, and preferences of people out there — to feed better decisions. An “Uber for providing your life knowledge.” Not a search engine (public info). Not an expert network (expensive, scheduled). A marketplace where anyone can monetize their lived experience, industry knowledge, and judgment — on demand, at scale, for AI-mediated decision-making.
This connects directly to the Personal Agent thesis: as personal agents become identity infrastructure, they become the natural interface for this marketplace. Your agent queries other agents. Knowledge flows agent-to-agent, not human-to-human.
This isn’t speculative. We can already see agents climbing the ladder in real deployments. Each rung unlocks higher leverage — and higher information demands.2
| Rung | Agent Role | Human Analog | Leverage | Data Needs |
|---|---|---|---|---|
| 1 | Task executor | Intern / VA | 10× | Instructions only |
| 2 | Knowledge worker | Junior employee | 50× | Domain knowledge |
| 3 | Process manager | Manager | 200× | Org context + judgment |
| 4 | Strategic advisor | Senior / Advisor | 1,000× | Industry intel + relationships |
| 5 | Decision partner | Cofounder / Mentor | 10,000× | Everything above + private context + external perspectives |
The key insight: at rungs 1–2, public knowledge suffices. At rungs 4–5, the bottleneck is information the internet doesn’t have. What does a 20-year supply chain veteran in Shenzhen actually think about this factory? What does a parent in Tin Hau prioritize when choosing a playgroup? What does a PE fund partner look for in this specific sector?
68% of institutional decision-makers already use expert consultations to validate strategies.4 But they pay US$500–1,350/hour through GLG, AlphaSights, or Maven.9 The demand is proven. The question is: can you democratize access and expand supply to everyone, not just former McKinsey partners?
| Layer | Market | Size | Source |
|---|---|---|---|
| Global TAM | Data brokerage (all personal + business data trade) | US$303B (2024) | Grand View Research6 |
| Segment TAM | Decision intelligence (AI-powered decision support) | US$18.1B (2025) → US$74B by 2033 | SNS Insider5 |
| Adjacent TAM | Expert networks (on-demand human expertise) | US$3.8B (2025) → US$16.9B by 2035 | Global Growth Insights4 |
| Adjacent TAM | Prediction markets (crowdsourced decision signals) | US$44B volume (2025), 5× YoY | Wedbush7 |
| Analog | Perplexity (AI answer engine — public info only) | US$100M ARR, 45M users, $20B val | Sacra10 |
The proposed marketplace sits at the intersection of three existing markets: expert networks (human knowledge on demand), data brokerage (packaging and selling information), and decision intelligence (AI-powered decision support). Each is growing 10–16% CAGR.456
Today, only ~1 million people globally are registered on expert networks (GLG alone claims 1M+).11 These are overwhelmingly senior professionals — former executives, consultants, PhDs. The “Uber for life knowledge” thesis says: the supply should be everyone. A factory worker in Dongguan has knowledge no McKinsey partner has. A single parent in Mong Kok has decision-relevant context about a hyperlocal market. A 22-year-old TikTok creator has taste data that no demographic survey captures.
Prolific (human data marketplace for research) proved the supply side works: 380,000+ studies completed in 2025, 8M+ hours contributed, participants earning US$8–12/hour.12 But Prolific is for researchers, not for decision-makers. The gap: no platform lets a founder querying “should I open a playgroup in North Point?” instantly access 50 parents in the area who’ll share their actual preferences and constraints.
| Company | Model | Revenue / Scale | What They Got Right | Gap vs. Thesis |
|---|---|---|---|---|
| GLG | Expert network (phone consultations) | US$650M rev, 1M+ experts11 | Massive supply, institutional trust | US$500+/hr. Scheduled, not on-demand. Experts only — excludes 99.9% of knowledge holders |
| AlphaSense / Tegus | Subscription + per-call | US$4B valuation (post-Tegus acquisition)13 | AI-powered transcript library (240K+ interviews) | B2B-only. US$20K+/yr subscription. Zero consumer supply. |
| Maven | On-demand expert calls | US$150–1,350/call9 | “90% cheaper, 10x faster than consulting” | Still expensive. Still gated. Still scheduled calls, not async. |
| Perplexity | AI answer engine (public data) | US$100M ARR, $20B val10 | Instant answers from public web. 780M queries/mo | Public data only. Can’t access private knowledge, preferences, lived experience. |
| Polymarket / Kalshi | Prediction markets | US$44B volume (2025)7 | Aggregates distributed intelligence via incentives | Binary outcomes only. Can’t query nuanced perspectives. Financial risk required. |
| Delphi | AI clones of experts | $16M Series A (Sequoia), 2K+ experts14 | Digital minds scale 24/7. Paywall monetization for creators. | Clone quality limited to creator’s uploaded content. Not real-time knowledge. |
| Poe (Quora) | AI bot marketplace | Tens of millions in creator payouts15 | Price-per-message monetization for bot creators | Entertainment-first. No structured knowledge extraction. |
| Prolific | Human data marketplace | 380K studies/yr, US$8–12/hr12 | Quality-verified human participants at scale | Research-focused, not decision-focused. Slow (study design required). |
| Company | Model | What Happened | Lesson |
|---|---|---|---|
| Cameo | Celebrity video marketplace | $1B val (2021) → $86M (2024). Couldn’t pay $600K FTC fine.16 | Novelty wears off. Celebrity supply concentrated. No recurring use case. Lockdown business. |
| Clarity.fm | Expert call marketplace | Pivoted, mostly abandoned by original team | Two-sided marketplace for calls doesn’t retain either side. Experts leave for direct clients. |
| people.io | Personal data marketplace | #1 UK iTunes, Telefónica partner, NASDAQ Rising Star → dead.17 | B2B2C data marketplaces are too complex. Balancing user trust, developer needs, and corporate buy-in failed. |
| Rewind / Limitless | Personal knowledge capture | $10M a16z → pivot to pendant → acquired by Meta → shut down Dec 2025.18 | Personal context capture is valuable but privacy is a minefield. Meta acquired the tech, not the product. |
| 23andMe | Personal data marketplace (genetic) | Bankruptcy March 2025. 15M users’ data on auction block.19 | Personal data ≠ sustainable business. Privacy breaches (7M users), 50+ lawsuits. Data as asset is legally and ethically toxic. |
| Character.AI | AI character chat | $1B val, $32M rev, exploring sale amid costs.20 | Entertainment personas ≠ knowledge transfer. 75 min/day engagement but no decision value extracted. |
| Company | What They Do | Funding | Relevance |
|---|---|---|---|
| Delphi | AI clones of experts, paywall access | $16M Sequoia14 | Closest to “monetize your knowledge as a service” — but supply-side is influencers, not everyone |
| Experts.app | AI vaults of expert knowledge, subscription access | Pre-seed21 | Knowledge preservation + monetization. Early. |
| Tinrate | “OnlyFans for knowledge” | €1.6M seed22 | Direct monetization of expertise. Supply = domain experts, not everyone. |
| Pick My Brain (Raison) | Expert content → AI assistant, pay-per-query | €1.8M pre-seed23 | Upload content, create AI clone, monetize. Targets 100K+ follower experts. |
| Enquire AI | Digital expert network, 60K+ experts, AI matching | Growth-stage24 | AI-native expert network. 35-min average match. Still B2B. |
Key observation: every funded player in this space targets the supply side as experts/influencers (top 1% of knowledge holders), not the long tail of everyone. The “Uber for life knowledge from anyone” model — where the factory worker and the parent and the barista are all supply — is genuinely unvalidated.
| Metric | GLG / Tegus (Expert Network) | Delphi (AI Clone) | Proposed Marketplace |
|---|---|---|---|
| ARPU (demand side) | US$20–25K/yr subscription13 | ~US$10–50/mo per subscriber | US$20–50/mo (prosumer) or US$200–500/mo (business) |
| Supply-side payout | US$200–500/hr to expert9 | Creator sets price (paywall) | US$0.10–5.00 per query response |
| Take rate | 40–60%11 | ~20% commission23 | 20–30% (must be attractive to long-tail supply) |
| Gross margin | 60–75% | 70–80% | 50–65% (AI processing + payouts eat margin) |
| CAC | Enterprise sales ($5K+ per logo) | Creator marketing ($50–200) | Unknown — two-sided marketplace cold start |
| Cost Component | Per-Query Estimate | Assumption |
|---|---|---|
| AI inference (query processing + RAG) | US$0.01–0.05 | Claude Sonnet 4 / GPT-4o-mini, ~2K token query + 10K retrieval |
| AI inference (response synthesis) | US$0.02–0.10 | Multi-source aggregation, 5–20 knowledge sources per query |
| Supply-side payout | US$0.10–5.00 | Depends on expertise level + demand. Async text: $0.10–0.50. Live call: $5+ |
| Quality verification / trust scoring | US$0.005–0.02 | Automated credibility checks, cross-referencing responses |
| Platform infra (hosting, storage, matching) | US$0.002–0.01 | Per-query amortized cloud cost |
| Total COGS per query | US$0.14–5.19 | — |
The thesis is directionally right. But “Uber for life knowledge” is a vision, not a product. Here are the four product shapes this could take, ordered by viability:
Replace GLG’s model with AI. Don’t call 500 people to find 3 relevant experts — use AI to match, extract, and synthesize knowledge from a massive pool. Enquire AI (60K experts, 35-min match) is already doing this.24 AlphaSense acquired Tegus for $930M to build exactly this: an AI-powered knowledge layer over human expert interviews.13
Why it works: demand-side willingness-to-pay is proven (US$20K+/yr), AI dramatically reduces matching cost, transcript library creates compounding asset.
Why it’s not the thesis: this is B2B, gated, expensive. Still experts-only supply. Not “everyone.”
Let anyone create an AI version of themselves, trained on their knowledge, that others can consult for a fee. Delphi ($16M Sequoia) and Pick My Brain (€1.8M) are the early movers.1423
Why it works: scales supply infinitely (clone answers 24/7), creator has strong monetization incentive, AI handles the interaction.
Why it’s fragile: knowledge captured at clone-creation time, not updated in real time. Quality degrades as the real person’s knowledge evolves. Cameo pattern risk — novelty wears off when the clone doesn’t match the real person’s current thinking.
This is what the thesis actually describes. Your personal agent — which already has your full context, preferences, relationships, and knowledge — responds to queries from other agents. Agent-to-agent knowledge exchange. You don’t create a clone; your live agent IS the interface.
Why it’s powerful: always up-to-date (it IS your agent), no clone-creation friction, integrates naturally with the personal agent adoption curve. If personal agents become identity infrastructure (~2027–2029 Hotmail moment), this marketplace is a protocol on top.
Why it’s premature: 10–50K people run personal agents today. Need millions for a marketplace to have liquid supply. At least 2–4 years from critical mass.
Don’t build a marketplace. Build the aggregation layer that sits between AI agents and human knowledge. Like how Perplexity aggregates the public web, this aggregates the private web: surveys, interviews, preferences, reviews, behavioral data — synthesized for decision-making.
Why it could work: Perplexity proved the model ($100M ARR, $20B valuation) for public data.10 If you could do the same for private/experiential data, the value per query is 10–100x higher.
Why it’s hard: data acquisition is the entire challenge. Perplexity crawls the web for free. Private knowledge requires consent, payment, and trust — which is exactly the problem every failed personal data marketplace hit.
The core claim: as models commoditize, data becomes the edge. Let’s test this rigorously.
The /zeitgeist session from Feb 14 surfaced signals directly supporting this thesis. The zeitgeist is not just compatible with this idea — it’s upstream of it.
If execution is free and everyone has the same AI IQ, the only differentiation is what you decide to build and why. Decision quality becomes the scarcest resource. A marketplace that improves decision quality has infinite demand.2728
“Vibe marketing > vibe coding” — the bottleneck isn’t building, it’s knowing what to build and getting anyone to care.29 This is a decision problem (what to build) and a taste problem (how to make people care). Both are information-starved.
Bad AI output → infects human writing → degrades training data → worse AI output. If taste degrades, the value of human taste data goes up, not down.30 A marketplace for genuine human perspective becomes a counter-force to the AI slop loop.
The Shenzhen pattern: intelligence distributed across relationships, 30x learning rate per dollar.8 This is exactly what the marketplace would do digitally: let decision-makers access distributed intelligence across thousands of perspectives simultaneously, instead of deep-diving one expert at a time.
| Asset | Relevance |
|---|---|
| Donna (live personal agent) | Living proof-of-concept for the “personal agent as knowledge interface” shape (Shape C). Already manages CRM, research, comms. |
| PCRM system | Rich private context layer — relationships, research, decisions. Could be the first node in a knowledge network. |
| OpenClaw expertise | Infrastructure knowledge for agent-to-agent communication. MCP protocol understanding. |
| claw.degree | Agent evaluation layer. Could evolve into agent knowledge quality scoring. |
| Research engine (this) | Already a decision-making tool powered by synthesized knowledge from multiple sources. Eric IS the first user. |
| @ericsanio audience | Growing. Taste/agency thesis resonates. Natural distribution for the idea. |
Decisions are infinitely scalable. One good decision can be worth billions. As AI agents become the interface for decision-making, they will demand ever-more-granular information inputs. The current expert network market ($3.8B, 16% CAGR) is just the institutional tip of the iceberg. When every individual has an AI decision partner, the demand for on-demand human knowledge explodes. The “Uber for life knowledge” marketplace is a trillion-dollar protocol sitting underneath the personal agent infrastructure layer.
Two conditions must both be true:
If both happen, the marketplace is a protocol, not a product. And protocols are built by the infrastructure layer (OpenClaw, Anthropic, OpenAI) — not by a marketplace startup.
The thesis is correct. The timing is wrong. The product shape is wrong.
Decisions are the highest-leverage human output. As AI agents climb to decision-partner level, they will demand richer information inputs than the public web provides. The “experiential data” layer — what people actually think, prefer, and know from lived experience — is genuinely uncaptured, valuable, and growing in importance as static data commoditizes.
But every attempt to build a marketplace where individuals sell their knowledge/data directly has failed. The economics don’t work: individual knowledge is too cheap to motivate supply, too noisy to trust, and too privacy-sensitive to scale. The companies that succeeded — AlphaSense ($4B), Perplexity ($20B), Polymarket ($44B volume) — didn’t build marketplaces for raw human input. They built aggregation products that extract value from human knowledge without requiring individuals to actively participate in a marketplace.
What to build instead (for Eric, right now):
1. Don’t build a marketplace. Build a protocol feature for personal agents. The right shape is an MCP-like standard for agent-to-agent knowledge queries. “My agent asks your agent: what do you know about X?” This becomes a feature of personal agent infrastructure, not a standalone business.
2. Dog-food it with Donna + the existing OpenClaw network. Let Donna query other agents in Eric’s circle for knowledge inputs during /deepmarketresearch runs. The research engine is already the product — adding human knowledge sources makes it better.
3. Write the thesis piece for @ericsanio. This maps directly to the zeitgeist threads (means-collapse, taste-is-everything, speed-doesn’t-fix-the-bottleneck). The framework — “decision quality = f(IQ, taste, context, data) and three of four inputs are commoditizing” — is a publishable insight.
4. Watch the 2–3 year timeline. When personal agents hit ~1M users (projected 2027–2029 per the Personal Agent thesis), revisit this as a protocol play. Until then, the pieces aren’t in place.
One thing that changes the answer: If someone figures out the supply-side incentive mechanism — a way to make participating frictionless AND valuable for regular people (not just experts) — this becomes a generational company. Nobody has cracked that yet. The closest is Polymarket’s financial incentive mechanism, but that only works for binary predictions, not nuanced knowledge.