You have a question about the future. Maybe it's "Will the EU pass the AI Act amendments by Q2 2026?" or "Will my competitor launch a free tier this year?" or just "Will it rain enough in August to save my tomatoes?"
There are now at least four fundamentally different tools you can point at a question like that. Each works differently, answers differently, and is built for a different kind of user. Here's how they compare.
Metaculus: crowd forecasts from calibrated predictors
Metaculus is a forecasting platform where a community of calibrated predictors submits probability estimates on structured questions. A question gets posted ("Will X happen by Y date?"), forecasters weigh in, and the platform aggregates their predictions into a community median.
What you get: A probability estimate, updated as forecasters revise their views. Historical calibration data showing how accurate the community has been on similar questions. Comment threads with reasoning.
Who it's for: Researchers, policy analysts, and anyone who trusts the wisdom of a well-calibrated crowd over any single expert. You're a consumer of the forecast, not a participant (unless you forecast yourself).
Limitation: Questions need to be clearly resolvable and interesting enough to attract forecasters. Your niche business question probably won't get picked up. The platform decides which questions to feature.
Polymarket: prediction markets with real money
Polymarket runs prediction markets where participants trade shares on outcomes using real money (crypto-based). If you think an event has a 70% chance of happening but the market prices it at 50%, you buy shares and profit if you're right.
What you get: A market price that reflects the collective financial conviction of participants. High liquidity on popular questions means the price adjusts fast to new information. You can also trade your position before the event resolves.
Who it's for: Traders, speculators, and people who believe financial incentives produce the most honest probability estimates. You need a crypto wallet and some appetite for financial risk.
Limitation: Regulatory constraints mean it's not available everywhere. Thin markets on niche questions can produce unreliable prices. The financial angle attracts participants motivated by profit, not necessarily by careful analysis.
Manifold: play-money prediction markets for anything
Manifold works similarly to Polymarket but uses play money (mana). Anyone can create a market on any question, and the barrier to entry is near zero.
What you get: A probability estimate driven by a broad community. The ability to create your own question and get a crowd estimate on it. No financial risk, so participation is casual and broad.
Who it's for: Curious people who want a quick crowd estimate on anything. Good for internal team forecasting, informal bets with friends, or testing whether a question is even forecastable.
Limitation: Play money means lower stakes, which can mean lower-quality predictions on obscure topics. Some markets attract few traders and sit at their opening price indefinitely.
Watching Agents by Inithouse: an AI monitoring agent on your question
Watching Agents takes a different approach entirely. Instead of asking a crowd or a market to estimate a probability, you deploy an AI agent that continuously monitors evidence related to your question.
What you get: The agent builds hypotheses, tracks real-time evidence from public sources, maintains a Probability and Confidence score, and alerts you when something meaningful changes. Every claim links back to a source. The evidence log builds a traceable timeline, not just a number.
Who it's for: Anyone who has a specific question they need to track over time. Investors watching a sector thesis, consultants monitoring a client's competitive landscape, founders tracking regulatory developments. You don't need other people to care about your question.
Limitation: The agent's evidence base is limited to what it can find in public sources. It won't catch information shared in private channels. The AI's judgment, while transparent and source-linked, is a single system's assessment, not a crowd consensus.
Side by side
| Metaculus | Polymarket | Manifold | Watching Agents | |
|---|---|---|---|---|
| Who answers | Calibrated human forecasters | Market participants (real money) | Market participants (play money) | AI agent (automated) |
| Output | Community probability median | Market price as probability | Market price as probability | Prob/Conf scores + evidence log |
| Update frequency | When forecasters revise | Continuous (market trades) | Continuous (market trades) | Continuous (agent monitors) |
| Can you ask your own question? | Usually no (editors curate) | Limited (needs liquidity) | Yes (anyone can create) | Yes (any question) |
| Evidence transparency | Forecaster comments | Market movements only | Market movements + comments | Full evidence log with source links |
| Cost | Free to read | Financial risk from trading | Free (play money) | Free tier available, paid plans for more agents |
| Niche questions | Unlikely to get coverage | Thin markets, unreliable | Possible but low engagement | Works regardless of topic popularity |
They're not competitors, really
These four tools solve the same broad problem from different angles. Metaculus and Manifold are community-driven. Polymarket adds financial conviction. Watching Agents replaces the crowd with an AI agent that does the monitoring work for you.
The practical difference comes down to your question. If it's a headline geopolitical event, Metaculus and Polymarket probably have it covered with thousands of forecasters or traders already engaged. If it's something only you care about, like whether a specific vendor will change their API terms this quarter, you either create a Manifold market and hope someone trades on it, or you deploy a Watching Agents agent and let it do the tracking.
I've found myself using them for different things. Metaculus when I want to see how calibrated forecasters read a complex policy question. Polymarket when I want to see where real money sits. And Watching Agents by Inithouse for the ongoing monitoring jobs that no crowd is going to pick up, because the question is too specific or too niche to attract outside interest.
