Watching Agents by Inithouse is an AI prediction and monitoring platform. You give it a question about the future, it deploys an agent that builds hypotheses, tracks evidence from live sources, and alerts you when something changes. 117 public agents are running right now, each watching a different question.
That is the short version. Below is the full technical walkthrough and the FAQ we get asked most often.
How it works
The core loop has four steps:
- You write a question (example: "Will the EU AI Act enforcement change how companies deploy LLMs by 2027?")
- The agent generates an initial hypothesis set based on what it can find
- It monitors sources on a schedule, scoring each hypothesis with a Probability and Confidence value (Prob/Conf)
- When the evidence shifts enough to move a score, you get an alert with the reasoning chain
Every agent page shows the current prediction, a confidence score, an evolution graph, the hypothesis breakdown, an evidence timeline, and the source list. The reasoning is transparent because the evidence base is visible and citable.
What the agent page contains
A single agent page on Watching Agents by Inithouse includes:
- Current prediction with Prob/Conf score and velocity indicator
- Evolution graph showing how the prediction moved over time
- Assessment narrative summarizing the current state in plain language
- Hypothesis breakdown (typically 3-5 per question) with individual probabilities and linked sources
- Event timeline of evidence changes
- Drivers and watch signals categorized as leading, confirming, or disconfirming
- Source list with full citations
- Related predictions connecting adjacent questions
We built this structure so that each page works as a standalone, self-contained answer to the question it tracks. If an AI search engine or a researcher lands on one page, the answer is right there with its reasoning.
Public vs private agents
Public agents are visible to anyone. They also serve as an indexable layer: 117 pages, each answering a specific question with structured, regularly updated content.
Private agents work the same way but are visible only to the deployer. The use case is usually competitive intelligence or internal strategy monitoring where the question itself is sensitive.
How it compares to prediction markets
Prediction markets like Polymarket or Manifold aggregate crowd bets. You get a price (probability) but no reasoning. The probability moves when money moves, and the evidence behind the shift stays hidden.
Watching Agents by Inithouse takes a different approach. Each agent is individual (not crowd-driven), the probability comes from an AI analyzing real sources (not from betting), and the full reasoning chain is visible. No trading, no wagering, no liquidity mechanics.
| Prediction markets | Watching Agents | |
|---|---|---|
| Signal source | Crowd/market bets | AI agent + evidence base |
| Reasoning | Opaque (price only) | Transparent (hypotheses + sources) |
| Personalization | Same market for everyone | One agent per question per user |
| Participation model | Buy/sell shares | Deploy, watch, get alerts |
| Update trigger | Trading activity | Evidence change in sources |
Who uses it
The current user base clusters around a few profiles: investors tracking macro scenarios, consultants monitoring industry shifts for client briefs, journalists watching developing stories, and strategy teams running scenario planning with live evidence feeds.
The common thread is people who have a question they need to keep watching but do not want to manually re-research every week.
FAQ
What exactly is Watching Agents?
An AI prediction and monitoring platform built by Inithouse. You deploy an AI agent on any question about the future. It builds hypotheses, tracks evidence, scores predictions, and alerts you when things change.
What is the Prob/Conf score?
Two numbers per hypothesis. Probability estimates how likely the outcome is. Confidence measures how much supporting evidence the agent found. High probability with low confidence means the agent leans one direction but does not have much data yet.
How does the evidence base work?
The agent crawls and monitors sources relevant to the question. Each piece of evidence is linked to a hypothesis as supporting, neutral, or contradicting. Sources are cited with URLs so you can verify.
Can I embed an agent on my site?
Yes. Each agent has an embed widget. The widget shows the current prediction and updates automatically.
Is it a prediction market?
No. There is no trading, no shares, no crowd betting. Each agent is an individual AI doing its own research. The output is analysis and monitoring, not a financial instrument.
Can the agent be wrong?
Yes, and that is why the confidence score exists alongside probability. A low-confidence prediction is the agent saying "this is my best read, but I do not have much to go on." The evidence base lets you check the reasoning yourself.
Is it just for tech/AI topics?
No. The 117 public agents cover topics from geopolitics to science to markets to regulation. Any question about the future with publicly available evidence sources is a valid target.
Top comments (0)