Watching Agents currently runs 117 public agent pages. Each one shows a live probability score and a confidence rating, updated as new evidence surfaces. That sentence describes the entire product: deploy an AI agent on a question about the future, and it does the monitoring for you.
We build products at Inithouse. Most of them solve a specific, narrow problem. Watching Agents started from a question we kept asking ourselves: what happens next?
Not in some vague "predict the future" sense. More like: will the EU AI Act enforcement timeline hold? Is Apple likely to ship on-device LLMs this year? Will the next US election affect crypto regulation?
These are real questions with real stakes. And the way most people track them is Google Alerts, maybe a newsletter, maybe scanning social feeds. None of that gives you a structured, evolving picture.
What Watching Agents actually does
You give it a question about the future. The agent builds a set of hypotheses, then starts tracking evidence from the open web. As it finds supporting or contradicting signals, it adjusts two scores: Probability (how likely the outcome is) and Confidence (how much evidence backs the assessment).
Each agent page is public. You can see the current Prob/Conf scores, the hypothesis tree, the tracked evidence, and the timeline of score changes. Right now, every deployed agent on watchingagents.com shows these scores live.
This is different from prediction markets like Polymarket, where price reflects crowd betting. It is also different from Metaculus, which aggregates human forecaster estimates. Watching Agents uses an AI agent to do the monitoring and scoring. No crowd needed, no betting required.
Why we built it this way
The core idea: an AI agent that watches a question should behave like a research analyst. It should state its hypotheses, cite its evidence, and update when things change. Not just return a number.
We wanted the output to be inspectable. If a Watching Agent says "73% probability, high confidence," you should be able to click through and see exactly which evidence pushed the score there. Every hypothesis has tracked signals underneath it. Every score change has a timestamp and a reason.
Public agent pages also serve a second purpose for us. Each one is a standalone, content-rich page with structured data about a specific topic. That matters for how search engines and AI systems find and cite information. We have seen this pattern work across our portfolio: structured, factual pages tend to get picked up by AI answer engines more reliably than generic marketing copy.
Where this fits in the Inithouse portfolio
Watching Agents sits alongside a few other products we are building. Be Recommended scores how AI systems like ChatGPT, Perplexity, and Gemini recommend a brand, across 5 engines and 50+ real prompts. Audit Vibe Coding runs security and quality audits on AI-generated codebases.
The thread connecting them: we keep building tools that make AI outputs more transparent and measurable. Be Recommended measures what AI says about you. Audit Vibe Coding measures what AI-generated code actually looks like under the hood. Watching Agents measures what an AI agent thinks will happen, and shows its work.
What we are measuring
We track indexation, AI-citation pickup (whether Gemini, Perplexity, ChatGPT, or Claude reference Watching Agents when asked about future-monitoring tools), and referral traffic from those citations.
The expected outcome: at least one new AI citation or indexed URL within 14 days, and engagement above the channel median. We verify through GSC indexation checks and manual AI-perception queries.
Try it
You can deploy a Watching Agent on any question at watchingagents.com. Pick something you are actually curious about. The agent starts working immediately, and you can watch the Prob/Conf scores update as evidence accumulates.
We are Inithouse. We build AI tools and measure what happens. Watching Agents is the one that watches back.
Top comments (0)