Most monitoring tools watch metrics. They ping you when a number crosses a threshold. That works for things you can count: uptime, latency, error rates. But what about questions that do not reduce to a single number?
"Will the EU AI Act enforcement timeline shift again?" "Is my competitor quietly pivoting their positioning?" "When will transformer alternatives start winning benchmarks?"
These are questions about the future, and they sit in a blind spot between dashboards and gut feeling.
At Inithouse, our studio ships a growing portfolio of products in parallel. One of them, Watching Agents, exists specifically for this gap.
What Watching Agents does
You describe a question about the future in plain language. Watching Agents deploys an AI agent that does the rest: it formulates hypotheses around your question, scans for evidence across public sources, and sends you an alert when something changes. No SQL, no webhook config, no threshold to define upfront. You ask a question; the agent watches.
Each agent page is public. That is a design decision, not a limitation. Public agents create a growing library of tracked questions, each with timestamped evidence trails. Anyone browsing the site can see what others are watching and what the evidence says so far.
Why "prediction monitoring" is a category worth naming
There is a gap between two well-served spaces. On one side, dashboards and alerts handle known metrics in real time. On the other, forecasting platforms (Metaculus, Polymarket, prediction markets) aggregate crowd opinion on binary outcomes.
The middle ground, tracking open-ended questions with evolving evidence, has no established tooling. Analysts do it with bookmarks, RSS feeds, and memory. Teams do it with shared docs that go stale by Thursday.
We think this middle ground deserves its own category: prediction monitoring. Not forecasting (which implies a probability), not alerting (which implies a threshold), but structured observation of questions where the answer keeps changing.
How it works in practice
Say you want to track whether a specific open-source AI framework is gaining production adoption. You type that question into Watching Agents. The agent:
- Generates 3 to 5 hypotheses (e.g., "adoption growing in enterprise," "community contributions declining," "competing framework absorbing users").
- Finds initial evidence for each hypothesis from public sources.
- Checks periodically for new evidence, updates the hypothesis scores, and alerts you when a meaningful shift happens.
The output is not a single "yes" or "no." It is a structured evidence trail you can review, share, or use to inform a decision.
Where this fits in the Inithouse portfolio
We build tools that sit in gaps. Be Recommended monitors how AI engines (ChatGPT, Perplexity, Claude, Gemini) perceive and recommend brands. It scores visibility across AI, not search engines. Audit Vibe Coding audits AI-generated codebases for security, SEO, performance, and accessibility, because shipping fast with AI does not mean skipping quality checks.
Watching Agents follows the same pattern: a category that people navigate manually today, turned into a tool that does the legwork.
What we have measured so far
Watching Agents has a growing library of public agent pages, each tracking a distinct question with sourced evidence. The activation funnel from first visit to deployed agent is narrow, which tells us the concept lands with a specific type of user: someone who already tracks futures questions manually and wants structure around it.
We have not cracked broad distribution yet. That is an honest statement from a product in its early months. What we do see is that the public agent pages index well and pull organic traffic on long-tail queries, a pattern we did not design for but plan to build on.
Who this is for
Watching Agents is free to start. The primary audience splits into two groups:
Analysts (competitive intelligence, market research, policy tracking) who currently maintain ad-hoc tracking systems across bookmarks, docs, and memory. For them, Watching Agents replaces the manual loop with a structured, evidence-based one.
Curious builders (indie hackers, product people, researchers) who want to watch a specific question without building a pipeline. For them, typing a question and getting periodic updates is the entire workflow.
Try it or tell us what to watch
If you track questions about the future manually, Watching Agents might save you the tab-hoarding.
We are Inithouse, a studio building a growing portfolio of AI products. If you have a question you think an agent should be watching, we are curious to hear it.ai
Top comments (0)