Guide

How AI research agents work

A research agent run is a bounded loop: each agent is given a subject, a set of sources it may read and a time budget, it fetches and normalises what it finds, and it hands the result to the next stage. Most of the pipeline is deterministic code rather than model calls, which is what makes a run reproducible, affordable and debuggable.

How the work is divided

Specialisation by source and by question. One agent reads job boards and knows how six of them shape their responses. Another compares dated captures of pricing pages. Another reads filings. A scoring agent decides what survives. A corroboration agent counts independent sources. A writing agent produces the prose at the end.

The alternative - one agent with access to everything - is harder to test and impossible to explain. When output is wrong, the question is always which stage produced it.

Everything is bounded

Each agent runs with a timeout, a concurrency limit and an error boundary. A slow source is abandoned rather than allowed to consume the run. A failing source is recorded as failed and the run continues without it.

This sounds like plumbing, and it is what separates a system that produces something every night from one that produces nothing on the night one API changes shape.

Why the model comes last

Collection, deduplication and scoring are deterministic problems with well-understood algorithms. Using a model for them is slower, more expensive and less reproducible, and it makes the selection unexplainable.

The model earns its place at the end, where the task is genuinely linguistic: turning a small set of selected documents into a paragraph a person can act on, citing the documents it was handed.

What a run should report about itself

A trustworthy run produces a second output alongside the findings: which sources answered, which were skipped for a legitimate reason such as the company not being listed on that board, which failed, how long each took, and how many documents each contributed.

That report is how you tell a quiet week from a broken pipeline, and the two look identical without it.

Where QuikSignal fits

QuikSignal runs 11 agents against thirty sources on this pattern, with per-source outcomes recorded for every run and visible in the workspace.

A company that is not listed on a given job board is reported as not listed rather than as a failure, because a company is on one board and the others correctly have nothing to say.

See it on your own market →

What it does not do
  • Agents read a defined source set rather than browsing freely, which is a constraint chosen for reproducibility.
  • No agent acts outside the product: nothing is sent, posted or purchased.
  • Runs are bounded by time. A long run returns what it has rather than blocking, and says how far it got.
Questions

What people ask

Is this the same as an autonomous agent?
No, and deliberately. These agents have a fixed source list and a fixed job. Nothing decides on its own to go and read something else.
How much does a run cost to operate?
Dominated by the model stage, which is why keeping the model at the end matters. If the deterministic filtering removes ninety-nine per cent of the material, the model cost is one per cent of what it would otherwise be.
What happens when a source changes its format?
That agent starts failing, the run reports it, and the rest continues. Without per-source reporting the same event produces a slightly thinner brief and nobody investigates.

Read your own market the same way.

Eleven agents, the companies you choose, every night, with the document behind every line.