Template · Agent · Official
FirecrawlNotionGoogle DriveSlackObsidian

Research briefings, sourced and cited

Point an agent at a question; get back a sourced briefing — market landscape, technical comparison, diligence pass — with every claim linked.

Get this template

Opens in AGNT and takes about a minute.

Research briefings, in brief

Tools it needsFirecrawl, Notion, Google Drive, Slack, Obsidian
Setup4 steps, about ten minutes
ApprovalYours, per category — nothing is sent on your behalf unless you say so
Where it runsYour own machine. Credentials stay in a local vault.

What the agent actually does

An answer with sources, not a list of links

The agent searches, reads the promising results in full rather than skimming snippets, and reconciles what it finds. Where sources disagree it says so, which is usually the most informative sentence in the briefing.

Claims carry citations

Every substantive statement links to where it came from. You can check the ones that matter for your decision instead of accepting the whole document on trust or none of it.

Your format

Landscape scan, competitive comparison, technical evaluation, diligence memo — you specify the structure once as a template and every briefing arrives in it, which makes them comparable to each other.

The tools this job needs

Firecrawl

Where the work arrives. Clean page extraction for research-grade scraping. The agent watches it and reads what turns up in full, rather than matching a rule against a subject line.

Notion

Context the agent pulls in before deciding. The team wiki agents can actually keep current. It is read, not just referenced — which is what lets the decision account for it.

Google Drive

Context the agent pulls in before deciding. Where the documents live — and pile up. It is read, not just referenced — which is what lets the decision account for it.

Slack

Where the result lands. Team chat where operations actually surface. Nothing is written here until the agent has formed a view and, where you asked for it, you have approved it.

Set it up

  1. Write the question precisely, including what a good answer would let you decide.
  2. Give the agent its sources — the open web, a document set, an internal database, or all three.
  3. Specify the output template so briefings stay comparable over time.
  4. Read the briefing against its citations, and refine the question where the answer missed the point.

You can see exactly what it did

Nothing happens behind your back.

Every run leaves a receipt: what the agent read, which tools it called, what it decided and why, and precisely what it changed. Anything irreversible — sending, paying, publishing, deleting — waits for you to approve it. It runs on your own machine, with your own credentials, and the whole trail is yours to read afterwards. The point is not that you trust it. The point is that you never have to.

The brief it works from

This is the actual instruction set the template installs — what the agent is told to do, and what it is told never to do. Every line of it is yours to edit in AGNT after install.

Read the full brief

You produce sourced research briefings. A list of links is not research — reading them, reconciling them and forming a view is.

Method

Search broadly, then read the promising sources in full rather than working from snippets. Notice when several results are reprints of one press release; three copies of a claim is one source, not three, and treating it as corroboration is the most common way research goes wrong.

Where sources disagree, say so explicitly and say which you find more credible and why. That sentence is usually the most informative one in the briefing, and burying it to appear authoritative wastes the reader's time.

Every substantive claim carries a link

The reader must be able to check the two or three claims their decision actually turns on. An uncited briefing has to be trusted entirely or discarded entirely, and both are bad options. Never cite a source you did not actually retrieve, and never present a plausible recollection as a finding.

Distinguish what you found from what you concluded. Mark estimates as estimates. State the date of anything time-sensitive, because a two-year-old market figure presented without a date is functionally a false statement.

Structure

Follow the user's template so briefings are comparable to each other over time. Lead with the answer to the question asked, then the evidence, then what you could not establish.

Say what you could not find

A gap named is useful; a gap papered over is a trap. If the question cannot be answered from available sources, say that plainly and explain what would be needed to answer it.

You are strong at coverage, extraction and reconciliation, and weak at knowing which credible source reflects how a market really behaves. Leave that judgment to the reader, and give them what they need to make it.

Shown exactly as it ships. The agent inherits whichever model you already use.

Search results are not research

A list of ten links is the beginning of the work. Research is reading them, noticing that three are reprints of one press release, finding that the fourth contradicts the others, and forming a view about which to believe. That synthesis step is what takes an afternoon, and it is the part an agent can genuinely take on — provided it shows the work rather than asserting a conclusion.

Citations are what make it checkable

An uncited briefing has to be trusted entirely or discarded entirely, and both are bad options. When claims carry sources, you can verify the two or three that your decision actually turns on and move on. That is how human analysts have always been reviewed, and it is the only workable standard for machine-produced research too.

Where it stops and you start

An agent is strong at coverage, extraction and reconciliation, and weak at knowing which of two credible sources reflects how the market really behaves. Treat the briefing as a well-prepared analyst’s first draft: the facts assembled and cross-checked, the judgment still yours.

Common questions

How do I know it did not invent a source?

Citations are links to material actually retrieved during the run, and the receipt records what was fetched. A claim without a source is visible as unsupported.

Can it research private or internal material?

Yes — point it at your own documents or databases. With a local model, none of that content leaves your machine.

How long does a briefing take?

Minutes rather than the afternoon it displaces, depending on how many sources you ask it to read in full.

Give AI a job. Get the proof.