- questionWhat have other cities tried to add winter shelter capacity?
- source typeMunicipal reports and council minutes
- source typePeer-reviewed housing research
- source typeLocal news coverage
- constraintTwo pages. Plain language. Sources listed.
AI research briefs with every claim traced to a source
A director needs a sourced answer before a meeting, and someone loses two days to tabs and half-remembered links.
A research agent searches the source types you trust, records why it dropped the ones it didn't keep, and drafts a short brief that says plainly what is still unknown. Every claim links to the passage it came from, so checking it is quick. A person reads it before it goes anywhere.
Who it suits: Policy, program and advocacy teams that need short, well-sourced briefs on a deadline.
This is an example workflow, an illustration of what gets built. Its numbers are examples, not client results.
A two-page research brief
An agent searches for sources, keeps the credible ones, pulls out the facts that answer the question, and drafts a short brief with every source listed.
In this example: A two-page brief from 12 vetted sources, every claim traced back. A policy lead reads it before it goes anywhere.
Example workflow. Given a policy question and a two-page limit, the agent searches municipal reports, housing research and news, and finds 34 sources. It keeps 12 and records why the other 22 were dropped. It extracts 48 facts and drafts a two-page brief with a summary, a section on approaches and costs, and a section on what is still unknown. The draft waits for the policy lead, who swaps one source. An inspection step confirms every claim has a citation, every source link resolves, and every quote matches its source. The brief is then delivered with its source list.
trace
0/11
- drawing
- wf-04 · research brief
- status
- example workflow
- stations
- 5, one sign-off
- run
- waiting
- The problem
- A director needs a short, sourced answer before a meeting. Someone loses two days to tabs, notes and half-remembered links.
- What the agent does
- Searches the source types you trust, screens out weak or off-topic material with a reason for each, pulls out the facts that answer the question, and drafts a short brief that says plainly what is still unknown.
- Where people stay in the loop
- A person reads the draft before it goes anywhere and can swap sources or cut claims. Every claim links back to the passage it came from, so checking is quick.
Common questions
- How do you stop it inventing sources?
- Every claim must point to a passage in a retrieved source, and an automatic check confirms each link resolves and each quote matches before the draft reaches a person.
Something like this in your week?
Book a discovery call- AI intake triage for housing providers and nonprofitsAn example AI agent workflow that reads a shared inbox and web forms, sorts each request, flags urgent cases for a person, and routes the rest to the right team.
- AI for nonprofit grant and impact reportingAn example AI agent workflow that pulls program spreadsheets and case notes, maps outcomes to a funder's framework, and drafts a report with every figure linked to its source. Staff review it before export.
- AI invoice reconciliation for small finance teamsAn example AI agent workflow that matches invoices against purchase orders and ledger lines, reconciles the clean matches, and sends every mismatch to a review queue.
- AI meeting follow-up: action items, emails and calendar holdsAn example AI agent workflow that turns meeting notes into action items, draft follow-up emails and calendar holds. Nothing is sent or booked until someone approves it.