Stage 2 — Employer research

Stage 2 takes the postings you selected in Stage 1 and produces an “employer profile” for each. The profile is what the cv-adapter reads in Stage 3 to tailor your CV — without it, generation falls back to generic bullets.

What the agent produces

Per company:

  • Overview — what they do, industry, stage/size, founded year, HQ, key markets.
  • Mission — single-sentence company mission.
  • Values — the values they advertise (filtered for the substantive ones, not “we work hard”).
  • Recent activity — public news from the last ~6 months.
  • Role needs — key responsibilities, must-have skills, nice-to-have skills, soft skills.
  • Open positions — other open roles at the same company (a hint for “should I apply to two?”).
  • Fit points — three direct/strong/gap notes comparing your CV to the role needs.
  • Cover-letter angle — a one-sentence narrative the cv-adapter uses as the cover-letter’s spine.

The sub-agent fan-out

employer-profiler is the orchestrator. It launches one company-researcher sub-agent per unique company in your selection — they run in parallel, each one independently fetching the company’s website, looking up recent news, and synthesizing the profile. The orchestrator then writes everything to companyResearch + userCompanies/userJobFit.

Parallel = fast for small batches but watch the cost. Five companies = five Sonnet runs. The active-runs indicator shows the parent run; cancelling cancels every sub-agent.

Re-running one company

If a profile looks wrong (the agent picked up the wrong company name, or the research page was stale), Stage 2 has a per-row Refresh action. It re-runs only that one company-researcher; the other rows stay as they are.

Selection cost

Selecting a profile in Stage 2 marks it as “feed this into Stage 3 generation”. Stage 3 only generates for selected rows. You can de-select before Stage 3 if you change your mind — no documents have been generated yet at this point.