Agent competitive analysis team

Competitive analysis team of agents

A competitive scan by one agent is five marketing pages summarized and called research. Split it so the output is comparable and sourced: a lead fixes one frame for everyone, an analyst tears down each competitor on that frame, a checker enforces where claims come from, and a synthesizer turns the grid into a decision. Without the checker, the report repeats each vendor's own copy back as fact.

Team

• Lead (product-manager-role): scopes competitors and the comparison dimensions.
• Analysts (ux-researcher-role): one competitor each, on a shared template.
• Evidence checker: enforces sourcing and kills marketing-only claims.
• Synthesizer (research-synthesis): builds the matrix and the implications.

Shape: parallel fan-out of teardowns, an evidence gate, then synthesis.

Method

1. Lead fixes the comparison frame first. Pick three to five competitors and the same dimensions for each (pricing, onboarding, core workflow, integrations, gaps), so teardowns are comparable, not five essays. Output scope.md with a shared template.
2. Assign one analyst per competitor, same template. Parallel fan-out. Each fills teardown-<competitor>.md dimension by dimension so the synthesizer can lay them side by side.
3. Enforce evidence rules: source every claim, rank the source. Each line cites where it came from (docs page, pricing page, hands-on trial, changelog, third-party review) with a date. Primary and hands-on beat the vendor's marketing, which is a claim, not a fact.
4. Separate observed from inferred. "Free tier caps at 3 seats," observed on the pricing page, versus "they likely target small teams," an inference. The checker labels each; an inference dressed as fact is the failure mode.
5. Trial the product where you can, do not read about it. For workflow claims, the analyst signs up and completes the core task, capturing what it took. A teardown built only from the marketing site describes the marketing site.
6. Checker gates claims before synthesis. It downgrades unsourced or marketing-only claims and flags anything stale, for example pricing older than 90 days. Only surviving claims reach the synthesizer.
7. Synthesize into a decision, not a table dump. landscape.md gives the matrix, then two or three implications for your product: where you win, where you are exposed, the gap to close. A matrix with no "so what" is trivia.

Run it

In Claude Code, run the lead as one subagent, spawn analysts as parallel subagents each with web access and one competitor over a shared directory, then run the evidence checker as its own subagent over the teardowns before the synthesizer reads only surviving claims. Port it to CrewAI as a scoping task feeding parallel teardown tasks then a synthesis task, to AutoGen as a GroupChat with a dedicated fact-checker, or to LangGraph as a fan-out then fan-in graph with an evidence node between teardown and synthesis.

Signals it works

• Every teardown covers the same dimensions, so the matrix lines up.
• Each load-bearing claim cites a dated source, and marketing copy is labeled.
• The synthesis ends in implications for your roadmap, not a feature grid.

Boundaries

This organizes competitive research; it does not access non-public data (roadmaps, contracts, internal pricing) and must not obtain it through deception or by scraping behind a login you lack rights to. Agents misread stale pages and vendor spin, so treat the output as a sourced draft for a human to sanity-check. Positioning calls stay with product leadership; vendor-evaluation covers the buy-side version.