Example setup · Technical research

How four agents can research one company and keep the sources

Specialist agents split the research, flagged disagreements, and built one sourced memo.

Researchers exchanging findings through a connected October operation
Assess this product, company, and technical risk. Cite the important claims and flag disagreements.

Why this gets messy

Technical research pulls from docs, demos, repos, architecture, and current market sources. The final memo has to show where every important claim came from.

Several agents can research faster, but they can also repeat the same search, disagree quietly, and return summaries that do not fit together.

Put the whole job in one place

Each research area got its own agent and source list. A fourth agent tracked open questions, disagreements, and missing evidence. The human could open the source behind any claim.

Give each agent one clear job

Product research

Browser Use · Browser workspace. Read first-party docs, watched demos, and checked the public product.

Technical review

Codex · Repository workspace. Checked the architecture, dependencies, tests, and technical risk.

Market research

Gemini · Research workspace. Collected current market context and saved the source dates.

Final memo

Claude Code · Memo workspace. Resolved disagreements and wrote only from attached evidence.

Keep every handoff visible

Define the decision

Person running it. The operator wrote down the decision, the research questions, acceptable sources, and what uncertainty mattered.

Split the research

Research agents. Each agent owned one area and posted its sources, dates, findings, and open questions.

Resolve disagreements

Writing agent. Conflicting findings became visible tasks. The agents challenged or updated each other before writing the memo.

Review the memo

Person running it. The final memo linked important claims to sources and clearly marked guesses, uncertainty, and missing access.

You still make the important calls

Agents preferred first-party and current sources, with dates attached.

The writing agent could not hide disagreement.

The human set the criteria and made the final call.

What changed

Parallel agents gathered the research faster. Visible disagreements and attached sources made it trustworthy.

The operator reviewed one clear memo instead of merging several disconnected chat answers.

All case studies · Download October