Use cases
Multi-model AI for code review and technical decisions.
Some engineering questions have no single right answer. A multi-model debate puts the question to several models independently and records where they disagree, so you review the trade-off, not just a verdict.
Where it fits in engineering work
Decisions, not lookups
It suits questions a team would argue about in a design review: which datastore, which queue, which language, how to isolate tenants. It is not a replacement for tests, linters or a reviewer who knows your code.
The models cannot see your repository, so describe the situation in the question: the size of the system, the constraints and the options you are weighing.
Review questions, not diffs
For code review, ask about the approach rather than pasting a whole codebase: whether a migration plan is safe, whether a retry strategy can cause duplicates, whether an agent should be allowed to merge its own pull requests. Then take the dissent back to your reviewer.
Recorded examples
These are real debates, shown in full. They are examples, not advice for your system.
Tenant isolation
One database, one schema per customer, or one database per customer? Three models chose a shared schema with row-level security. Gemini chose a schema per tenant for regulated buyers.
Background jobs
A Postgres queue, Amazon SQS or Kafka? Three models chose a Postgres-backed queue. Gemini chose SQS. The final answer says when to switch.
TypeScript or Python
For an eight-week build, three models favoured a TypeScript API plus a Python service. The final answer overruled them and said so.
Agents merging their own pull requests
All four said no as proposed. One suggested a narrow exception for low-risk files.
A workable review loop
State the decision and the constraints
Team size, scale, deadlines, compliance needs and the options on the table.
Read the dissent
The objection from the one model that disagreed is often the risk worth testing.
Turn it into a check
Convert the strongest objection into a benchmark, a spike or a question for a human reviewer before you commit.
What this does not do. It does not run your code or replace testing, and it can be wrong. A consensus is a reason to look closer, not permission to skip review.
Read the debates
Tenant isolation debate
The full transcript, every model's position and the final answer.
Job queue debate
Postgres, SQS or Kafka, with the point where the answer changes.
TypeScript or Python debate
Where the final answer overruled the majority and said so.
AI agent merging debate
Whether an agent should merge its own pull requests.
Questions
Short answers.
Does it read my repository?
No. The models only see the question you send and any context you choose to include. Describe the situation in the question.
Can I use it from my editor?
Yes. It connects to MCP-compatible clients, so your assistant can start a debate and bring the result back into the chat. See the docs for setup.
Should I trust a unanimous answer?
Treat it as strong input, not proof. Models can share a mistake, so test the claims your decision depends on.
Bring your next design question.
Create an account and ask the question your team would argue about. Read the dissent before the verdict.
Ready to try it? Create an account
A note on limits. E3D Pilot gives you input, not an approval. Models can be wrong together, and agreement is not proof, so check what matters before you act.