Why not just use Claude Code?
You should use it. Symphony runs inside it. The real question is what surrounds it when the product does not exist yet.
- Prompt
- ?
- ?
- ?
- Output
The product decisions still get made. You do not see them.
- Brief
- Define
- Scope
- Design
- Check
- Release
You decide at each gate. Every answer and every check stays on the record.
One prompt can rebuild a product that already exists.
A finished product supplies its own answers: what it is for, how it behaves, what it feels like. A new idea has notes, research and ambition. The decisions still have to be made, and a coding agent cannot make them for you.
Serras is for the product that does not exist yet. It asks the right questions at the right time, records who answered, and carries every answer through design, build and release.
What changes when Symphony runs around it
| Claude Code alone | Claude Code with Symphony | |
|---|---|---|
| Who decides | You, in the chat, as questions come up | You, at set decision points, with the context attached. Every answer is recorded |
| What gets built | What the prompt describes | A frozen scope with requirements and acceptance criteria you approved |
| Design | Generated along with the code | Every screen and state designed and approved before the build |
| Checks | The tests the agent chooses to write | A test for every requirement before code; design, feature and behavior checks after |
| The record | The chat history | Decisions, checks and fixes on one record |
| Around the product | Up to you | Estimates, proposals, status reports, training and documentation |
Outside evidence: people still matter.
In September 2026, Sierra, a company that builds customer-service AI agents, published hyper-τ-bench, a test of how well AI models can build agents. A top model working alone in Claude Code passed under a quarter of the tasks. Paired with an engineer who had deep context, the same class of model passed more than four in five.
- A top model working alone
- 23.9%
- The same class of model, with an engineer who has deep context
- 82.2%
Claude Opus 5 (max reasoning) running in Claude Code — passes just 23.9%… Paired with an engineer with deep context, the same class of model reaches 82.2%.
Symphony is built on the same idea. The agent does the work. Your team keeps the decisions that matter, at the moments they matter.
Keep the tools you trust.
Symphony runs inside the Claude Code or Codex account you already have. AI usage stays on your plan with your provider. No new AI bill.