Chat vs Build: When to Compare Models and When to Chain Them
Chat is one shared thread: @ one agent, then another that continues, or send to everyone. Build chains agents from one prompt. The orchestrator is useful routing, not magic.
By the flAim team
Chat is one shared thread you steer. Build is a chain you inspect. Parallel compare is a useful Chat move — it is not the whole of Chat.
In flAim both live in the same Smart Bar. The difference is the contract with the models, not a new product.
What Chat does
Chat is a single conversation that every agent in the workspace can join. A typical loop:
- 1
@ model A. Ask the question. A replies in the thread.
- 2
@ model B.It already has A's reply and the shared notes. It continues — critiques, tightens, or takes the next step. You did not paste anything into a second tab.
- 3
Ask both. Send the next question without a mention (or to several agents). Both reply with that context. When you need a bake-off, send the same prompt to everyone at once and judge side by side.
Comparison is still Chat. Sequential handoff is also Chat. The thread is the product.
What Build does
Build takes one prompt and chains agents in order. Each step can plan, draft, critique, or refine. Then flAim synthesizes a stronger result. The plan is visible in the workspace. You can inspect the thread. You are not supposed to re-run the playbook by hand.
Decision table
| You need | Use | Why |
|---|---|---|
| To see where models disagree | Chat | Same prompt, side by side. You score the task. |
| A then B, then both | Chat | @ A, then @ B with that context, then fan out. You steer. |
| A first-pass vs a critic vs a long-doc reader | Chat | Same thread. Mentions pick who speaks next. |
| Research, then outline, then draft | Build | Order matters. Synthesis is the artifact. |
| One send instead of eleven copy-pastes | Build | Handoffs are the product. Inspect the plan. |
| A teammate to jump in mid-work | Either | Share the workspace. Mode is independent of multiplayer. |
A simple rule
If you would open three tabs to talk to them in sequence, use Chat. If you would open three tabs to look at the same question, also Chat (fan out). If you would open three tabs to run a playbook you do not want to steer, use Build. If you only have one question for one model, you do not need either — stay in ChatGPT.
Credits: Build runs, long threads, and frontier models use more than a short everyday Chat. Details live in what 100 credits actually buys and the FAQ.
Try the shared-thread loop
@ one agent, then another, then send to both. Or switch to Build for a chain. Free, no API key.
Related: Docs · How we compare models · Pipeline
We're building this in the open and we read everything. If you try it and something's clumsy, tell us. We'd rather hear it from you than not hear it.