Ask for agents on big work
When to spawn a specialist agent, when to skip it, and three patterns that work.
Claudioo brings in specialist agents on its own when a task is complex enough to need one. But you don't have to wait for it to decide. Asking for agents explicitly — for the right kinds of work — is one of the biggest habits you can build to get more out of Claudioo.
When an agent is worth asking for
- Deep research that would crowd your chat. Anything that involves reading dozens of files or tracing patterns across modules — the kind of work where the answer is short but the path to it is long.
- Two investigations that don't depend on each other. An agent can explore one thread while the main session works the other. They run in parallel and both finish faster.
- Work risky enough to want a second pair of eyes. A fresh agent, with no memory of how you got there, will catch things you and Claudioo missed together.
Three patterns to keep handy
Run two investigations in parallel
If two questions can be answered independently, ask Claudioo to spawn one agent per question and run them at the same time. The cards land side by side; you read both summaries when they're done.
Spawn one agent to compare our purchase customizations module against base Odoo 18, and another to find every test that touches how we calculate the total on purchase orders. Run them in parallel.
Each agent gets the question, does its own reading, and reports back. Your main conversation never holds the raw search output — only the answers.
Ask a fresh agent to review the work
After Claudioo finishes something tricky, spawn a separate agent to review it cold — no memory of how the solution was reached, just the final result and what to check.
Spawn a reviewer agent. Give it the change we just made to the margin calculation on sales orders, and ask it to verify it against base Odoo — timing, ordering, and dependencies. The reviewer should not see our prior reasoning, and should cite source locations for every finding.
The point of the no-context start is anti-bias: a fresh reader catches things the original author had already accepted as obvious. If the reviewer flags issues, fix them and spawn a fresh reviewer again. Stop only when the reviewer has nothing to flag.
Worth doing for: payment flows, access control, multi-tenant logic, anything that touches money or permissions. Skip for: typo fixes and small UI tweaks.
Send an agent into deep research
When the answer needs a lot of reading before anything useful comes out, hand the whole job to an agent.
Spawn an agent to find every place across our custom modules where we change how the unit price is set on sales orders. Summarize the patterns, flag anywhere standard Odoo behavior is overridden, and tell me which Odoo versions each customization is safe on.
The agent churns through the codebase, you keep working on something else, and you get a tight summary back instead of a 50-message search log in your main thread.
When NOT to ask for one
Agents have setup cost
Spawning an agent takes a few seconds and the agent starts with no context — you have to tell it everything it needs to know in the prompt. For a single file read, a quick lookup, or a one-line clarification, it's slower than asking Claudioo directly. Save agents for the work where the overhead actually pays off.
Also skip agents when the task depends on the back-and-forth of your current chat. Agents can't see your prior turns, so you'd have to re-explain everything they need from scratch.
A note on long sessions
Sessions that lean on agents can take longer to come back. Thirty minutes for a deep, multi-agent task is normal — that's the cost of doing it well, not a smell. Two things to keep an eye on:
- Don't fan out endlessly. Two or three parallel agents is a healthy max. More than that and you're paying overhead instead of buying speed.
- Stay on one topic. Agents don't cure long-thread fatigue. If you've already burned a lot of conversation on one thing and want to switch, start a fresh session — don't try to extend the same thread by piling on agents.
Learn more
- One topic per session — Why focused threads beat sprawling ones
- Run a skill — Skills already spawn agents under the hood
- FAQ: What are agents? — The short answer