Let the agent choose models for each step
A user let Perplexity Computer handle several tasks on its own, and it switched models as needed, including for code writing.
- What they did
- @dani_avila7 launched Perplexity Computer and let it keep working without вмешательство until the jobs were done. The agent decided on its own which model to use at each stage, using Gemini 3 Flash for several steps and then Opus 4.6 when it needed to write code.
- How it went
- It finished the observed tasks through automatic routing between models. The run showed that the agent could change models mid-task without help from the user.
- Tips
- This works best when you can describe the end goal clearly and let the agent break the work into parts. Ask it to stop and ask before any external send, payment, or submission.
Try it yourself with Gemini
I want you to complete this [multi-step goal]. Break it into steps, choose the best available tools and models for each step, and keep going until you have produced [final deliverable]. Before you take any irreversible action, such as sending, paying, deleting, or submitting anything, pause and ask for my approval. At the end, tell me which model or tool you used for each step and why.
Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.
Get 3 agent uses like this every morning
With the day's top AI news. Free, and one click to unsubscribe.
Related agent uses
Set up a chat agent for clinic questions
An n8n workflow used Gemini and Google Sheets to reply to patient chat enquiries with context kept across messages.
Use Gemini to clear email and compare trip options
A writer used Gemini Agent Mode to sort recent mail and help book a Tokyo hotel while leaving final approval for later.
Set up a Gmail summarizing flow with a custom Gem
They built a Workspace Studio flow that summarizes incoming Gmail and writes reply drafts for review.
Watch construction risks and ping the lead
A builder wired n8n and Gemini to watch project data and notify the lead when one issue stands out.
Turn course evaluations into revisions with Claude Code
A professor fed evaluation PDFs to Claude Code, used the results to guide edits, and repeated the process for three classes.
Set up a shared team agent for calls and tasks
A startup team used a shared Hermes Agent to surface call context, manage approvals, and turn decisions into tasks from their shared notes.