Hey Makers
We’ve just released the new Make AI Agents — and this one is different.
This isn’t a surface update. It’s a full rebuild shaped by your feedback, your builds, and your ambitions.
What’s new (and why it matters for users)
Reasoning panel (Full transparency)
You can now see exactly how your agent thinks, step by step, directly on the canvas.
Every tool call. Every decision. No mystery outputs.
Debug faster. Refine with confidence. Build production-ready systems.
Built natively in the canvas
Agents now live directly inside your scenario builder.
No context switching. No separate setup screens.
You build, run, test, and refine in one place just like the rest of Make.
In-canvas chat for refinement
You can chat with your agent while building to test behavior, probe logic, and iterate quickly without running the entire scenario.
Faster feedback loops = better agents.
File & Multi-Modal Support
Agents can now accept and generate files (PDFs, CSVs, documents, more coming).
Unstructured data → structured action.
Share Across Teams & Workflows
Agents can now be shared across workflows and teams, reducing excess time spent rebuilding.
Build once. Improve continuously.
Library of Agents
Ready-made agents built for real workflows, not demos.
Start with proven patterns for different functions. Deploy in minutes. Customise as needed.
From black canvas to working solutions faster.
Built with the Community
This release wouldn’t exist without you.
Your comments, feedback, edge cases and experiments.
You’ve directly influenced:
- The visual reasoning view
- The simplified agent builder
- The way tools are configured
- The shareability model
And we’re not done.
What’s Next: AI Agents Builder Series
We’re kicking off a new webinar series focused entirely on helping you:
- Understand when to use automation vs AI vs Agents
- Build your first production-ready agent from scratch
- Design with tools, knowledge, and guardrails
- Debug and scale responsibly
More details dropping soon 
We still want to get better
This is in open beta for a reason.
Try it. Stress-test it.
Then tell us:
- What worked?
- What confused you?
- What still feels clunky?
- What would make this your default building method?
Drop feedback here in the community or leave it in the feedback form here.
We’re moving from “AI experiments” to AI you can build and trust inside real workflows.
Thanks for shaping it with us 
Now let’s build what’s next.