This month’s release is one of those that looks quiet at first glance… until you realize a lot of the hard problems just got handled.
Less friction. More control. Smarter agents. And fewer “why is this broken?” moments.
Let’s get into it.
The big shift: agents you can actually trust in production
We’ve been saying it for a while: AI coding isn’t just about generating code. It’s about operating safely, predictably, and at scale inside real environments.
This release pushes that forward in a meaningful way:
- Plan Mode in the CLI
Preview what the agent intends to do before it does it. No surprises. No rogue commands. Just a clear plan you can approve.
- CLI Sandboxed execution + tighter tool restrictions
CLI Agents now operate with clearer boundaries. They stay where they’re supposed to stay. Security teams can breathe a little easier.
- More granular command approvals
Control what runs, how it runs, and when it needs human approval. This is how you move from “cool demo” to “enterprise ready.”
Cost, usage, and accountability (finally, real visibility)
AI costs are easy to spin up and hard to track. That’s not sustainable.
- Token consumption & cost APIs
Get real visibility into usage. Feed it into your own reporting. Build chargeback models if you need to.
- Per-team quota enforcement
Not all usage is equal. Now you can control it at the team level, not just org-wide.
CLI is growing up (fast)
The CLI is quickly becoming one of the most powerful surfaces in the platform:
- CLI Extensions
Extend functionality with your own or third-party modules.
- /ide command
Bridge CLI and IDE workflows directly.
- Generalist Agent mode
Broader, multi-domain problem solving without needing to handhold the agent.
A quieter but important change: code review evolves
The old IDE review tab is gone.
Not because review is less important. The opposite.
Code review is now:
- Agent-driven
- Context-aware
- Powered by coaching guidelines
In practice, that means better feedback, earlier in the process, and fewer cycles getting PRs over the line.
Why this matters
This release isn’t about one flashy feature.
It’s about closing the gap between:
“AI that can generate code”
and
“AI that can operate reliably inside your organization”
That gap is where most teams struggle today.
This release moves you meaningfully closer to the second.
More soon. We’ve got some (really) bigger things coming next.
— Tabnine
