All posts
Product
6 min read
Why every agent needs an approval step
Autonomy is the goal, but trust is earned one decision at a time. Here is how we designed the approval inbox and why it ships on every plan.
The first question every team asks before handing work to an agent is simple: what happens when it gets something wrong? We think the honest answer is that it will, sometimes, and that the product should be built around that fact instead of hiding it.
Approvals as a design primitive
In Kite, any step that sends, pays, deletes, or changes a record outside your workspace pauses by default. The agent prepares everything it needs, shows you exactly what it is about to do, and waits. One click approves it. Another sends it back with a note.
Drafted emails show the full message and recipient list
Record updates show a before and after diff
Payments show the amount, the payee, and the source of the request
Loosening the leash over time
Approvals are not meant to be permanent. Once an agent has run the same workflow cleanly a few dozen times, you can mark routine steps as trusted and keep review only for the parts that carry real risk. Most teams get there within a couple of weeks.
Next post