Why every change in Kesher starts as a question

Most AI products describe themselves with words that put the AI on one side, acting, and you on the other, watching: agent, autopilot, autonomous. That framing is a big part of why AI feels a little unsettling to a lot of people — it’s built to sound powerful and separate, and powerful-and-separate is the opposite of trustworthy.

The mechanic, not the marketing

Kesher’s actual product mechanic — the agent proposes, and nothing happens until you say yes — isn’t a safety feature bolted onto a powerful system. It’s the same claim the product makes, said in one word instead of a sentence.

When the agent wants to change something, it doesn’t log a change and move on. It asks:

“Can I add a priority field to Invoices? Just adds one new field — nothing existing gets touched.”

Not:

“AI Agent Proposal #4471: Schema modification pending approval.”

(Swap this for a real product screenshot — this is just a placeholder showing how an inline markdown image renders.)

What this means as the product grows

Even as more of the work gets automated — more fields, more views, more of the day-to-day upkeep — the same rule holds: nothing crosses from the agent’s side to yours without your yes. That’s not the whole pitch, though. It’s the floor everything else stands on.

The bigger claim is what it enables: because every change stays structured configuration instead of generated code, you can keep making those changes yourself, indefinitely, without ever needing to hire someone to maintain what the agent built. The approval step is what makes that safe to trust. It isn’t the headline — it’s what makes the headline true.