AI is moving fast into places where the stakes are high. Performance reviews. Promotions. Development conversations. The moments that shape someone’s career.

At Zal, we’ve landed on a simple belief: if we’re going to incorporate AI into any feature, our acceptance criteria requires it to be safe, secure, and governable by design. Full stop.

These aren’t three separate concerns. They’re three legs of a stool. Remove any one of them and the whole thing falls over.

Safety means preserving human judgment.

The risk isn’t that AI gets the data wrong. It’s that AI gradually becomes the decision-maker and no one notices. In Zal, every AI-generated output requires a human to review and edit before it goes anywhere. That’s not a guardrail bolted on after the fact. It’s how the system is designed.

Security means treating people data like what it is.

We use masking techniques to ensure no personally identifiable information is ever sent to the LLM providers we work with.

Integrations and data access are configured at both the organizational and individual level, because each employee holds different beliefs about data sharing, and a responsible system accommodates that range rather than overriding it.

Governance means full visibility into what happened and why.

Every AI-assisted output should be auditable.

At Zal, we track edits and surface that data to HR during calibration. If a manager submits a review without changing a single word of the AI-generated draft, that’s visible, and our calibration process is designed to catch it.

Responsible deployment also requires cultural fidelity.

A performance management system should scale what makes your organization yours: your values, your language, your definition of what great looks like, not force you into a generic happy path with minor customizations around the edges.

We think constantly about how AI can improve performance management. But we think just as hard about the people it touches and what they deserve from the systems shaping their careers. That’s what drives every decision described here, and it always will.