The CPOs of Harvey, Glean, and Rubrik Agree on One Thing: Don't Ship an Agent Without a Human Approval Gate

Rubrik’s chief product officer Anneka Gupta told a SaaStr panel that building agents into an existing product turned out to be “a much more challenging problem” than the team expected a year earlier — because doing it right means rebuilding the whole product as a conversational interface, not bolting a chat window onto the old one. She was on stage with the CPOs of Harvey and Glean, and all three converged, independently, on the same design constraint: require human approval before an agent executes a plan. For Rubrik and Harvey specifically, that’s not a nice-to-have — the stakes involved (cyber recovery, legal filings) make unreviewed autonomous execution a non-starter. Harvey now runs 180-plus ex-lawyer “legal engineers” who customize agents to each law firm’s own playbooks, and partner attorneys review the agent’s plan, not just its final draft. Glean’s Emrecan Dogan added a less obvious cost: heavy AI users at Glean now spend roughly half their workday just building context for the agent — feeding it documents, past decisions, and patterns — before it can act usefully at all.

That approval-gate instinct matches what’s already working at government scale. Alberta’s Ministry of Technology and Innovation ran roughly 50 Claude Code agents in parallel to scan 466 million lines of code across 3,400 repositories, work it estimates would have taken 6.5 years manually — but every generated patch still went through human engineer review before shipping, the same design principle the SaaStr panel arrived at independently a year later.

Gupta’s closing point is the one worth sitting with when evaluating an agent vendor’s pitch: companies get held accountable for agent behavior they didn’t explicitly design, and customers “blame us, and rightfully so” when it errs. An approval gate isn’t friction slowing an agent down — on the evidence from a legal-tech CPO, a security CPO, and a government IT ministry alike, it’s the design decision that makes shipping the agent survivable in production.