The New Vanity Metric Is an Agent Count
Elena Verna named it directly this week: a wave of leaders publicly claiming AI is "running their business" — 17 agents, autonomous workflows, full automation — while producing zero revenue lift, zero pipeline movement, zero user growth to anchor the claim. It's a performance. And it's spreading.
For operators trying to make real decisions about where AI fits in their stack, this creates a specific problem: signal degrades. When every team and vendor is confidently narrating their AI transformation, the executives and operators who haven't yet found a working use case start second-guessing their own judgment. The noise looks like proof. It isn't.
What the Theater Actually Costs
The business risk here isn't just wasted vendor spend — though that's real. The deeper cost is opportunity cost on your own team's attention. When internal champions are incentivized to appear AI-forward rather than be AI-effective, you get decks about agents instead of data about what the agents did. You get tool adoption metrics instead of outcome metrics. You get a growing gap between what leadership believes is happening and what's actually moving the business.
This is particularly acute in non-tech companies, where AI literacy at the leadership level is still forming. The executives most likely to be impressed by confident AI theater are often the ones who haven't yet developed a framework for stress-testing the claims.
The Vetting Standard That Fixes It
Verna's prescription is straightforward and worth operationalizing immediately: treat AI vendor and internal team claims the same way you'd treat a new hire or a major capital investment. Demand case studies with named outcomes. Require work trials before committing. Ask for the denominator — not just the wins, but the attempts, the failure modes, the time to value.
For internal teams, this means setting a simple standard before any AI initiative gets air time in a leadership meeting: what metric moves, by how much, over what timeframe? If the answer is a narrative rather than a number, it goes back for more work.
For vendor evaluation, it means asking for a customer reference who will tell you what didn't work before it worked — because production deployments always have failure modes, and vendors who can't name theirs haven't run one at scale.
Actionable takeaway: Before your next AI review — internal or vendor-facing — build one slide that lists every active AI initiative alongside a single measurable outcome it owns. If that slide is mostly blank, that's the finding. Start there.