The most honest feedback you'll ever get might come from something that doesn't worry about getting fired.
The Summary
- Coinbase launched "Coinbase Coach," an AI agent that analyzes internal data — Slack, Google Docs, meeting transcripts, Github — to give employees private performance feedback
- CTO Rob Witoff says he's getting better feedback than from humans: "too in the weeds," not direct enough
- The reports are private, opt-in during beta, and not shared with managers or used in performance reviews
The Signal
Coinbase just crossed a line most companies haven't thought about yet. They built an AI agent that watches everything you do at work and tells you how you're doing. Not a quarterly review. Not peer feedback forms. Daily coaching based on your Slack tone, meeting transcripts, and code commits.
CTO Rob Witoff told Business Insider the agent tells him he's "too in the weeds" and needs to be more direct. This is happening every day. The kind of feedback that usually requires someone to work up the courage, pick the right moment, and hedge their language so it doesn't blow up. Now it's automated, contextualized across multiple data sources, and delivered without the social friction that makes most workplace feedback useless or nonexistent.
"That's the kind of feedback I'm getting every single day now."
The agent isn't evaluating one meeting or one message. It's looking at patterns across Google Docs, Github commits, Slack interactions, and transcripts. It knows if you dominate conversations. It knows if your written communication is vague. It knows if you're doing IC work when you should be delegating. This is the difference between a chatbot and an agent: it has persistent context and it takes action without being asked.
Three reasons this matters now:
- Most performance feedback is garbage because humans avoid conflict
- AI agents can see patterns across data sources humans can't synthesize
- Private, opt-in coaching sidesteps the surveillance problem (for now)
Coinbase is running this in beta. Opt-in only. Private reports, not shared with managers. That's the wedge. If it works, if people actually want it, the next version won't be opt-in. Because if your peers are getting daily feedback and you're not, you're falling behind. And once adoption hits critical mass, the "not used in performance reviews" promise becomes harder to maintain. Not because anyone's lying. Because managers will notice who's improving and who's not.
The Fourth Web business model is starting to show up in unlikely places. Coinbase is a crypto company, but this isn't about tokens. It's about agents that work inside your company's data infrastructure and make everyone a little better at their job without adding headcount. The agent doesn't need a desk or equity. It just needs API access.
The Implication
If you're building for enterprise, this is the wedge. Not "replace workers" but "make workers better at the parts of their job they're bad at because no one tells them." Performance feedback has been broken forever. AI agents can fix it because they don't care about being liked.
If you're a worker, pay attention to which companies adopt this first. The ones that do will move faster. The gap between companies with embedded coaching agents and companies still doing annual reviews is going to be obvious within 18 months.