Coinbase’s CTO says the company’s AI coach gives him the ‘best feedback’ of his career
No one wants to hear from a colleague that they monopolized a meeting or failed to get their point across — and few colleagues want to be the one to share that with their boss.
While thoughtful feedback can be valuable, it’s not always welcome when it’s unsolicited. Nowadays, Coinbase CTO Rob Witoff is getting that kind of candid advice regularly from a source he doesn’t have to worry about awkward workplace dynamics with: an AI agent.
“That’s the kind of feedback I’m getting every single day now,” Witoff told Business Insider.
The company recently rolled out a beta version of “Coinbase Coach,” an internal AI tool that’s been embedded across data sources that workers have access to, including Google Docs, meeting transcripts, Github commits, and Slack messages. It provides employees with private feedback based on interactions with people throughout the company. The reports aren’t used in performance reviews or available to managers, Witoff said.
Employees can opt in for the tool while it’s in beta, and the tool can review those transcripts alongside their Slack interactions and activity across other internal tools to provide weekly feedback, the CTO said.
When Coinbase linked its internal AI models to company data sources, employees across the organization began building their own agents to automate tasks and get feedback on their work. That experimentation helped inspire Coinbase Coach. The company built an early version in March for a small testing group before rolling out the beta internally in June.
Coinbase CEO Brian Armstrong talked about his own personal usage of the tooling in an episode of the “All-In Podcast” in January. He said he used the AI assistant through what he calls “reverse prompting.” Rather than giving the model a task, he asks it what he should do based on the data it’s gathered throughout the company. It might, for example, flag a disagreement within a team over strategy.
Armstrong said he also asked the model questions about what he’s changed his mind about most, or how he could become a better CEO. In response, the model might analyze how the CEO spent his time over the previous quarter and compare that with how he’d intended to spend it.
Pushing for peak performance
In an era hyper-focused on results, continuous feedback from AI coaches may be the next iteration in improving performance.
Witoff said the tool is part of a broader effort for the company to rethink what it means to improve efficiency.
“We are moving toward a culture of proactive coaching,” Witoff said.
The cryptocurrency platform laid off 14% of its staff, roughly 700 workers, in May, eliminating “pure managers” as it shifts toward smaller teams and leans more on individual contributors. On some teams, a single employee may serve as engineer, designer, and product manager.
Witoff said he’s getting some of the best advice of his career because he has an impartial agent who isn’t afraid to tell him the hard things.
“Not so I can be micromanaged,” Witoff said. “So I can get good feedback to be the best version of myself I can be.”
Witoff said the feedback tends to center on two themes. One is communication: The coach will tell him when he isn’t being as direct or clear as he could be, prompting him to sharpen his point, communicate it more efficiently, or articulate a clearer strategy.
The other is knowing when to step back. Witoff said he can sometimes get too tactical, prompting the coach to tell him he’s “too in the weeds” and should focus more on strategy while leaving execution to others.