AI & Technology

Coinbase CEO gives an AI chief of staff the chasing, and keeps the sign-off

Written by
Full Name
November 14, 2026
Coinbase chief executive Brian Armstrong says always-on AI agents now write code, chase reviews and feed him one ranked queue of decisions, six months after he scrapped “pure managers” at the exchange to cut what he called coordination tax.

Brian Armstrong’s chief of staff is now software. The Coinbase co-founder and chief executive wrote on X on Saturday 7 November that he runs an always-on lead agent for each area of focus, with a “chief of staff” agent overseeing them, and that everything needing his decision now arrives in a single ranked queue.

The post, which drew close to 10,000 likes, is one of the most detailed accounts yet from a public-company chief executive of what delegating to agents looks like day to day. Armstrong has handed over the tracking, chasing and coordinating that usually falls to a chief of staff or a layer of middle management. He says he has kept the sign-off. That split, between coordination that software can absorb and judgement that stays with a person, is the question facing marketing teams as agent tools move from drafting copy into running workflows.

What has Brian Armstrong actually built?

Armstrong’s setup has four layers: Armstrong, the chief of staff agent, a lead agent per focus area, and the sub-agents each lead manages. According to his post, the sub-agents write code, test it, chase reviews and coordinate with Coinbase staff. The leads also bring back ideas he had not considered.

The output lands in what he described as a simple local web app that ranks every item awaiting his decision. Product reviews arrive with screenshots and a video demo, and he clears the queue in batches, mostly by voice and keyboard shortcut. He said working from one queue, rather than switching between dozens of chat windows, has cut his mental overhead.

The system is also meant to learn. Armstrong said it distils his preferences into what he calls a “Brian brain”, so recommendations and first drafts improve before they reach him. The stated aim is fewer corrections per change. Asked whether the setup ran on an internal harness, he replied that it did not, but that it borrowed ideas from xAI’s Grok bot, according to i10x.

Armstrong did not name the models underneath, how many agents are running, what the setup costs or how the “Brian brain” is evaluated. Coinbase has issued no statement on it, and nothing in the post says the arrangement extends beyond his own desk.

Why does one CEO’s agent queue matter beyond Coinbase?

Coinbase has spent 2026 reorganising around this idea. On 5 May Armstrong cut about 14% of the workforce, roughly 700 jobs, and announced a flatter structure of no more than five layers below the chief executive and chief operating officer, with “player-coaches” replacing pure managers. “Layers slow things down and create coordination tax,” he wrote at the time, as reported by Fortune. A chief of staff agent applies the same argument to his own working day.

The company had already put agents into staff channels. In April Coinbase began testing two Slack and email agents modelled on co-founder Fred Ehrsam and former chief technology officer Balaji Srinivasan, and Armstrong predicted the company would have more agents than human employees at some point soon. In September 2025 he said about 40% of Coinbase’s daily code was AI-generated.

Armstrong is not the only chief executive describing agents this way. Microsoft’s Satya Nadella has said custom agents built in Copilot Studio serve as his AI chiefs of staff, and Google has started giving its Gemini agent a work account of its own. What sets Armstrong’s account apart is the plumbing: a named hierarchy of agents feeding one review queue, with each layer’s job spelled out.

For marketing teams, the relevant detail is which work moved. Chasing approvals, following up on reviews and keeping parallel workstreams in step are familiar tasks for campaign and marketing operations managers. Armstrong’s setup automates those tasks and leaves the decision on what ships with him.

How much does the human review actually cover?

Armstrong’s safeguard is one line: every change still goes through the same human review and security checks as before. “It is eliminating tasks, not human judgment,” he wrote. For a listed exchange that holds customer funds, that line carries weight, and it rests on his word. The post does not say who carries out the reviews, whether other Coinbase staff approve work before it reaches his queue, or how approvals and overrides are recorded. i10x noted that the checks cover code changes, not drift in what the agents choose to propose.

Not everyone wants the arrangement. Replying on X, William Morgan said deciding what to work on was the last thing he would want to hand to an AI. Open-source developer Eric S. Raymond said he keeps his own AI use deliberately simple, and argued that elaborate agent stacks are a way of coping with frontier models that are not yet smart enough.

Survey data suggests review is where much of the cost of agents sits. G2’s survey of 1,019 B2B decision-makers, fielded in July and published on 14 October, found 66% of agent buyers had agents in limited or scaled production. It also found 31% spend at least six hours a week overseeing agent work, and 45% say oversight and rework consume more than a quarter of the gains they expected. G2 calls this the “Distrust Tax”. Only 14% reported cutting headcount because of agents, while 35% had moved staff into higher-value roles.

Armstrong has not published the figure he says he is optimising for, corrections per change, or said whether Coinbase will give the same setup to the rest of its staff.

Subscribe to our newsletter

By subscribing you agree to with our Privacy Policy
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.
Share article

Recommended Reading