Will this role survive the next round?

The slower thinking

The Coinbase morning

On the morning of May 5, 2026, Coinbase laid off 700 employees — fourteen percent of its workforce — and the memo Brian Armstrong sent to staff is more useful than the headlines that followed. It is, possibly, the cleanest statement of the displacement thesis any public-company CEO has put on record. "We are not just reducing headcount and cutting costs," Armstrong wrote, "we're fundamentally changing how we operate: rebuilding Coinbase as an intelligence, with humans around the edge aligning it."

The structural change underneath that sentence: Coinbase did not cut a department. It removed a layer. Pure managers — the kind whose job is coordinating individual contributors without contributing themselves — were eliminated. The remaining managers are "player-coaches" required to also ship work, carrying fifteen-plus direct reports. The bigger move sits below them. Coinbase replaced the standard triad of engineer + designer + product manager with what Armstrong called "AI-native pods" — small teams, in some cases one person, who "directs agents that encompass the responsibilities of engineers, designers, and product managers all in one role". The seven-layer org chart was reduced to five.

There is a piece of irony worth naming, because we're going to be honest in this piece and not just analytical. The Coinbase careers page, on the day the memo went out, disclosed that the company "is piloting an AI tool based on machine learning technologies to conduct initial screening interviews" — meaning agents are already replacing the recruiter-level conversation for the very roles being eliminated. The displacement is on the application page. We will sit with that for the rest of the essay.

The framework: look at the parts, not the role

The question of whether a role survives the next round is almost never answered at the role level. It is answered at the part level — by which we mean the daily, weekly, and quarterly outputs that constitute the role's actual work. Most roles have between four and twelve such parts. Some scale with software. Some don't. Some can be done by a careful generalist with the right tools; some require the kind of judgment that compounds over years of context. The parts are usually visible in the job posting, hidden in the form of bullet points that the candidate is expected to skim.

The exercise — and we'll do it in a minute with a real Coinbase posting — is to take a single job description and label each line of it, honestly, with the kind of work it is. We use four labels: knowledge (records, documents, structured information — work that an AI can do honestly with citations), data (numbers, KPIs, trend lines — work that scales with a model that reads spreadsheets), compliance (rules, audits, regulatory edges — work that benefits from a system that doesn't sleep), and human (conversations, judgment, mentorship, accountability — work that doesn't scale with software because the value lives in the relationship, not the output). The proportions tell you the answer. A role that is sixty percent human in its load-bearing parts is not the same job as one that is sixty percent knowledge + data, even if both have "Manager" in the title.

This framework is not ours. The intellectual heritage is older — it's what Eliyahu Goldratt was pointing at in *The Goal* when he separated the work that flows through a constraint from the work that doesn't. What's new is that the constraint, increasingly, is human attention. The parts of a role that can be done by a system that doesn't sleep are the parts that move out of human attention first. Whether the role survives is a question about how much of the constraint work remains after that.

A worked example: the Senior PM, looked at honestly

Coinbase is, at this writing, hiring a Senior Product Manager (or PM — the person who owns what a product team builds and why) for its Exchange product. The salary is $207,485 to $244,100. The role expects seven-plus years of experience. The posting names eight responsibilities and seven requirements. We'll read it the way the previous section described — not as a "PM job" or a "crypto job," but as a list of parts, each with a label.

The responsibilities, listed verbatim:

Own product development from conception through launch and strategically expand core offerings. Collaborate with engineering, design, and cross-functional teams on product roadmap development. Define and analyze metrics to guide product decisions. Align teams around shared vision and steer execution. Communicate plans, progress, and product benefits to internal and external stakeholders.

Read at the part level: "define and analyze metrics" is data work — the kind of thing an agent that reads the warehouse honestly, with citations, will absorb in the next two years, the way a competent analyst absorbed it ten years ago. "Communicate plans, progress, and product benefits" is knowledge work in its written form — drafts, status notes, briefs — and it's already mostly automated for anyone willing to use the tools. The work that doesn't move is "align teams around shared vision" and "collaborate with engineering, design, and cross-functional teams on product roadmap development." Not because the writing-down of the roadmap can't be done by a system — it can, and the agents that Coinbase's surviving PMs are directing will do that part — but because the aligning, the calibration of seven engineers and three designers and a regulator and a treasury team around a single bet, requires somebody who has been in the room with each of those people, repeatedly, for long enough to know what each of them is afraid of. That is not a chat thread. It is a constraint that compounds.

The requirements line is the same exercise. "Strong analytical, prioritization, and problem-solving abilities" — the analytical part scales, the prioritization doesn't, because prioritization is the act of putting one number ahead of another in front of people whose careers depend on the answer. "Demonstrated leadership guiding product strategy with senior engineers and designers" — guiding is the human part, and it doesn't scale because the senior engineers and designers won't let it; they will read your taste, your character, and your willingness to say no, and they will keep an agent in the loop only as a tool, not as a leader. "Institutional grade focus on reliability, scale, and risk mitigation" — the mitigation part is compliance, and an agent can watch the rules better than a human can; the focus is the constraint, the human bit that decides where attention goes when three risks land in the same week.

If we tally honestly: this role, at the senior-IC level Coinbase is currently hiring for — individual contributor, meaning a senior person who does the work themselves rather than managing others — is probably forty percent parts that scale and sixty percent parts that don't. That ratio is why Coinbase is hiring at this seniority while cutting at the layer above it. The pure-management layer was almost entirely composed of the scaling parts — coordination, status synthesis, calendar logistics, written communication — and the value of the human in the seat was decided by how much of the constraint work the rest of the role contained. The middle layer at most companies has very little. The senior IC has more. That is why the cuts move up the org chart from the middle, not from the bottom.

Where Armstrong is right, and where he is narrative

Armstrong's framing of the Coinbase cuts is partly true and partly executive-narrative — and the worker reading this deserves to know which is which, because the difference changes the answer to the question.

The part that is true: engineers at Coinbase, and at almost every well-tooled software company, can now use AI to ship in days what used to take a team weeks. Armstrong said exactly that, and Goldman Sachs's Joseph Briggs has been tracking productivity gains that support a real, measurable shift in the engineering function specifically. The non-technical teams shipping production code at Coinbase is a real phenomenon, not a press release.

The part that is narrative: Coinbase's actual revenue pressure, the thing that decided the timing of the cut, is not AI productivity — it's the crypto down-market that has compressed Coinbase's trading volumes through Q1 and Q2. The "AI-native pods" framing is partly the genuine restructuring underneath, and partly the version of the story that lets a public-company CEO talk to Wall Street about a layoff without saying "the market is bad and we got expensive." Sam Altman, of all people, has warned about "AI-washing" — companies blaming AI for cuts that would have happened anyway, because "AI-driven efficiency" is the only narrative the market currently rewards. We mention this not to discount Coinbase's restructuring, which is real, but to be honest about the answer to the question. The Senior PM role at Coinbase is not surviving because Coinbase's product strategy is more clever than its peers'; it is surviving because the parts of the role that compound were never the parts that the down-market made expensive.

The part to hold both at once: the cuts happened, the productivity gains are real, the narrative is partly true. The worker whose role just got cut is not consoled by the productivity gains, and shouldn't be. The role that survives is the one whose constraint parts couldn't be moved into the agent loop, regardless of whether the cuts were AI-driven or market-driven. The honest question — will this role survive the next round — is answered at the part level, in the posting, with the four labels above, regardless of what the memo says.

A note on what we do, plainly

FidelicAI is, in the simplest description, the company that does what Armstrong said Coinbase is doing now — building agents that absorb the parts of roles that scale, around humans who hold the parts that don't. We sell agents named ZADO-01 (knowledge), DRYN-01 (data), VELA-01 (compliance), and a few others — each one designed to take a specific part of a role, not the whole role. If you've gotten this far in the essay and want to see the same exercise applied to your own posting, the teardown generator takes a JD and produces the four-label version above. We make money when the answer to "which parts of this role scale" is honest, which is why we have written the essay this way.

We also believe the worker whose role just got cut is owed the same honesty as the operator deciding which roles to consolidate. Both audiences read this page. The answer to the question is the same for both: look inside the role, count the parts, and the structural answer is more available than the comforting one.