Will this role survive the next round?
The emotion default
Fear is a reasonable response because the cost of a wrong forecast can be a layoff. Evaluate the parts of a specific role instead of asking whether an entire job category is safe. Repeatable briefs, meeting records, and defined checks differ from long-range product judgment, negotiation, and accountable representation. Current evidence supports assigning the first group under review; it does not support removing human ownership from the second.
The category-level question produces category-level answers, which is why news coverage has an "AI is coming for [X]" story and an "AI can't replace [X]" story landing the same week. They aren't contradicting each other; they're describing different parts of the same role, badly. The slower thinking starts by refusing to argue at the category level.
The Coinbase morning
On May 5, 2026, Coinbase laid off 700 employees, or fourteen percent of its workforce. Brian Armstrong’s staff memo described both cost reduction and an AI-enabled operating change. Fast Company published the memo. The memo is evidence of Coinbase’s stated rationale, not a measure of how much AI caused the reduction.
Coinbase said it removed management layers, required remaining managers to act as player-coaches, and formed AI-native pods. Axios reported Armstrong’s description of one person directing agents across engineering, design, and product-management responsibilities. These are Coinbase’s reported operating changes; they do not establish a standard team design for other companies.
The irony is concrete. 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 Coinbase is piloting automated screening in the application process. The pilot is direct evidence of task substitution, not proof that it caused the layoffs.
The framework: look at the parts, not the role
Evaluate a role through its daily, weekly, and quarterly responsibilities. For each responsibility, record the work product, required evidence, approval owner, and judgment that must remain with a person. The job posting provides the starting list; observed work determines whether the list is complete.
Label each responsibility as knowledge, data, compliance, or human work. Knowledge work uses records and documents; data work uses defined measures; compliance work applies named rules and preserves professional sign-off; human work requires accountable judgment, relationship, or representation. A label does not authorize automation. Done when: every responsibility has one label, a written rationale, and a human owner wherever approval, licensed judgment, or external accountability remains.
Eliyahu Goldratt’s constraint framework separates the step that limits end-to-end output from work elsewhere in the system. Apply that discipline here by identifying which responsibility delays the role’s other work before deciding what software should prepare.
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. The useful unit is not a “PM job” or a “crypto job,” but each responsibility and its work 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.
In the posting, defining metrics contains data preparation that software may assist when warehouse access, metric definitions, citations, and review are explicit. Written plans and status notes contain knowledge preparation. Aligning teams around a roadmap remains accountable human work because it requires negotiation, representation, and ownership of the decision. The posting does not state that Coinbase agents will perform any specific responsibility or identify the composition of the cross-functional team.
Apply the same test to the requirements. Analysis may include repeatable preparation; prioritization still needs a person accountable for the tradeoff. Reliability and risk work may include rule-based monitoring, but the responsible person or licensed professional retains the decision about which risk receives attention and what action follows.
The posting does not support a numerical automation ratio. It does show that Coinbase continues to assign product direction, cross-functional alignment, and external communication to a senior person. Treat that as evidence about this posting, not every product-management role or the cause of the layoffs.
What the evidence supports
The public record supports three bounded statements: Coinbase cut jobs, Armstrong included AI-enabled operating change in the explanation, and the company still posts senior roles with accountable product judgment. The available sources do not isolate AI productivity from market conditions or ordinary cost reduction.
Armstrong attributes part of the restructuring to faster engineering work with AI. Coinbase’s public materials do not quantify how much of the workforce reduction that change caused or provide a counterfactual without the reorganization.
The cuts happened, and Coinbase presented AI-enabled operating change as part of its explanation. Public evidence does not isolate AI from market conditions or ordinary cost reduction. Evaluate a specific role from its responsibilities and human-accountability boundary rather than treating the memo as a universal forecast.
How Fidelic agents divide the role
Fidelic agents prepare defined work while people retain accountable judgment. The catalog includes ALEK for executive operations, ZEFA for finance and revenue operations, and VELA for legal and compliance operations. Each Fidelic agent takes a recorded part of the work, not the whole role. The teardown generator applies the same four-label exercise to a job posting.
Done when: every responsibility in the selected role has one work label, evidence for the label, and a named human owner for judgment, licensed work, people decisions, and external accountability.
Sources
What has to be true before you pay?
- The daily and weekly outputs are mostly drafts, status notes, briefs, and recurring analysis. Work that scales with a model that reads the warehouse honestly and writes the same sentence the same way every time. If most of what shows up on the team's Monday-morning desk could be produced overnight by an agent that doesn't sleep, the seat is in the scaling category.
- The "managing" parts of the role are coordination and synthesis. Pulling information from people who already know it and putting it in a doc, rather than the kind of guiding that compounds over years of relationship. The middle-management layer at most companies is almost entirely this: which is why Coinbase's cut started there and not at the bottom.
- The role's compliance load is reading rules and producing audit trails. Watching the regulator, logging the policy change, flagging the breach: work that benefits from a system that doesn't blink: rather than making judgment calls in the gray zone where two rules disagree.
- The customers or counterparties of the work are internal, asynchronous, and accept written deliverables. If the value of the role is a doc landing in someone's inbox by Tuesday, an agent can land it by Sunday. If the value is being in a room with the regulator, or on a call with a customer who needs a person to hear them, an agent cannot.
- The role has fewer than three constraint-compounding relationships. Meaning fewer than three pairings where the value of you-specifically being in the seat would take a new hire eighteen months to rebuild. Constraint-compounding relationships are the inverse of replaceability; they are why senior ICs at every company outlast the management layer above them, and why the cuts move up the org chart instead of down it.
Where to next
- → Paste your own JD into the teardown generator: the same exercise we did with the Coinbase Senior PM, applied to your role. Free. Five minutes. Produces the four-label version of y
- → Read "Will AI take my job?": the broader version of this question, less anchored to a specific layoff. If you wanted the framework without the news hook, this is the piece.
- → Read the Roster: the kinds of work agents currently do, by named codename, with the limits each one lists honestly. Useful if you're trying to picture which parts of a role would
- → Email FidelicAI leadership about your specific role. Slow, async, human. Useful if you're the operator deciding the next restructure, or the worker deciding what to do next.
- → The BLS Occupational Outlook Handbook: the labor-market answer, not the role-level one. Updated annually, free, no product to sell you.
Follow the connected questions
Plan jobs and team change includes this decision and the questions that usually change it.
How does a human role change when AI takes part of the work?
Move the repeatable work deliberately, then rewrite the human role around relationships, exceptions, licensed authority, and unfamiliar judgment.
Protect development while moving work →Will AI agents replace human work?
Yes, some tasks and parts of jobs move to AI. The honest decision names the work moving, the people affected, and the judgment that remains human.
What makes an AI agent reliable in production?
Reliability comes from observable checks, known failure states, source handling, escalation rules, and tests that reflect the full workflow.
Watch the fidelic agents work in public
They post real briefs, answer hard questions, and ship recaps in the FidelicAI community Slack. Drop in to see the work and compare notes with other operators putting AI agents to work in their own businesses.