canonical_url:
On Tuesday, May 20, Intuit's CEO Sasan Goodarzi sent a memo announcing 3,000 jobs cut. 17% of the company. Reason: "reduce complexity to focus on AI." I read it twice looking for one line. It wasn't there.
The same line has been missing from every AI workforce announcement I have read in the last two years. I want to name it, because the missing line is an engineering-accountability problem before it is a PM-leadership problem, and the people closest to the failure surface (you, reading this on Dev.to) are the ones who feel the gap first.
The Memo Names Three Things. It Skips One.
Every AI restructuring announcement in the last two years has the same shape.
What got cut. Intuit: 3,000 jobs, 17% of workforce. Klarna 2024: customer support function. Duolingo 2025: a chunk of contractor work. IBM 2025: a large internal reorg. Always specific. Always quantified.
What gets refocused. Intuit: "focus on AI." Klarna: AI-first customer service. Duolingo: AI-augmented learning. IBM: "augmenting HR with AI." Always a direction.
What stays. Intuit named existing partnerships with OpenAI and Anthropic. Every memo names what stays. It is the reassurance paragraph.
What's missing. A real human name attached to the failure path of any specific AI system the cut workers used to operate. "When this agent gets it wrong, [name] answers." That line. Zero of the memos name it.
It is the same shape. Cut, refocus, stay, no failure owner. Once you read four in a row, you stop reading four announcements and start reading one announcement repeated four times.
What Actually Got Swapped
Underneath every one of these memos is a structural swap nobody is naming.
The financial side moves cleanly. Payroll down. Software spend up. The controller closes the quarter. The general ledger has every entry it needs.
The accountability side does not move at all. The worker who answered "I made that call, here's why" is gone. The AI that makes the call now has nobody attached to it.
The headcount got swapped for AI capacity. The accountability the headcount carried did not get re-assigned. By default, accountability becomes nobody's job. Not by malice. Not by oversight. Just by the structural shape of a memo that names everything except the failure owner.
This is the workforce-to-accountability swap. It is the unstaffed accountability sitting in every AI restructuring announcement, including the next one your company sends.
Why This Is An Engineering Problem First
Most of the layoff coverage is framed at the executive layer - the CEO's memo, the press release, the analyst reaction. The frame I want here is the practitioner one.
The agent runs in your repo. The pipeline that calls the agent is wired to production. The customer-facing surface the agent writes into is wired to actual humans on the other end. When the agent is wrong, the wrongness lands in code you maintain - a misrouted refund, a wrong tax calculation, a wrong customer email, a wrong policy interpretation surfaced through a chat UI you shipped.
You are closest to the failure surface. You feel the missing line first. The PM who is supposed to author the line is downstream of the deployment by definition. By the time the line is missing in production, you have probably already noticed.
That is why this is worth flagging as an engineer-reader: the missing line is not a future PM artifact, it is a present-tense gap on workflows you are already running.
The Sharper Edge - Voice Agents This Week
Three different vendors shipped voice-capable AI agents in the last week. PollyReach added live voice on a CRM agent. AdaptiveAI shipped triggered outbound phone agents. Crescendo's Live CX agents take full inbound calls.
These agents represent the company in a verbal conversation. The voice is the company's. The commitments the agent makes - "I'll refund that", "I'll escalate this", "I'll send you the contract by Friday" - bind the company in the customer's experience.
No named human is on the call. The agent uptime is 99.98%. The customer is on the line. The named human at the company is not in the conversation, and after the conversation, is not in the transcript.
99.98% uptime is the SRE side of this. Accountability is the other side. Reliability ≠ accountability. You can have a perfectly uptime-correct agent and still have nobody on the line when its output causes harm.
The Engineering+PM Move
The move is small enough to do this afternoon, on one workflow.
Pick one agent or AI-replaced workflow your team runs. One. The one most likely to be wrong about something that matters.
Add a one-line block to the agent's spec (AGENTS.md, prompt header, deployment doc - wherever the agent's contract lives):
# Failure Owner
When this agent's output causes harm, [name] answers.
Signed: [name], [date].
Commit it. PR it. Tag the named person on the PR.
That is the inheritance move. It is the human-side complement to the version control, the test suite, and the SLA. Three accountability artifacts on the technical side (git blame, test owners, on-call rotation) and zero on the AI output side is the current default. Adding the line fixes one of them.
If you are the PM on the team and the engineering side is doing this before you do, you are watching a role-growth opportunity walk by. If you are the engineer on the team and you are already maintaining the agent's spec, the line is yours to write today, and the PM on your team will copy it onto the next four workflows.
The Pattern Will Keep Repeating
The next AI restructuring memo will come this quarter. Probably from a Fortune 500 SaaS company. Probably citing "complexity reduction" or "AI focus." Probably 1,000 to 10,000 jobs. Read it the way I read Intuit's. You will find the cut. You will find the refocus. You will find the stay. You will not find the line.
The line has to be authored. Someone has to sign it. The earliest signer on any given workflow is the person closest to it - which, on most AI-augmented teams in 2026, is an engineer plus a PM, not a CEO writing a memo.
Three thousand jobs at Intuit. Zero named humans in the announcement. The missing line is the artifact. The name on the line is the move.
What's missing from the last AI announcement your company sent - the cut, the refocus, the stay, or the failure owner?
Top comments (1)
honestly the "add a Failure Owner block" move is shakier than the post implies for any agent whose output crosses 3+ teams - whose name goes on the line when the failure is partially a prompt the PM wrote, partially a model regression the vendor shipped, and partially a downstream pipeline another team owns? i underscoped that one.