The cut is immediate. The automation is not.

The worksheet below is invented. It is the kind a leadership team might sketch the morning after reading that a well-known company plans to run with roughly 40 percent fewer people. The top half took ten minutes: a headcount, a percentage, a savings line. The bottom half is the list of what those people actually did on a bad Tuesday, the escalation, the review, the fix nobody wrote down. That half is mostly blank, because nobody has been asked to fill it in.

Last Thursday, February 26, after the market closed, Block, the company behind Square and Cash App, said it would reduce its workforce by more than 40 percent. The letter to shareholders from Jack Dorsey, its co-founder and chief executive, put it more starkly: nearly half, from over 10,000 people to just under 6,000.¹² The stated reason was blunt. "Intelligence tools have changed what it means to build and run a company," Dorsey wrote, and a much smaller team using them can do more and do it better. The letter added a prediction: within the next year, most companies will reach the same conclusion.² On X, as quoted by CNBC, Dorsey wrote that the business is strong, "But something has changed," and explained the choice of one cut over repeated rounds: rounds damage morale, focus and trust.⁴⁵ The shares rose as much as 24 percent in extended trading on Thursday and closed up 17 percent on Friday.⁴⁵

My argument is that the cut and the capability arrive on different schedules, and the second one is where the risk sits. The people you can cut is a number a meeting can set in a day. The work you can hand off is a list nobody has until someone writes it: which duties a tool now does and someone checks, which moved to a person who knows they own them, and which the company has decided to stop. The work does not leave with the people. It is automated, reassigned or dropped, and only the first two are choices anyone can see.

This is not an argument against using AI to run leaner. The tools do change how much a team can do, and Block's own engineering numbers may hold up. It is not a prediction about Block's results, which nobody can know yet, and it is not a comment on the people leaving, who deserve respect and nothing less. Nor is it advice to cut slowly for its own sake; Dorsey's point about repeated rounds is fair. The claim is narrower: whatever the size of the cut, the work it leaves behind needs an owner on the first day.

What was counted, and what comes later

Some things were true on Thursday. Block expects charges of about $450 million to $500 million, mostly this quarter, and the plan to be substantially complete by the end of the second quarter.¹ It guided to an adjusted operating margin of 26 percent this year, against 20 percent in 2025, and CNBC reported a goal of more than $2 million in gross profit per employee.²⁶⁵ Those are targets, and Block's own filings warn that actual results may differ materially. But they are the kind of number a market can price in an evening.

Other things become true later, by Block's own account. The letter says the full effect on profit arrives in the second half of the year. On the call, the chief financial officer said the first quarter would show less because of notice periods for employees outside the United States, and described the automation work candidly: some workstreams are "nearly fully rolled out; others are earlier in their maturity." The CFO added that the long-term effect depends on implementing it well at scale, and that Block expects to hire more senior AI engineering talent.²³ The letter's list of risks now includes the reorganization itself and a greater reliance on AI tools.² That is an honest description of a capability still being built. It is also exactly the gap between the two halves of the worksheet.

Five weeks ago I argued that a saved minute becomes a business result only through a decision. Block made that decision, decisively. What it has not yet shown, and could not show on a Thursday, is that the minutes are really there.

Why the skeptics belong in the room

The doubts are not only about Block. In January, Forrester warned that many companies announcing AI-related layoffs do not have mature, vetted AI applications ready to fill those roles, and predicted that more than half of such layoffs would be quietly reversed.⁷ Challenger, Gray & Christmas counted AI as the stated reason for 7,624 of the 108,435 job cuts announced in January, and observed that "the market appears to be rewarding companies that mention it."⁸ On Block specifically, a Morningstar analyst told Reuters the margin gain was near-term and its durability unclear.⁶

There is also context. Reuters reported that Block grew from about 3,800 employees in 2019 to more than 10,000 in 2025, and Dorsey has called earlier overhiring a mistake, according to CNBC.⁶⁵ An analyst at Hargreaves Lansdown read the cut as part AI efficiency and part overdue cleanup.⁶ None of this proves that AI is or is not the cause at Block. It does mean that any leader copying the headline should know which kind of cut they are making. Cleanup removes work that did not need doing. Automation moves work that still does.

The work that is left

Even where a tool does take over a duty, the job does not disappear; it changes. As an earlier essay put it, the exception is the business: when the routine cases are automated, the cases left for people are the harder ones, and they need more judgment each, not less. A thinner team inherits the escalations, the reviews and the knowledge of why things were built as they were. If no one is named for those, they do not stop happening. They simply happen late.

Block's call offered a useful number: production code shipped per engineer up more than 40 percent since September.³ It is a measure of output, and Block reported it as its own metric. Output per person, like gross profit per head, is easy to improve, which is why the dashboard is part of the machine. The things a smaller team can let slip, response times, error rates, review backlogs, are worth watching alongside it.

The work does not leave with the people. Someone has to decide where it goes.

What to check before the number is set

• List the work, not the roles. Recurring duties, exceptions, approvals, incident and customer escalations, and the knowledge of why things are built the way they are.
• Give every duty one of three words. Automated, by which tool and checked by whom. Reassigned, to a named person who has agreed. Stopped, on purpose, with the risk accepted by someone entitled to accept it.
• Match the schedule of the cut to the maturity of the tools. If a workstream is still early, the people carrying it are part of the plan, not the first cost to remove.
• Measure outcomes as well as output. Track response times, error rates and backlogs next to output per person, and decide in advance what would make you slow down.

Back to the worksheet. The top half is finished, and it will look right in any board deck. The bottom half is where the plan either becomes true or does not. Three of seven is not a reason to stop. It is the list of what to do next.