What 4,000 People Knew That Jack Dorsey Didn't Ask
There’s a letter that Jack Dorsey sent to Block’s employees in late February announcing that 4,000 of them no longer had jobs. He wrote it in lowercase. The letter to investors, released the same day, was formatted correctly.
Technology commentator John Gruber noticed. “A telling sign,” he wrote, “about who he respects.”
It was a small thing, and probably unintentional — the kind of detail that only lands because it rhymes with everything else happening at once. But it’s worth sitting with, because it’s actually the whole story in miniature.
In September 2025, Block flew 8,000 employees to Oakland for a three-day company celebration. Jay-Z performed. So did Anderson .Paak and T-Pain. The event cost CAD $91 million, which Block logged as a general and administrative expense in its earnings. Five months later, Dorsey cut 40% of the company.
He was clear about the reason: not financial distress, but AI. “A significantly smaller team, using the tools we’re building, can do more and do it better,” he wrote. “Intelligence tool capabilities are compounding faster every week.”
Block’s stock rose 24% on the announcement. The market, at least, approved.
The framing Dorsey offered was forward-looking and, in its own way, honest. He acknowledged the company had overhired during the pandemic years — Block’s headcount had tripled between 2019 and 2023 — and he argued that it was better to make one decisive cut than to drag employees through years of serial uncertainty. “Repeated rounds of cuts are destructive to morale,” he wrote. On that narrow point, the research supports him.
What the research does not obviously support is the rest of the premise. A 2025 McKinsey report found that most firms are still experimenting with AI implementation, and nearly two-thirds have yet to scale the technology meaningfully. An Oxford Economics analysis found that many layoffs CEOs attributed to AI were more straightforwardly explained by overhiring they were looking to correct. The AI narrative was, as Bloomberg put it, a “convenient cover for structural inefficiencies” — or as Sam Altman has called the broader trend: “AI washing.”
Dorsey may be right. The models did get significantly more capable late last year, and Block’s internal AI tools had produced real results — a greater than 40% increase in production code shipped per engineer since September. He may genuinely be ahead of a wave. Or he may be a founder who over-hired, spent $91 million on a party, and needed a story that didn’t make him look incompetent.
Both things can be true simultaneously. That’s not the interesting question.
The interesting question is what Block’s 10,000 employees knew that never made it into the decisions being made at the top.
Because some of them knew. They always do.
Some engineer on the Goose team understood exactly what the AI tooling could replace and what it couldn’t, and had opinions about the sequencing that nobody solicited. Some manager three levels below Dorsey had watched headcount balloon past any reasonable justification and written it up in a document that went nowhere. Some product person had been watching Cash App’s competitive position deteriorate and had thoughts — specific, informed, operational thoughts — about what was actually going wrong and what might fix it.
None of this is speculation. It’s how organizations work. The information that would most help senior decision-makers is almost always already present somewhere in the organization. The hierarchy’s job, as currently designed, is to summarize and filter it — which is a polite way of saying that most of it disappears before it reaches anyone with the authority to act.
This is not a management failure, exactly. It’s a structural one. The gap between what an organization knows collectively and what its leadership knows individually is a feature of every hierarchy, not a bug in poorly run ones.
Dorsey’s announcement landed as a surprise to employees. The investor letter and the employee note were written for different audiences in different registers, because the decision had been made in a room that didn’t include the people most affected by it. That’s not unusual. It’s normal. It’s the default.
The question Actual exists to answer is: what would it look like if it weren’t?
There’s a fair counterargument here. “Employee voice” platforms have existed for decades. Survey tools, engagement indices, anonymous feedback mechanisms — the consulting industry has been selling versions of this idea since at least the 1990s. If the information existed in organizations and could be surfaced, why hasn’t it been?
Two reasons, and they’re related.
The first is that most of the tools built to surface employee voice were built to measure employee happiness, not organizational intelligence. The annual engagement survey asks whether people feel valued and whether they’d recommend the company as a place to work. These are fine questions with limited strategic utility. They don’t ask what the frontline team knows about the product that leadership doesn’t. They don’t map the divergence between what senior executives believe about the transformation and what the people executing it are experiencing. They produce a number — 72% engaged, up 3 points from last year — that tells you roughly nothing about what’s actually happening inside the organization.
The second is that point-in-time measurement is architecturally unsuited to the problem. Organizations don’t have fixed states that can be accurately captured on a Tuesday in November. They’re dynamic systems, and the signals that matter are the changes over time — the shift in how the engineering team talks about the product six months into a new initiative, the growing divergence between what leadership says about culture and what employees describe experiencing, the early warnings that a transformation is losing credibility with the people being asked to carry it out.
Dorsey didn’t need to know in February that employees were unhappy. He needed to know, in August, that the headcount was becoming indefensible and the organization was starting to sense it. He needed to know, in October, that the AI tooling was generating genuine productivity improvements in some areas and genuine anxiety in others, and that the anxiety was patterned in ways that had predictive value. He needed the organizational intelligence to be longitudinal, multi-stakeholder, and actionable — not a satisfaction score attached to a vague recommendation to “improve communication.”
None of the tools that existed to give him that information were designed for it.
The lowercase note might have been a formatting accident. The $91 million party might have made sense at the time — culture investment, they called it; a signal that the company was healthy and invested in its people.
But the decision to cut 4,000 jobs was made without any systematic mechanism for understanding what those 4,000 people knew, what they were experiencing, or what early warnings were sitting inside the organization that might have changed the calculus.
That’s not a critique of Dorsey specifically. It’s a description of how most large organizations operate. The intelligence exists. The architecture to surface it, mostly, doesn’t.
Block’s employees knew things. They wrote them in Slack and said them in 1:1s and mentioned them in all-hands Q&As that got polite, non-answers. Some of them, probably, were already updating their CVs by the time the Jay-Z concert ended, because they could read the signals even if nobody was formally reading them.
They just wrote in lowercase, to each other, where nobody official was paying attention.
Actual Intelligence builds longitudinal organisational assessment infrastructure for consulting firms delivering transformation work. If the gap between what your client’s employees know and what leadership believes is a recurring problem in your practice, we should talk.