What the AI Layoffs Don't Prove

Last week's AI-driven layoffs in tech, from PayPal and Coinbase to Cloudflare and Atlassian, are colliding directly with data that shows AI is not "working" in large enterprises. So what's actually going on?

Some companies are likely using AI as cover to correct past overhiring. Others have a real AI strategy.

But when you look across multiple reports instead of just one, some patterns emerge.

Tools Accumulate. Knowledge Compounds.

Based on McKinsey's State of Organizations report, an essay from Block's Jack Dorsey that explains the rebuild behind their February AI layoffs, and HBR's piece on the last mile slowing AI transformation, the takeaway is that a lot of companies are stockpiling AI tools with nothing to show for it yet. Only a few standouts are building something that makes the company smarter as it operates, akin to a company brain (more on that below), and that asset is what will pull them ahead. The two paths look the same in a press release, but produce completely different outcomes.

McKinsey's report, a survey of more than 10,000 executives across 15 countries and 16 industries, lays out the picture: 88% of organizations are deploying AI in some form, yet fewer than 20% see significant bottom-line impact. Only 1% of US C-suite executives describe their generative AI rollouts as mature. 86% of leaders say their organizations are not very prepared to use AI in day-to-day operations. From the report:

"Most current efforts to integrate AI focus on fragmented use cases that augment the efficiency of individual contributors. Enterprise-wide rewiring of companies to become agentic organizations remains a challenge."

In other words, a stockpile. Lots of pilots, lots of AI chat tools, and a ton of head scratching - is this actually making these companies better?

Pilot-rich but transformation-poor

In March, Harvard Business Review published "The Last Mile Problem Slowing AI Transformation," based on a Harvard / Microsoft Frontier Firm Initiative summit with senior leaders from healthcare, banking, and industrial manufacturing. They paint the same picture in different words: "pilot-rich but transformation-poor." The examples they cite are striking:

* A global investment bank with more than 250 LLM-connected apps * A global apparel firm with more than 18,000 automated finance processes * A global payments network where 99% of employees use copilots

These are the same companies whose individual workers became more productive but whose enterprise-level value did not move. The HBR authors describe the result as "islands of productivity" that never connect into anything that changes how the business runs. The question is, why?

The companies running these pilots are missing the point. AI shouldn't only be used to take work off people's plates or cut costs by replacing people with automated workflows. It should also be used to capture what the business already knows: its decisions, its customer history, its operational judgment, and so on, so the company as a whole can make smarter decisions. Used this way, AI builds a "brain" that learns as the company operates, gets smarter over time, and makes its people more effective.

"If I Only Had A Brain"

A "brain," in this sense, is not a chatbot or a tool stack. It's a layer that sits on top of the data, documents, and decisions a business already generates, and turns them into context an AI can use. The brain stays fresh because the company's own operating activity keeps updating it.

HBR calls this "Strategic Knowledge Capture": treating tribal knowledge as a strategic asset, pairing senior experts with knowledge designers to externalize judgment, and building systems where that judgment compounds rather than stays static or walks out the door. In March, Block published an essay co-written by Jack Dorsey and Sequoia's Roelof Botha titled "From Hierarchy to Intelligence." They describe building Block as a "company world model": a continuously updated representation of the business itself drawn from the artifacts the company already produces. This includes things like meetings, decisions, code, and customer transactions. The model gets richer every time the company operates. That is the asset. The AI that runs on top of it is just what makes the asset compound at speed.

Companies, especially over the last two decades, have improved their ability to have data and insights at their fingertips that they could use to make their businesses smarter/better/more efficient. But it took teams, time, and debate to try to do so. Now with AI, it can be done much more quickly and easily. The brain learns and grows on its own, naturally, as the company operates.

Is It For Everyone?

Block may be seen as uniquely positioned to do this. It has a transaction firehose and a remote-first culture that makes most work machine-readable. Most operating companies do not have that. What they have is decades of field knowledge, customer history, technician notes, post-mortems, and operational decisions that mostly live in long-tenured heads, spreadsheets, and CRMs that hardly anyone queries. The work of becoming a learning organization is to accumulate institutional knowledge into something the business can ask questions of and act on. AI is what newly makes it possible to build a brain atop what a company already generates. Without the signal, AI runs on top of the cost structure and trims it.

That is why this batch of AI-attributed layoffs got the reaction it did. PayPal, Coinbase, Freshworks, Cloudflare, Upwork, Atlassian, and WiseTech all attributed their cuts to AI, and investors mostly punished them for it. Cloudflare announced 1,100 cuts on the same day it reported a record-revenue quarter and the stock dropped 24%. The market read those announcements as a tell: "We deployed a lot of tools, but we did not build a knowledge system. So we just shored up our financials."

The question to sit with is which pattern describes your company. Tools that augment individuals can be bought. Knowledge that compounds has to be built.