The Great Rebuild

That's Jack Dorsey, February 26, 2026. The same day he cut 4,000 people from Block — nearly half the company. The stock jumped 20%.

He's right about the conclusion. He's wrong about how to get there.

Dorsey's pitch was clean: "A significantly smaller team, using the tools we're building, can do more and do it better." AI makes people redundant. Smaller teams win. Get there first or get left behind.

Wall Street loved it. The narrative was tidy. AI eats jobs, stock goes up, future arrives on schedule.

Then Bloomberg started pulling threads.

Turns out Block tripled its headcount during COVID — from 3,835 to over 12,400. The pandemic hiring binge. And in March 2025, just eleven months before the AI-themed layoffs, Dorsey sent an internal memo explicitly stating the cuts were "not about replacing folks with AI." That was when he let go of about a thousand people. Now the same restructuring, four times larger, suddenly had an AI story attached to it.

Zachary Gunn at Financial Technology Partners told Bloomberg: "This is more about the business being bloated for so long than it is about AI."

He's not alone in that assessment. A Harvard Business Review survey of 600+ executives found that only 2% had made large layoffs due to actual AI implementation. The rest were cutting based on AI's potential, not its performance. Meanwhile, Challenger, Gray & Christmas reported that 60% of hiring managers emphasize AI's role in layoffs because it's "viewed more favorably than financial constraints."

Oxford Economics put it bluntly, seven weeks before Block made headlines: "We suspect some firms are trying to dress up layoffs as a good news story rather than a bad one."

Goldman Sachs data added context: AI is eliminating roughly 5,000 to 10,000 jobs per month across all U.S. sectors. Block alone accounted for 4,000 in a single day. The math doesn't add up unless most of those cuts had nothing to do with AI.

This isn't a story about AI replacing workers. It's a story about a convenient narrative.

The week before Dorsey's announcement, a viral Citrini Research essay modeled a scenario where AI pushes unemployment past 10% by 2028. It triggered roughly $300 billion in selloffs. Citadel Securities fired back within 48 hours — job postings up, Fed data stable, history says productivity expands consumption.

I don't think either side is asking the right question.

The doomsayers assume AI slots into existing structures and subtracts people. The optimists assume it slots into existing structures and adds productivity. Both take the current structure as given. That's the blind spot. The structure itself is what needs to change.

Here's what I keep coming back to: the real split isn't between companies that adopt AI and companies that don't. It's between companies that rebuild around AI and companies that renovate — bolting new tools onto old architectures and hoping for the best.

Most companies are renovating. McKinsey's global survey found that 88% of organizations use AI somewhere, but only 21% have fundamentally reshaped their workflows. The other 79% added tools to processes designed for humans. And the gap in outcomes is enormous — the companies that actually redesigned are seeing three times the financial impact.

That tracks with what I'm seeing everywhere. Deloitte named an entire chapter of their 2026 report "The Great Rebuild" — and in the same report showed only 11% of organizations are actually running agentic AI in production. Gartner predicts half of all agentic AI projects will be canceled by end of 2027. Everyone knows the rebuild is necessary. Almost nobody is doing it.

And the ones who try to skip the rebuild? They pay twice.

Klarna is the clearest example. They cut roughly 700 customer support roles and layered AI onto existing workflows. Fast, decisive, financially clean. Then quality declined. Customers noticed. The CEO admitted they went too far. They started rehiring into a hybrid model — humans back in the loop, AI handling the routine, people handling the nuance. The same transformation, paid for twice.

Zapier did the opposite. They didn't cut the company and hand survivors a chatbot. They rebuilt workflows around what agents could actually do — 800+ AI agents deployed internally, woven into how work happens. 97% adoption across the org. Not through mandates. Through redesign. When AI is part of the workflow itself, not using it feels like working without email. TELUS took a similar path — 21,000 custom copilots across 70,000 people, $90 million in benefits, nobody replaced.

The difference isn't that one company used AI and the other used better AI. It's that one subtracted people from existing processes, and the other redesigned the processes themselves. Subtraction is fast. Redesign is durable.

"Human augmentation, not replacement" has become the safe thing to say. It's on every enterprise AI slide deck. And it's becoming a platitude that lets companies avoid the harder work.

Because what most companies mean by "augmentation" is: give everyone a Copilot license and call it transformation. That's not augmentation. That's decoration.

The data tells a different story about what real augmentation looks like. The companies in McKinsey's high-performer category didn't just give people AI tools. They changed what people do. Roles shifted. Workflows were redesigned. Decision-making authority moved. The org chart itself had to change, because the old structure was designed around execution constraints that no longer exist.

The World Economic Forum captured this perfectly: "The enterprise was designed for a world where execution was the primary constraint. Today, execution is cheap, abundant and instantaneous. Now the constraint is orchestration."

That's a structural observation, not a technology observation. If execution is no longer the bottleneck, then an organization built around execution — with its hierarchies, approval chains, handoff points, and QA processes all designed to manage human execution speed — is an organization designed for a constraint that no longer exists.

Giving that organization AI tools is like giving a horse-drawn carriage a faster horse. The carriage is the problem.

I spent twenty years at Microsoft. Most of those years were in the field, migrating Fortune 100 companies to Office 365 and Azure. I watched hundreds of organizations go through the cloud transition, and I saw the same split playing out that I see now.

Some companies went cloud-native. They redesigned how teams collaborated, how data moved, how security worked. It was expensive upfront. It required rethinking roles, retraining people and structure, and rebuilding processes that had been in place for a decade. In 2012, it looked like they were overspending. By 2018, they were operating at a different level entirely — faster, more resilient, more adaptive.

Other companies did “lift-and-shift”. They took their on-prem Exchange servers, their file shares, their existing workflows, and moved them to the cloud. Same architecture, different address. It was faster, cheaper, and less disruptive. It also preserved every inefficiency they already had. Those companies spent the next three to five years rearchitecting anyway — paying for the transformation twice, just like Klarna.

The pattern is identical. The technology is different. The stakes are higher. And the clock is faster.

Cloud migration played out over a decade. Companies had time to learn, experiment, catch up. AI isn't giving anyone that runway. The capability curve is steeper, the competitive gap opens faster, and the organizations that wait for a "proven playbook" will find the playbook was written by the companies that started without one.

Here's what makes this urgent, even for leaders who are skeptical of the hype.

The ROI of AI-native operations compounds. Every workflow you redesign becomes a foundation for the next one. Every agent you deploy generates data about how your organization actually works. Every process you rebuild creates institutional knowledge about how to rebuild processes — a meta-capability that accelerates everything that follows.

The McKinsey high performers aren't just 3x more likely to have redesigned workflows. They're pulling further ahead over time, because the rebuild itself is a capability that improves with practice.

Meanwhile, the cost of waiting isn't static. It compounds too — in the other direction. As competitors redesign, the gap between "AI-assisted" and "AI-native" operations widens. The talent market shifts toward people who know how to work in redesigned environments. The vendor ecosystem evolves for rebuilt architectures. Every quarter of delay makes the eventual transformation more expensive, not less.

Starting imperfectly beats waiting for certainty. The companies that rebuilt for cloud in 2012 didn't have a perfect plan. They had a directional bet and the willingness to learn as they went. The ones who waited for the playbook found that the playbook was written by the companies already running on the new architecture.

Dorsey was right about one thing: most companies are late. But not in the way he meant.

They're not late to cutting headcount. They're late to rethinking what the headcount does. They're late to redesigning workflows, not around what AI can replace, but around what humans and AI can do together that neither could do alone. They're late to the rebuild.

And the thing that feels safe — incremental adoption, bolt-on copilots, tactical cuts — is actually the risky bet. It's the lift-and-shift of the AI era. It preserves the old architecture and calls it transformation. It pays now and pays again later when the real work still needs to be done.

The rebuild is already happening. The 21% in McKinsey's data who've fundamentally reshaped their workflows aren't waiting for permission. They're not waiting for proven ROI models. They're learning by doing, iterating, getting it wrong sometimes, and building the institutional muscle that will define the next decade.

The question isn't whether your organization will rebuild. It's whether you'll design the rebuild — or have it done to you.

This dispatch was researched using parallel AI agents gathering data across multiple angles, then synthesized and written in collaboration with Claude Code. The research, the writing, the system, it's all part of the same rebuild I'm describing. Pattern recognition from inside the pattern.

P.S. Speaking of systems - Registration for the course closes tomorrow. Last call at 25% off with code EXPLORER25

Save your spot today - https://maven.com/sashmohapatra/applied-ai-for-kw?promoCode=EXPLORER25

Thanks for reading through.
I’d love to know how you felt about today’s newsletter. This will help me make the newsletter better.