The Squeeze
AI is compressing the org chart. The question is where you land.
This Thursday I’m publishing Part 2: Surviving the Squeeze. If the org chart is compressing, what do you actually do about it? I cover five moves: why generalists are winning, how to think of yourself as an orchestrator running multiple agents and workstreams, why surrendering the details matters more than perfecting them, developing taste as a core skill, and how to evaluate whether the company you work at is going to make it through this.
Subscribe so you don’t miss it.
About nine months ago, one of my team members asked if they could help with the social media app I’d been building on the side. Run it like a normal project. Track it, manage it, and bring some structure.
I said no. I was flying. I was the product manager, the project manager, the developer, and QA. My agents were my staff. I couldn’t justify bringing in a project manager. The reason was simple: by the time I explain the intent, the goals, the constraints, the assumptions, by the time we segment ownership and schedule ceremonies, I would already have built it, tested it, and gathered real signal from users.
I wrote about this in S2S a few months ago and framed it as a process problem, a mismatch between how we’ve learned to build software and our new reality. Since then I’ve started to think the problem is bigger than process. Agile was the first thing to stop making sense. The entire org chart is next.
The Memos
Between January and March of this year, a handful of CEOs published detailed plans to restructure their companies around AI. Specific memos with headcounts and new role definitions.
In February, Jack Dorsey posted a note on X to Block’s 10,000 employees. The company was cutting to under 6,000. Forty percent of the workforce, gone in a single announcement. His reasoning was simple: “Intelligence tools have changed what it means to build and run a company. A significantly smaller team, using the tools we’re building, can do more and do it better.”
Stock went up 25%.
Five weeks later, Dorsey and Sequoia’s Roelof Botha published a 4,000-word essay called “From Hierarchy to Intelligence.” The thesis: corporate hierarchy is a 2,000-year-old information routing hack inherited from the Roman army, and AI is the first technology capable of replacing what it does. Middle managers, in this framing, were human routers. They existed to relay information between layers and precompute decisions so executives didn’t have to process everything themselves. AI can do that now, continuously, without losing context or playing telephone.
Dorsey flattened Block into three explicit roles:
Individual Contributors: the builders. They write code, ship product, do the actual work.
Directly Responsible Individuals (DRIs): they own specific outcomes on 90-day cycles. One person, one outcome, full accountability.
Player-Coaches: senior people who mentor and guide Individual Contributors while still building themselves. Hands-on leadership, no pure managers.
That’s the entire org. The management layer between “person doing the work” and “person deciding what work matters” compressed into nothing.
At Meta, a leaked memo revealed a pilot reorganizing roughly 1,000 people in Reality Labs into what they call “AI-native pods.” Every employee gets one of three titles: AI Builder, AI Pod Lead, or AI Org Lead. The ratio: up to 50 individual contributors per manager. Zuckerberg told investors that 2026 would “dramatically change the way we work” and that he’d become convinced “small, talent-dense teams” are the optimal configuration. Meta set a target for H1 2026: 65% of engineers should write more than 75% of their committed code with AI assistance.
Their CFO reported output per engineer up 30% since the start of 2025. Power users were at 80%.
The same pattern is playing out everywhere. Shopify’s Tobi Lutke told managers they must prove a job can be done exclusively by AI before they’re allowed to hire. Shopify’s headcount dropped 30% while revenue grew 21% annually. Salesforce cut 4,000 customer support roles after Benioff said he’d hire no engineers this year. Duolingo declared itself “AI-first” and began replacing contractors with GPT-4.
In Q1 2026 alone, 78,557 tech workers were laid off. Nearly half of those cuts were attributed directly to AI.
Follow the Handoffs
So where do the roles go?
Think of any organization as a graph. There are nodes: the work itself. Analyzing product data. Writing requirements. Creating designs. Building the software. Testing it. Deploying it. Measuring whether it worked. And there are edges: the handoffs between each of those functions. A product manager translates strategy into a requirements document. A designer translates requirements into mockups. An engineer translates mockups into code. A QA team translates code into test results. Each translation is a boundary where context degrades, priorities get reinterpreted, and intent drifts.
The tools for doing the work at each node are maturing fast. Amplitude runs product analytics on a prompt. Figma and AI design tools generate interfaces from descriptions. Claude Code and Cursor write and test software. The individual jobs are getting easier. That part of the story is well covered.
The part that isn’t: the edges. The translation cost between functions. The overhead of turning one person’s output into another person’s input.
That’s what I was describing in my opening. I couldn’t justify bringing someone in because the coordination cost, the explaining, the documenting, the translating, the scheduling, exceeded the cost of just building. The edges were more expensive than the nodes.
In a traditional org, those edges are where most of the headcount lives. Project managers exist to synchronize the edges. Meetings exist to re-align after information degrades at a boundary. Documentation exists because the person writing the requirements and the person reading them sit in different rooms, on different teams, with different assumptions. The entire middle layer of most companies is edge infrastructure.
AI is compressing the edges. When one person can go from product insight to requirements to design to working code to deployed feature, the translation layers between those steps disappear. You don’t need a PRD when the person who understands the user is also the person prompting the build. You don’t need a design handoff document when the builder is reviewing AI-generated interfaces directly against their own intent. You don’t need a QA handoff when tests are written alongside implementation.
The nodes still exist. Someone still has to understand the user. Someone still has to make design decisions. Someone still has to write and review code. The work hasn’t vanished. But the friction between each piece of work, the part that used to require dedicated roles and recurring meetings and Jira boards and handoff ceremonies, is collapsing into a single workflow.
That’s the real squeeze. Map the edges in your own organization. The ones AI can eliminate are where roles disappear. The ones it can’t are where the humans still matter.
The Product Manager Squeeze
Start at the top of the chain. The product manager’s job has always been a hybrid: strategy, analysis, translation, stakeholder management, and occasionally therapy. PMs looked at platform data, user behavior, reviews, interviews, feedback. They made informed decisions about what to build and tracked specific metrics to know if it worked.
Most of that analytical work can now be done by an AI agent in seconds. Amplitude launched what they call “Agentic AI Analytics” in February: AI agents that continuously analyze product usage, identify problems, build dashboards, and communicate findings via Slack. Mixpanel and ThoughtSpot are converging on the same idea from different angles. The whole category is racing toward natural language as the primary interface for data.
The analytical side of product management was hard before. Most PMs didn’t do it well. The tooling required SQL or a data engineer or both.
Now it requires a question typed in English.
The strategic work remains. Setting direction, communicating with stakeholders, making judgment calls about what metrics matter and what tradeoffs are acceptable. That’s still human. But the analytical infrastructure that used to justify a dedicated analyst role, or a full data team, now runs on a prompt.
Design, QA, and the Rise of the Builder
Further down the chain: requirements and design. The design-to-development handoff tools exist today, but they still require heavy supervision. The gap between “what we want it to do” and “working software” is closing fast.
I think the designer and the product manager will merge. Design still matters, but the translation layer between “what we want” and “what it looks like” is shrinking, and the person who understands the user and sets direction will also be the person reviewing the AI’s output and saying yes or no.
QA will be largely automated. It already is for builders working this way. AI agents write tests alongside implementation. They catch regressions. They run coverage. The manual QA cycle where a human clicks through every flow after every sprint is already an artifact.
The builder.
The builder, in this new structure, owns the full chain. Requirements to development to testing to deployment. One person, working with AI agents, shipping a feature that used to require a pod of five or six. This is what Meta’s “AI Builder” title and Dorsey’s “Individual Contributor” role both assume. It’s what I described in S2S: the emerging operator who articulates intent with precision, works directly with AI to produce code and interfaces, and owns outcomes end to end.
There will still be coordination above the builder. Someone needs to ensure the work across builders aligns with company goals, that deployments don’t conflict, that the product holds together. A thin layer. A Player-Coach in Dorsey’s framing. A Pod Lead in Meta’s.
What Happens to Everyone Else
If the builder owns the full chain, what happens to everyone who used to own a piece of it? The project manager holding sprint planning for a team that no longer needs formal 2-week sprints. The QA lead running a test cycle that AI handles in minutes. The data analyst building dashboards that an agent builds on demand.
Some of those roles disappear entirely. Others get absorbed into the builder role. A few shift upward into the thin coordination layer.
And the ceremonies that held these teams together were designed for a world where execution was expensive and coordination was the price you paid. Daily standups lose their purpose when most of your collaborators are machines. Sprint planning falls apart when task estimation is meaningless. Everything is either trivial (seconds to minutes) or so large it’s an entire product. There’s no size 13 on the Fibonacci scale anymore. The XL task is an app.
This compression will be messy. Klarna cut 47% of its workforce and replaced 700 customer service workers with AI. Then they quietly started rehiring humans when the quality collapsed. The line between “AI can handle this” and “AI can handle this well enough” is thinner than the memos suggest. Companies that move too fast will overshoot and pull back. The ones that get it right will move deliberately, testing each collapsed handoff before committing to it.
The Taste Question
There’s a question underneath all of this about taste.
You could argue taste is purely data. Users engage more or they don’t. They come back or they don’t. Run the A/B test. Let the numbers decide. If an AI agent identifies an optimization and the data supports it, ship it. Why wait for a designer’s opinion?
For most decisions, that’s probably right.
But there’s a layer above the data. How does the product make people feel? What do they associate with the brand? Is this thing we’re building getting us closer to what we actually want to be, or are we optimizing ourselves into something nobody intended?
I think about Rick Rubin. He doesn’t play instruments on the records he produces. He doesn’t write the songs. What he does is sit in the room and say “that’s it” or “not yet.” He holds the vision for what the thing is supposed to feel like, and he protects it from the thousand small compromises that would dilute it into nothing.
That’s the role that survives at the edges. Taste at the beginning: what are we building and why. Taste at the end: is this what we meant. One person, course-correcting at a high level, while builders and agents do the work between.
Dorsey said something in his February memo that stuck with me. “I think most companies are late. Within the next year, I believe the majority of companies will reach the same conclusion and make similar structural changes.”
Eleven months before that memo, in March 2025, Dorsey explicitly denied that Block’s layoffs had anything to do with AI. By February 2026, AI was the entire thesis. That’s how fast the handoffs are collapsing.
Every org chart is a map of handoffs. AI is redrawing that map, and the roles that survive are the ones that sit where the handoffs can’t collapse: at the top, where someone has to decide what matters, and at the bottom, where someone has to judge whether the machine got it right. Everything in between is the squeeze.
I publish a few times a week. If you want it in your inbox, subscribe. If you know a builder who’d get something out of this, send it their way.