Most organizations I work with are treating AI as an overlay. They bolt agents onto existing teams, sprinkle copilots through existing workflows, and hope that productivity compounds. Some of it does. But the real shift happening in 2026 is not about overlaying AI on what you have. It is about rebuilding the company around the premise that AI changes what a company architecturally is.
In the first week of January 2026, Jack Dorsey's leadership team at Block sat around a table and asked one question. Would we build this company the way we built it, if we had today's tools from scratch? The answer was uniformly no. Three weeks later, Block laid off 40% of its workforce, 4,000-plus people from a company of 10,000. The stock went up 26%. Quarterly gross profit was already up 24% year over year. Dorsey's public statement left no room for euphemism: "intelligence tools have changed what it means to build and run a company."
Most coverage filed the story as another tech layoff. Dorsey insists it is something else, and after reading his manifesto "From Hierarchy to Intelligence" (co-authored with Sequoia's Roelof Botha, March 2026) I think he has a point. What Block is trying to do is a structural bet on a fundamentally different company architecture.
The 2,000-year hierarchy problem
According to the manifesto, corporate hierarchy exists for one reason. It routes information through an organization too large for any single person to oversee. Roman army, Prussian General Staff, modern corporations, all stem from the same need. And naturally, when you narrow the span of control you need to add layers of command as you grow. Organizations have been trapped in this shape for roughly two millennia.
AI breaks the geometry at Block. Every artifact of modern work, from Slack messages to Google Docs to pull requests to meeting recordings, is machine-readable by default. Route all of that through a world model of the company, and the information-routing function of middle management becomes a feature built into the software layer. In Dorsey's words, the company becomes a "mini-AGI," meaning a queryable intelligence that aggregates every work artifact into one model anyone can interrogate. Any board member, any employee, any customer can now have a conversation with the company about how the company is doing.
Block is rebuilding around this premise with four structural layers. Capabilities are the atomic primitives. Interfaces are the delivery surfaces: Square, Cash App, Tidal in Block's case. A proactive intelligence layer prompts customers before they prompt the company. And two world models (one of the company, one of the customer) sit underneath everything. Money, Dorsey argues, is the most honest signal in the world. A transaction tells the truth, and billions of transactions become the signal fuel for the customer world model, which feeds everything else.
The three roles that survive
Block is collapsing the traditional org chart to three roles.
Individual Contributors build and operate, augmented by agents. One IC now does work a team of ten used to handle. Their durable human skill is judgment, taste, and creativity.
Directly Responsible Individuals own customer outcomes on 90-day cycles and assemble ICs around specific problems. Their durable human skill is ownership and accountability.
Player-Coaches build human capacity by doing the work alongside people rather than supervising from above. Their durable human skill is empathy and coaching.
These roles are assignments. And there is no reporting hierarchy between them. Notice how directly this maps to this issue's leadership section on Gardner's six-two split. The three surviving roles all require strength in the two intelligences AI cannot replicate: interpersonal and intrapersonal. The roles that absorb into software are the ones that were primarily about routing information between other humans.
What the economics say
Anthropic generates around 30 billion dollars in annualized revenue with roughly 3,000 to 5,000 employees. Revenue per employee: five to ten million dollars. Google needed 32,000 employees to reach 30 billion. Salesforce needed 79,000. Cursor crossed 100 million ARR with about 20 people. Midjourney runs at half a billion annual revenue with 40 employees and zero outside funding. Traditional SaaS has operated at 200 to 300 thousand revenue per employee for two decades. The AI-native companies are running ten to thirty times that.
Structure follows economics. Anthropic uses "Member of Technical Staff" as a default title across many roles. Reporting lines are deliberately flat, and people move across problems fluidly, re-forming around work as it emerges.
The skeptical counterweight
There is a counter view to consider. Josh Bersin's research across 70-plus companies argues that most AI-driven layoffs are performance management dressed in AI language. Replacing a 35,000 dollar support agent with AI often costs 15 to 20 thousand per year, not zero. For a 100,000 dollar engineer, inference tokens on complex reasoning can even exceed the human salary. Real ROI, Bersin argues, comes from business re-engineering, with Allianz, Travelers, and HubSpot scaling services rather than shrinking.
Both views can be true. Block may be genuinely rebuilding and also cutting faster than the new architecture is ready to catch. The tell is whether three things happen at once: actual architectural work (world models, composable capabilities, not just agent pilots), revenue per employee rising over time rather than ratios moving from denominator cuts, and remaining people spending their time differently (more prototyping, fewer status meetings). Without all three, "AI-native" is a cover story. With all three, it is a structural shift.
Your action step
Three questions for the next three leadership team meetings.
-
Dorsey's question. If we started this company today with current tools, would we build it the way we built it? Write down what would be different. That gap is your roadmap.
-
The role inventory. Sort every title in your company into four buckets: Individual Contributor, DRI, Player-Coach, and Information-Routing-Only. That fourth bucket is where middle management has historically lived. AI is absorbing it. What should replace it in your company?
-
The RPE trajectory. Track revenue per employee quarter over quarter. Flat RPE despite AI investment means you are overlaying. Rising RPE over multiple quarters means the structural bet is paying off.
The conversation this surfaces is about company architecture. Leaders who start having it now will be on the other side of the 2,000-year problem before their industry realizes the geometry just changed.
If you'd like to work through the AI-overlay vs AI-native question for your organization, or want me to run an executive session with your leadership team on company architecture in the age of AI, I'd love to help.