Inside Block’s AI-Native Organization

Like everyone, I’m very curious how AI will impact individual jobs and more importantly, company organizational models. Initially AI seemed to be affecting individual contributor roles. Now we are starting to see real innovation emerge for AI-enabled organizational design. One of the most interesting is what Block is doing under Jack Dorsey.

In February 2026, Block laid off 40% of its staff, citing gains from AI. Here's Jack's announcement:

The layoffs aren't the story. This is a public case study where a CEO is explicitly saying AI changes a company's fundamental structure. This is where AI impacts management structure.

“we’re already seeing that the intelligence tools we’re creating and using, paired with smaller and flatter teams, are enabling a new way of working which fundamentally changes what it means to build and run a company. and that’s accelerating rapidly.”

– Jack Dorsey, Block Head

If you want to hear more about this directly from Jack, this is worth watching:

Block is showing what it looks like to fundamentally rethink the role of AI and organization design, ultimately harnessing AI to increase speed as a compounding competitive advantage.

Traditional Organizational Models

Before we dive into what Block is doing, it helps to ground ourselves in how most companies currently operate. Most organizational design stems from the hierarchical models implemented during the industrial revolution. Those models evolved from military structures. For example, the military formalized the distinction between "line" and "staff" functions. Line advances the core mission. Staff provides specialized support. Corporations still use this vocabulary today.

Management structures are designed for command and control, top-down decision making, and "vertical" information flow (e.g. up/down the hierarchy). Most companies have a pyramid-based structure that looks like this:

• CEO at the top
• Layers of management underneath
• Strategy/direction flowing down
• Information flowing up
• Individual contributors "doing the work"

At its core, management is an information system. Layers exist to move, filter, and interpret information. Using humans introduces some well-known failure modes:

• Information is filtered and/or distorted. Human managers are notorious for telling management layers above them what they want to hear.
• Human managers inevitably get caught up in human concerns: avoiding risk, taking credit, competition among peers, and internal politics.
• Human managers may not challenge the views and strategy of senior management, choosing instead to "support their team". E.g. loyalty over truth.
• Humans are limited in the number of people they can directly manage. So larger companies naturally tend to have more layers and more bureaucracy.

Information flow can also be impeded by corporate bullsh*t, and more generally by the inertia of enterprise despair.

Even assuming accurate, timely information traditional "managerial" strategic planning faces an existential challenge. The world is just too complex and unpredictable. Responding to such a world requires greater adaptability, and adaptability is not a characteristic of rigid strategies and top-down hierarchies. See my post: Planning vs. Reacting.

General Stanley McChrystal (command of Joint Special Operations Command "JSOC" from 2003 to 2008) recognized that hierarchical command and control organizational models could not respond in real time to what was happening on the battlefield. In his book "Team of Teams" he describes how he restructured US forces from the ground up on principles of extremely transparent information sharing (what he called “shared consciousness”) and decentralized decision-making authority (“empowered execution”).

"Adaptability, not efficiency, must become our central competency."

– General Stanley McChrystal, Tantum Collins, David Silverman, and Chris Fussell, Team of Teams

Block is experimenting with something similar, with AI driving the "shared consciousness" and directly responsible individuals heading the "team of teams".

Block's "AI Inclusive" Model

We have known for years that tools like email, slack, and internal websites (remember "portals" and "intranets"?) can replace some of the functions of management layers. Yet these tools can also be used to further individual political agendas, spread corporate bullsh*t, and over time they become a source of noise. In other words, these tools still incorporate the failure modes of human management.

We have also tried to build data warehouses with reporting and analysis tools layered on top, executive dashboards, etc. Generally, these attempts fell short. Why? Data quality, timeliness, cycle time, lack of flexibility, etc.

The real insight is that it's not just about using AI in specific use cases like software engineering. AI can directly interact with the company’s sources of truth. Support tickets. Financial data. Operational signals. AI can read and summarize thousands of ZenDesk support tickets. It can see your general ledger transactions. It can monitor channels in Slack or Teams. It can see customer interactions via Salesforce.

Instead of relying on humans to summarize reality, AI can work directly on raw data. You can now have a “conversation” with your business that is closer to the truth than any management reporting chain. Closer to the customer. Closer to the work. Closer to reality.

Block is even attempting to create a "world model" of their business to further enable AI. The understanding of the "state" of the company would flow from this model. This leads to better decisions, better features, and ultimately better products.

Historically, organizational cohesion came from management layers. Now it comes from an intelligence layer consuming raw data. Block’s model is not a pyramid. It’s a system. Minimal management. Clear ownership plus real-time context.

In a conventional company, the intelligence is spread throughout the people and the hierarchy routes it. In this model, the intelligence lives in the system.

It's a network of humans coordinated by AI. Think:

• Nodes, not levels. The org becomes dynamic.
• Ownership, not titles. Teams form around problems.
• Flow, not reporting. No more filtered reporting chains. Everyone works off the same context.

The Human Roles that Remain

What are the "durable human skills" that a company requires, and how are those skills expressed in roles? Here's Block's take:

0. Leader: Perhaps this role can be viewed as a specialized "directly responsible individual" but there is still a CEO/Leadership role. Jack refers to himself as the "Block Head". Durable human skills at this level are judgement, the ability to attract the right people, set values, culture, and the "tone at the top". This role also drives customer focus and execution.
1. Directly Responsible Individuals (DRIs): DRIs own outcomes. They may own customer outcomes and strategy/strategic decisions. They may be accountable for a problem being solved. Not “managers”, not “approvers”, they pull in people as needed for the mission. Authority comes from ownership, not title. The durable human skills are agency, ownership and accountability.
2. Player-Coaches (PCs): They help others get better and master their craft. They teach, guide, not control. They still do real work (because they must demonstrate mastery be respected by ICs). The durable human skill is the ability to develop human capacity with compassion and empathy.
3. Individual Contributors (ICs): They focus on execution, not navigating hierarchy. They are salespeople, builders, engineers, designers, operators, etc. Their durable human skills are judgement, taste and creativity.

Block's Key Metrics

Block is anchoring the model with a simple but powerful metric: ~$2M+ gross profit per employee (up from ~$500k pre-COVID). This is hard to game. It exposes inefficiency quickly. In Jack's view is the best measurement of "bloat" inside an organization.

They’re also compressing organizational depth:

• Maximum layers between CEO and any employee: ~5 today
• Target: 2–3 layers in the future

The direction is clear: fewer layers, tighter feedback loops, faster decisions.

Four Key Business "Layers"

Block sees it's business as composed of fundamental capabilities which are composed into an interface for customers to access. On top sits a proactive intelligence that sees the customer using the capabilities and helps optimize both the customer experience and the capabilities and interface itself. Finally there are world models for the company and customers that enable AI to take an even larger role.

1. Capabilities. The things that Block can do for customers. These are like pure utilities. Card issuance, card acceptance, lending, banking, blockchain, etc.
2. Interface. The UI. Distribution. E.g. Square register, the cash app, the dashboard. It's where customers interact with and experience Block's capabilities.
3. Proactive Intelligence. Real-time understanding of customers, their activity, their pain points, etc. The intelligence layer can proactively prompt customers to alert them to things that they should pay attention to, or they might consider doing.
4. World Model. This is actually two world models: A world model for customers, where they can see all their behavior to optimize it, and a world model for the company that sees all its inputs and outputs, how it operates, risks and failure modes, etc.

Block's AI Tools/Tooling

1. Goose - Goose is an agentic harness that works on top of any model. Jack equates it to an internal Claude Code or OpenAI Codex. Think of it as a runtime for AI agents that lets agents:

   • Call tools (payments, data, APIs)
   • Manipulate CLI tools
   • Take actions (not just chat)
   • Coordinate workflows

Goose likely integrates deeply with Block’s ecosystem via internal MCP tools, APIs, data access, architectural knowledge, etc. Goose aligns with Dorsey’s vision of models that that do more than just think. Many other things are built on top of Goose. For example, Builder Bot, Manager Bot, Money Bot sit on top of Goose.

In the movie Top Gun, Goose was technically Maverick's Radar Intercept Officer (RIO) but most people think of him as the co-pilot.

New Capabilities

This model unlocks capabilities that weren’t practical before:

• Customers building features directly, within constraints. Imagine customers building their own features using Block's AI tooling. If the feature aligns with Block's strategy and Block's overall road map, go ahead and let the customer implement it at their point of need. If the feature doesn't align with Block's strategic road map, then the tooling would prevent the customer from doing so.
• Imagine an intelligence layer that sits on top of all customer care activities, a layer that's aware of all customer support calls, customer issues, and customer concerns. It can read call transcripts and can see the key issues that are generating customer care calls. It could use that information to proactively set priorities and feed that into builder bot.
• AI identifying patterns across support, usage, and operations
• Automatic prioritization of work based on real-world signals
• Shortening the loop from problem → solution dramatically
• Dynamic pricing and personalization at scale

It also introduces a fundamental shift from roadmap-driven → customer-driven. Instead of asking “what should we build next?”, you ask: "Why does a roadmap exist at all if customers can express needs in real time?"

Block's Mindset

• Want to think like a Frontier Lab that is building the company as the intelligence itself. Can't just layer on models and intelligence, it must be at the core of everything they do.

• It's not about a particular person or role doing more with AI. It's about how the entire company must change to leverage the capabilities of LLM models.

• Reduce layers of people. There should be very few layers between the head of the company (the Block Head) and individual contributors. Also want to make everyone closer to the customer.

• Companies aren’t productivity problems but information flow problems. Hierarchies were built because humans couldn’t process everything. Now that LLMs can, the structure can be simplified. It’s a "system" that effectively zeroes out the administrative burden of being a large organization.

• It’s not “what should we build next?” it’s “why does a roadmap even exist if Block can read/interpret customer signals in real time?”

• Make the core an intelligence model and humans exist at the edge. This replaces the "command and control" pyramid with a "circle" where customer intent dictates the roadmap in real-time.

• The key durable human values left are agency and ownership.

Summary

AI isn’t a productivity layer—it’s a forcing function for a new organizational model, with Block as the most concrete example of what that looks like in practice. (Especially the shift toward agents, metrics, and structure—versus just tools.) Block isn’t just flattening the org chart. They’re trying to delete it and replace it with:

• A layer of intelligence
• A network of responsibility, coupled with coaches and players
• A much shorter distance between idea and execution

The deeper insight is that companies' key challenges were never really management or productivity problems. They are information problems. Hierarchies exist to manage information flow. Now AI can see the state of the company in real time. Shared context, or "shared consciousness", can be available to anyone, everywhere at any time.

This moment feels both exciting and uncomfortable. Block made a massive bet on the future which looks:

• Less like “AI will make people more efficient.”
• More like “AI will force companies to become something entirely different.”

References

• AP News: Fintech company Block lays off 4,000 of its 10,000 staff, citing gains from AI
• The Org: How the Rise of the Railroads Gave Way to the World's First Org Chart
• Video: On the Block #1
• Block: From Hierarchy to Intelligence