The Simplest Company in the World

The traditional company hierarchy is collapsing. What replaces it will change everything about how businesses are built and run.

Think about the last time your organisation needed to respond to something fast.

A competitor moved. A customer escalated. An opportunity appeared with a short window. What happened next? Someone called a meeting. Someone else said they needed to loop in the relevant stakeholders. A deck got built. The deck went through two rounds of internal review before it reached the person who could actually make the call. By the time the decision arrived, the moment had either passed or the response was so diluted by the process that it barely resembled the original instinct.

Nobody in that sequence was being obstructive. Everyone was doing their job. The system was working exactly as it was designed to work.

That is the thing worth understanding. The complexity is not an accident. It is the answer to a real problem: how do you move information across a large organisation when human beings are the only available transmission mechanism? Management layers exist to solve that problem. They always have. The pyramid is not primarily a power structure. It is an information architecture.

That architecture has just been made redundant.

What Dorsey said, and why it matters

Jack Dorsey has spent the past year in what he describes as existential dread and optimism at the same time, sometimes within the same hour. His question is not about Block specifically. It is about what a company even is anymore, what these structures are for, and whether any of it survives the next few years in recognisable form.

On 2 April 2026, he appeared on Sequoia Capital’s Long Strange Trip podcast alongside Roelof Botha, Sequoia partner and Block board member, to discuss a manifesto he had published called “From Hierarchy to Intelligence.” This is the argument that matters.

Corporate hierarchy exists to solve one problem: how do you get the right information to the right people across a large organisation when information cannot move freely on its own? You build layers. Managers hold context and pass it upward and downward. Directors coordinate across functions. Executives synthesise and decide.

There is a second problem embedded inside the first, and it is one Botha names clearly in the conversation. Hierarchies do not just slow information down. They distort it. People carry agendas. Politics shapes what reaches the top. Emotion filters what gets surfaced. The information that arrives at the decision-maker is not the company’s reality. It is someone’s interpretation of it. Botha’s phrase for what the intelligence layer replaces this with is precise: ground truth. The company becomes entirely legible. Every signal, every action, every decision: visible, queryable, real.

Dorsey’s claim is that both problems have now been solved by something else. Large language models, sitting across every artefact an organisation produces, every Slack message, document, recorded meeting, customer interaction, financial transaction, can surface, synthesise and distribute context in real time. The function that justified the management layer is being performed by the intelligence layer. Which means the management layer is no longer the solution. It is the overhead.

Block has already cut roughly 40 per cent of its workforce. Dorsey wants to compress the distance between himself and his remaining 6,000 employees to zero layers. Not because he intends to manage 6,000 people. Because the intelligence layer manages the coordination and he provides the judgment.

Three roles going forward: builders and operators doing actual work, augmented by agents; owners of outcomes holding strategy and customer accountability; and coaches who raise the capability of the people around them by doing, not directing.

Everything else was scaffolding built around a problem that no longer exists.

One thing Dorsey is emphatic about: this is not a productivity story. Most organisations are treating AI as a copilot, an augmentation layer that makes individuals 10x more effective. His argument is categorically different. This is a structural shift. The architecture of the organisation itself needs to change. Not the tools on top of it. The foundation beneath it.

He also makes a point about timing that is easy to miss. He did not make this transformation under pressure, responding to a situation rather than shaping it. He moved ahead of the moment when others will be forced to make it reactively. That is a meaningful distinction. The companies that act now do so with integrity and when the decision is still theirs to make. The companies that wait will act under pressure, at a cost to the people involved, and to the business.

Why this is possible right now

Most analysis stops at the claim. The harder question is why now specifically. Three things have converged that did not exist together before.

Models that reason, not just retrieve.

Earlier language models were good at finding and reformatting information. What the current generation can do is different: genuine multi-step reasoning across complex, ambiguous, real-world context. Given a sales pipeline, a set of customer communications, a competitive landscape, and a product roadmap, a model now produces not a summary but an analysis. Where the risk is, what the pattern suggests, what the recommended action is and why. That is the cognitive work a thoughtful senior manager was doing. The quality is not perfect. But it is good enough to act on, instantly, at any scale, without a meeting.

Dorsey pushed the tools for three hours every morning throughout 2025, asking whether they could do something he did not think they were capable of. Every single day they surprised him. One year since these tools reached production maturity. One year. And the compounding in that single year has been extraordinary.

Agentic frameworks that turn thinking into doing.

A language model on its own is a capable thinking partner. An agentic system is something different in kind. It is a model connected to tools, data sources, and real-world systems, given a goal rather than a prompt, and allowed to reason through the steps required to achieve it. Frameworks like LangGraph and CrewAI, orchestration layers built on top of frontier models, have made this buildable by a small team in weeks. An agent given a goal monitors, drafts, routes, logs, and escalates. The human reviews the exceptions. The system handles the volume. That entire class of coordination work that previously required a layer of people now runs continuously and autonomously.

The interface has collapsed to a message.

This is the shift most analysis still misses. You no longer need to know which application to open, which workflow to navigate, which system to query. OpenClaw, which became the most starred software project in GitHub history earlier this year, demonstrated that an agentic harness connected to your data, your tools, and your context can be accessed through a WhatsApp message, a Telegram thread, a Slack channel, or a voice note. You describe what you need. The system reasons through the tools required, executes them, and returns the result. Salesforce just shipped 30 new agentic features for Slack, positioning it as the primary interface through which work gets done. Manus, now part of Meta, launched its autonomous agent directly inside Telegram.

The pattern is consistent: the interface is collapsing from hundreds of specialised applications into a single conversational layer connected to an intelligence infrastructure. The simplification is not just about people. It is about the entire application stack. Every tool your team was trained to navigate, every workflow they had to learn, every system they had to coordinate between: all of it accessible through one agentic system that knows the tools, holds the context, and reasons through the execution. The complexity has not disappeared. It has moved. From the surface, where people had to manage it, into the architecture, where the system handles it.

What this actually looks like

Three scenarios. Each one is operational now, not theoretical.

Client relationships at scale.

In the traditional model, managing a portfolio of client relationships requires account managers holding context in fragmented CRM notes, regular check-in calls, reporting processes to surface risks to leadership, and coordination overhead to keep the right people informed. At any reasonable scale this requires a team. The team requires management. The management requires process. The process requires overhead.

This is not automation in the traditional sense, and the distinction matters. Automation replaces a specific task. What replaces the coordination architecture around a set of tasks is something of a different order entirely.

In an intelligence-first operation, every client interaction, every email, every call transcript, every deliverable, every commercial signal feeds into a system that maintains a continuously updated model of each relationship. It surfaces anomalies: this client has not engaged in three weeks, their last two communications carried a different tone, their renewal is sixty days away. It drafts a recommended action and routes it to the one person whose judgment is needed. That person reviews, adjusts if necessary, decides. Everything else is handled.

A team of three operating this way manages a portfolio that previously required twelve. Not because they are working harder. Because they are only doing the work that requires them.

Brand marketing and campaign governance.

A campaign brief arrives. The idea is sharp, the market window is real. In the traditional model it goes to copy, then brand governance, then regional legal, then back to copy for revisions, then back to legal for approval across multiple markets. Six weeks. Three people have left the thread. The window has closed.

This is a familiar sequence for any CMO. But look at what it actually represents. Every handoff in that chain exists because the next person in line does not have the context the previous person had. The process is not bureaucracy for its own sake. It is an information transfer mechanism. A slow, lossy, distortion-prone information transfer mechanism.

In an intelligence-first operation, the brand’s truth, its guidelines, approved claims, market-specific legal parameters, and historical campaign evidence live inside a governed intelligence layer. A brand manager asks through some for of messaging platform: what claims can we make about this product in these markets and what evidence supports them? The system returns the answer in seconds, drawn from the actual regulatory framework and the brand’s own approved knowledge. The brief is written within certainty rather than outside it. Legal reviews intent rather than firefighting misalignment. One approved decision becomes a reusable global asset rather than a one-market call that gets repeated eleven times across eleven markets.

The team required to operate this is a fraction of the size. Not because people have been replaced. Because the coordination overhead that consumed the majority of their time has been absorbed by the intelligence layer. What remains for the humans is the work that actually requires humans: judgment about what the brand stands for, creativity in how it expresses itself, accountability for the decisions that matter.

A new business in a regulated market.

A founder with deep domain expertise and real market access wants to build. Two years ago the limiting factor was not the idea. It was everything required to operationalise it. Compliance infrastructure. Reporting processes. Onboarding administration. The operational overhead arrived before the revenue did, and it arrived with a headcount attached.

Today that overhead is largely deployable through an intelligence layer. Regulatory monitoring routes to an agent that tracks relevant changes and surfaces implications. Financial reporting runs through a system that pulls from source data and produces draft outputs for human review. Customer onboarding is handled by an agent that collects, verifies, and processes documentation, escalating only the cases that genuinely require human judgment. The founder builds. The system operates. The business reaches meaningful scale before it requires the overhead that would previously have been a precondition of it.

The barrier to starting has not just lowered. For people who know their market and have the credibility to access it, it has largely dissolved.

The bet

Several years ago I made a bet. Not based on a market forecast. Based on a pattern I had watched play out across three decades of technology shifts.

My hypothesis was this: software costs are falling toward zero. Model capability is compounding. The cost of coordinating a business, in people, capital, and operational complexity, would fall with it. The organisational structures we had built were workarounds for capability gaps that technology was closing. Once closed, the rationale for the structures would dissolve. Not gradually. In the way important things shift: suddenly, and then obviously.

Each time a genuine capability gap closes, the layer built to bridge it becomes redundant. Not improved, not disrupted. Made unnecessary. I could see the models becoming capable enough. I could see the agentic infrastructure maturing. I could see the cost of deployment collapsing. So I started building toward a world where those gaps had already closed.

I built a venture studio around this belief: that the conditions for building new businesses were about to change structurally. That the people who had struggled to build because of the cost, the capital required, the team size needed, the operational complexity, would find those barriers gone. That the real scarce resource, the one thing a model cannot produce, was distribution. Access to a market. Credibility with a customer. The trust that makes someone choose you over a well-funded alternative.

Dorsey is direct about this. Distribution and access are becoming the most valuable things. The cost of building the intelligence, the capability, the operational infrastructure: that cost is collapsing. What you cannot replicate with a model is the relationship, the reputation, the deep understanding of a specific customer’s world that comes from having been inside it for years.

The bet was right. What I did not anticipate was the speed of the final jump.

Twelve months ago I was building around real limitations, making architectural compromises, working through fragile edges in the tooling. Today those limitations are largely gone. Things I was constructing workarounds for are now solved baseline capabilities. The models reason well. The agentic frameworks are stable enough for production. The deployment infrastructure has matured to the point where a small team with the right architecture knowledge can deploy in weeks what would have been a six-month engineering project even a year ago. Reality has moved faster than the hypothesis.

The gap that is opening

New companies building today start clean. No coordination overhead. No identity wrapped up in roles that exist to solve a problem the intelligence layer now solves. No process debt inherited from an era when information could not move freely on its own.

The intelligence layer is not their destination. It is their starting point.

They will move at a speed that established organisations will find genuinely difficult to understand. Not because they are smarter. Because they are not maintaining infrastructure built for a problem that has been solved. The structural advantage compounds. Every capability release, every model improvement, every new agentic framework widens the gap between the company built around intelligence and the company still managing around it.

The established organisation faces a real choice. It can attempt the transformation Dorsey describes: real, disruptive, and expensive in the near term, and requiring the kind of decision that is far better made when the choice is still yours than when it is not. Or it can keep optimising the existing structure and watch the gap compound. Most will choose the latter until they cannot.

For those building from scratch, none of that calculus applies.

The real opportunity is not in the technology. The technology is available to everyone. It is in the combination: domain knowledge that took years to develop, access to a market that took years to earn, and an operating architecture that can now be deployed in weeks. If you have spent years inside a function, an industry, a customer relationship, you already have what cannot be generated on demand. The infrastructure to deploy around it now costs almost nothing compared to what it cost one year a ago.

That is the moment we are in. Not approaching. In.

The simplest company in the world is not a stripped down version of the complex one. It is what you get when you build without ever adding the complexity in the first place. When the intelligence layer is the foundation, not the renovation.

Build it that way from day one. The window is real and the moment will not wait.

---

Here are all the links to the Jack and Sequoia podcast and articles.

YouTube: https://seq.vc/6b8685

Spotify: https://seq.vc/f8h

Apple Podcasts: https://seq.vc/985228

Full Sequoia page with transcript: https://sequoiacap.com/podcast/jack-dorsey-every-company-can-now-be-a-mini-agi

And the companion article Dorsey wrote: https://sequoiacap.com/article/from-hierarchy-to-intelligence/

Craig Hepburn is an AI strategist and Perplexity Fellow who builds advanced agentic systems. He spent years leading digital transformation at Art Basel and UEFA. Now he works on the harder question: not whether organisations adopt AI, but how they govern it when they do.