Jun 12, 2026 4 min read Anthropic

Anthropic Management Structure: One Direct Report

Anthropic's org chart narrows to a single direct report between co-founders. That unusual structure shapes how decisions move — and where pressure builds.

Dario Amodei has just one direct report

We recently learned a strange detail about how top AI labs organize themselves. Dario Amodei is the CEO of Anthropic. He spoke on Bloomberg's The Circuit with Emily Chang. In that interview, he revealed that he has just one direct report. That report is his chief of staff, Avital Balwit.

Connie Loizos on TechCrunch reported more details on this setup. Every other executive at Anthropic reports to President Daniela Amodei. Daniela is Dario's sister and co-founder. Private investors recently valued the firm near $1 trillion. Daniela runs day-to-day work and answers to the board.

This setup is very rare for a tech firm of this size. It matters because it shields a technical leader from daily human resource tasks. Fast-growing startups usually spend a lot of time on these issues. By giving up daily management, Dario can focus on safety, research, and long-term strategy. He avoids tasks like performance reviews and budget fights. He also avoids coordinating departments.

The mechanics of the split-leadership model

Why use this model? We must look at what usually happens as startups grow. At first, a small team focuses only on building. As the company scales, the technical founder becomes a manager of managers. They spend their days solving fights, aligning teams, and doing paperwork.

Anthropic splits the work instead. Dario acts as the intellectual anchor. Daniela runs the business. The interview notes that this split protects Dario's mental energy. He can spend his time writing policy papers and setting safety benchmarks. He can also steer the direction of the models.

This structure also addresses a cultural risk that fast-growing startups face. Fast companies recruit heavily from big software firms. Amodei noted that you must show new hires how your company works. Otherwise, they will simply build the bureaucratic models of their old employers. Anthropic stops this by keeping the CEO's office small. They centralize management under Daniela to limit corporate structures inside the research core.

Spans of control across the AI sector

This one-report structure is a sharp departure from how other tech leaders operate. Spans of control in Silicon Valley are usually much wider, even at the top.

For example, OpenAI's Sam Altman reportedly has about six direct reports. This is a standard size for a tech CEO. Satya Nadella manages about 16 direct reports at Microsoft. He must balance a vast array of business units. Sundar Pichai oversees roughly 18 direct reports at Google.

Nvidia's Jensen Huang is on the other extreme. He famously has around 60 direct reports. He relies on a flat, highly collaborative approach. He calls this extreme co-design. He notes that this setup prevents corporate layers. However, it makes traditional one-on-one professional development talks impossible.

Dario Amodei’s approach is a structural outlier. It assumes that the main bottleneck for an AI lab is not communication speed or operational flatness. Instead, the real bottleneck is the cognitive load on the person guiding the technical vision.

The risk of single-point failure

This design helps a research-focused CEO, but it brings operational risks. Anthropic funnels all executive-level reports through Daniela Amodei. This creates an intense concentration of operational authority.

If Daniela Amodei were to step away, the company would face an immediate crisis. There is no obvious public succession plan for her role. Her job bridges the gap between a hands-off, research-focused CEO and the rest of the executive suite.

This setup also risks splitting the strategic vision from daily business reality. Friction can develop if the person steering the models is cut off from the people executing the business plan. The chief of staff role becomes incredibly powerful. In this model, that person acts as the sole filter for what information reaches the CEO.

The survival of the founding team

Amodei shared another surprising detail in his talk with Emily Chang. All seven of Anthropic's co-founders are still at the company. This level of stability is very rare in the volatile world of AI builders.

For comparison, almost all of OpenAI's original co-founders have left over the years. They split over strategic disagreements, governance disputes, or competing ventures. Anthropic has kept its core team together for over five years. This suggests their internal governance has worked well. Perhaps this division of labor has helped them manage the intense pressures of building a frontier AI business.

We do not know if this model can survive a transition to a public company. Public markets generally expect a CEO to be deeply involved in financial and operational details. Also, investor relations departments often demand a single, visible leader. That leader must answer for both the technology and the business execution.

What to watch next

This structure raises several questions that will only be answered as Anthropic continues to scale:

The public market test: If Anthropic proceeds toward an IPO, will institutional investors accept a CEO who does not manage the executive team? Or will they demand a more traditional reporting structure?
The filter problem: How will the CEO work with other leaders? This is a key question if all communication must go through the chief of staff or the president.
Replication: Will other research-driven AI startups adopt a similar dual-leadership model? Doing so could protect their technical founders from administrative burnout.

This split-leadership experiment is a reminder of an important fact. As AI companies reach massive valuations, their challenges are not just technical. Success depends on organizational design and human management just as much as training runs and cluster architecture.

At find-my-ip.net, I am involved in every product decision, such as the rate-limiting strategy and whether to run a blacklist check before the port checker. This means I can iterate quickly and not worry about admin tasks. I have dedicated agents for this.

Sources: