Orgs in the Age of Agents

In his 1983 classic High Output Management, Andrew Grove (Intel co-founder and third CEO) explores how to optimize information flow and managerial leverage. He argues that excessive organizational layers distort information, and that maintaining an efficient span of control is essential to keep decision-making close to where knowledge resides.

Well, 43 years later, in the AI era, Mark Zuckerberg is building a personal AI “CEO agent” to flatten Meta’s organization — an approach very much in line with what Grove envisioned.

These AI agents should enable him to access information more rapidly, bypassing the multiple layers of employees he would typically have to navigate. Indeed, there is no substitute for hearing the voice of the customer directly.

This ties into a broader shift: Meta has signaled plans to cut up to 20% of its workforce. With more than 78,000 employees — many in coordination-heavy roles where impact is hard to quantify — agents change the equation. Work becomes traceable, comparable, and priced in tokens. Once output is measurable, roles that existed because managers couldn’t clearly evaluate them become exposed, and easier to eliminate.

The underlying vision is that everyone inside Meta will operate with their own AI agents. To accelerate adoption, Meta’s remuneration model is expected to increasingly reflect AI usage, making AI central to how work is assessed and rewarded.

Conclusion

In the age of AI, core management principles — such as span of control and organizational layers — remain intact.

But with AI agents, individual spans can expand into the hundreds, and large companies (in terms of revenue) can be run by just a few hundred people, naturally morphing traditional organizational layers.

As each worker commands an army of agents, much of the ambiguity in white-collar work disappears, and jobs stop being defended by narrative and start being complemented by output, directly tied to measurable, agent-level productivity.

And with agents coordinating across teams, the boundary between organization and software dissolves — and companies begin to behave less like org charts and more like coordinated compute.