Spencer Saldana
← all writing

Automation Mayhem: An AI Revolution Update

June 22, 2025·Spencer Saldana

Price's Law, now on AI steroids

In any large team, a few people carry most of the load. Derek de Solla Price made this rigorous in 1963: roughly the square root of the number of contributors produces half the output. In a project of 100, around 10 people generate 50% of the result. In a project of 10,000, it's 100. The distribution is heavy-tailed and nobody's that surprised.

What's new is that the tail is getting heavier. The top performers in any field are now armed with tools that are themselves quietly superintelligent at the parts of their job that used to take half their week. The vital few were already vital. Give them Cursor, Claude, and an agent layer and they pull further ahead.

This isn't theoretical. AI startups are reaching $100M in annual revenue with under 100 employees. Lovable hit $17M ARR with 15 people, which works out to over a million dollars of revenue per head. Midjourney is reportedly somewhere north of $4M per employee. OpenAI's number is in the rough neighborhood of $1.5M. The historical baseline for a software company was tens of thousands. The new baseline is millions, and it's still moving.

A single sharp engineer plus a coherent product idea plus a real model can replace what used to require a department. The first one-person billion-dollar company is plausibly someone reading this. The implication for inequality inside a company is severe. A small number of AI-empowered people might contribute the bulk of value, leaving everyone else to find a new role or accept they're decorative.

The same dynamic plays out at the firm level. A few giant labs dominate the model layer. The companies with the most data, the most compute, and the strongest distribution have a flywheel that gets better with use. The digital era already concentrated wealth and productivity. AI is the same gradient, steeper.

Outsourcing to algorithms

Companies have been offshoring work for decades. The next move is not offshore. It's off-org. You don't hire the entry-level analyst, you point an agent at the data and it returns a report by lunch. The marginal cost of an additional "employee" is the cost of a query.

This is putting downward pressure on entry-level hiring in ways that show up in the data and even more in the anecdotes. Why hire a junior to draft the email when a model writes it in three seconds. Why hire the BPO seat when the chatbot handles the tier-one ticket without complaint. Why hire the analyst when the spreadsheet runs itself.

The result is a generation of leaner teams. Companies can stay small because most of the heavy lifting is no longer human. The Lean Startup turned out to be the warm-up act. The new ratio is whatever the senior team plus their AI tooling can sustain.

There's a related dynamic worth flagging. Robin Dunbar's research suggests we can hold maybe 150 stable relationships in our heads. Most large companies are larger than 150 by an order of magnitude, and they paid for that with bureaucracy. Hierarchies, processes, middle management. If AI lets you do big things with small teams, you no longer have to pay that tax. You can stay small and coherent. The Gore-Tex move (cap a plant at 150 people to preserve culture) was a strange exception in 1985. It's the obvious move in 2026.

The monkeysphere at work

Dunbar's number is the monkeysphere. The size of the group beyond which colleagues become "that random guy from accounting" rather than individuals. Human brains evolved to trust and coordinate in groups up to this scale. Beyond it, informal trust gives way to formal rules. Bureaucracy is what a primate species does when it grows past the limits of its own social cognition.

AI lets us route around this. Instead of one monolithic 5,000-person company, you can run a network of tight teams loosely coupled through software. Think Hollywood model. Small expert crews assemble for a project, ship, dissolve, reform. AI handles the coordination, the documentation, the institutional memory. The humans handle the relationships, the taste, the judgment calls.

Smaller teams under the Dunbar threshold are more agile and more cohesive. Above the threshold, communication silos form. Engagement drops. People stop knowing each other's names. The number isn't magic. It's a rough ceiling on how many people can be in your social model at once.

A "team of 5" with the right AI tooling can ship what used to take a 50-person org. A two-pizza team becomes a one-pizza team with a printer for second pizzas. The humans strategize and build relationships. The agents handle the support tickets and the data work and the first draft of every document.

The structural implication for large enterprises is that they should re-architect into many small semi-independent teams sharing common AI infrastructure. Internally it should feel less like a top-down org chart and more like a swarm of startups operating on shared rails. People are not wired to meaningfully connect with hundreds of coworkers. Keep the tribe small and let the machines bridge the gaps between tribes.

Automation at every level

When we talk about AI taking jobs, the image is usually a factory floor or a customer support seat. The real distribution is more interesting. Pressure is showing up at both ends of the org chart, often more aggressively at the top than people admit.

Entry-level and routine work is the easy story. AI can draft emails, schedule meetings, crunch numbers, write code, design basic graphics. Roles like administrative assistants, junior analysts, customer service reps, and copy editors are directly in the path. The McKinsey projection of 30% of US jobs fully automatable by 2030 is probably wrong on the exact number and roughly right on the direction. Many traditional junior roles will be eliminated or reshaped past recognition.

Senior leadership is being squeezed differently. AI is now genuinely useful for the work executives historically owned. Synthesizing large amounts of context. Running scenario analysis. Surfacing the option you didn't think of. There's no robo-CEO on the horizon, but there is an executive who is now competing with a tireless analyst that costs eight cents per query and never sleeps.

The role most directly under threat is middle management. AI is excellent at the things middle managers actually do all day. Project tracking. Status reports. Reviewing expense reports. Aggregating updates. The orgs that figure out how to flatten will do so faster than the ones that don't, and the ones that don't will lose to the ones that did. This is uncomfortable and also obviously true.

What's not happening, despite a lot of noise to the contrary, is that AI is replacing human leadership. The soft skills, the moral judgment, the political navigation, the relationship work, the personnel calls, the ability to read a room and absorb risk and apologize convincingly when needed. These remain stubbornly human. The future of management is AI-augmented, not AI-replaced. The good managers will figure this out and use the tooling. The bad managers will pretend they don't need it.

So what

The shape of the company is changing. Some will be very small. Some will be vast networks of small things. Most won't last long enough to find out which they were. The labor market will sort and re-sort. The colleague in the next chair will increasingly mean either a teammate or a model. The most successful organizations will figure out how to combine the brilliant 10%, the cohesive 150-person tribe, the tireless agents, the accountable humans, and the platforms that hold all of it together.

If you're in the workforce now, the move is to bring your brain, your taste, your relationships, and your AI. You'll need all four. The companies that don't are running their primate operating system in a world that's running something else.

Sources

← more writingchat about it →