Not subscribed? Sign up to get it in your inbox every week.
PRESENTED BY OMNISEND.COM
Anyone can claim email expertise. You can prove it.
The Omnisend Product Certification is free technical training on the skills that make campaigns actually run: deliverability, segmentation, automation, and reporting. Not "how to do email marketing." You've got that. This is the platform expertise behind it, with a badge and certificate to show for it. And if you're switching platforms, Omnisend does the migration for you, free, in 5 days.
Use code THEBOTTLENECK for 30% off your first 3 months.
Your org chart is missing a line
Jane Street wants to think faster than everyone else.
Last week SemiAnalysis reported that the ultrafast Cerebras capacity at OpenAI is sold out, mostly to Jane Street and a handful of other customers paying a steep premium for the fastest inference money can buy.
Trading firms have done this before. They paid for fiber lines and microwave towers to see prices a few milliseconds before everyone else. Now they're paying for intelligence that arrives a few milliseconds sooner.
It's the most dramatic version of a race every company just entered. For most companies, though, the bigger drag is the cost of getting a lot of people to work together, and that cost is about to get repriced.
A firm is an orchestration machine
Strip a company down and it's a machine for coordinating two things at scale: people and capital. Ronald Coase made this argument in 1937. Firms exist because it's cheaper to coordinate work inside one than to negotiate every task on the open market. The org chart, the management layers, the planning cycle, and the weekly staff meeting are all overhead we accepted so that hundreds of people could pull in the same direction.
That basic job hasn't changed. What changed is that there's now a third line of execution next to labor and capital: agents that do real work, around the clock, at a marginal cost close to zero.
I run two of them for this newsletter. One sends five sponsor emails a day and reviews its own results every Saturday. The other reads the week's AI research for me. Two years ago, the first job would have been a part-time hire, and I would have spent hours a week managing that person. Today it's an agent I check in on.
The old ratios are breaking
Most of what we know about running a company came from a world where every unit of output needed a person attached to it. That world gave us a set of ratios so familiar we stopped seeing them as choices.
Span of control. A manager handles somewhere around six to ten direct reports, because that's how many humans one person can coach, unblock, and keep aligned. Nobody has a good number yet for how many agents one person can direct, and it's probably a lot more than eight.
Support layers. We staffed coordinators, analysts, ops roles, and assistants in rough proportion to the people doing the core work. A lot of those roles existed to move information between humans, and that's the job agents are best at.
Headcount as the unit of growth. Planning a bigger year meant planning a bigger team. The budget conversation started with "how many people do we need?" because people were the only way to add capacity.
Revenue per employee. It used to be a rough proxy for efficiency inside a sector. Now it's starting to measure something closer to how well a company has redrawn itself.
Each of these ratios carries assumptions about how work gets done. Swap in a new line of execution and the assumptions stop holding. A manager's job becomes less about keeping people aligned and more about writing clear specs, defining what good looks like, and reviewing output. I wrote in May that every AI agent needs a manager. The harder question is what that manager's week looks like when half their reports run on tokens.
The shape compounds
Here's the part I think most leadership teams are underestimating.
Every company is running this experiment at the same moment. Nobody has the answer yet, so the advantage goes to whoever finds a better shape first. But the advantage doesn't stay fixed, because a company built around agents also changes itself faster. Updating a workflow means editing a prompt and an eval rather than retraining a team, so each iteration is cheaper and quicker than the last. Therefore the company that finds a good shape this quarter will find a better one next quarter, while its competitor is still running a reorg.
That's a second-derivative advantage. A firm that moves faster wins a race, and a firm that gets faster at getting faster wins the decade.
It's also why this can't wait for next year's planning cycle. A gap in speed can be closed later, but a gap in the rate of improvement keeps widening every month you leave it alone.
The edge you can buy, and the one you can't
Jane Street can write a check for faster chips, and so can its competitors. That edge is real, but it's rentable, which means it gets competed away.
A shape is harder to copy. It lives in how a company assigns work between people and agents, who owns which agent, what gets reviewed and by whom, and how quickly the whole system rewires itself when something better ships.
Your Monday test
Pull up your org chart and redraw it with the line it's missing.
List the work, not the roles. For one team, write down the recurring jobs that actually get done each week.
Mark each job by who should do it. Is it human judgment, an agent with human review, or an agent alone?
Count what's left for the people. How many of the original roles still make sense as written?
Ask the ratio question. If this team grew output by 3x next year, would you add people, add agents, or change the shape?
Do it for one team this week. You'll learn more about your company's real shape than any reorg deck will tell you.
The firms that redraw first will be very hard to catch.
Till next time,
Chris
