The Org Chart Is the Bottleneck
AI didn't create the problem. It just made it impossible to ignore.
David George at a16z published a piece this week that every software CEO and board is forwarding to each other. The argument is clean: there are only two paths left. Grow revenue 10 points with genuinely new AI products, or rebuild to 40% true operating margins. No middle lane. Pick one and execute in 12 months.
He’s right about the binary. He’s describing the output of a deeper problem.
The reason most software companies can’t credibly execute either path isn’t the AI products. It isn’t the margins math. It’s the organizational architecture underneath both. And the evidence for this has been accumulating quietly while the industry debated which AI model to buy.
The governance gap is already measurable
Here’s a number that should stop any executive: 79% of leaders report productivity gains from AI. Only 29% can measure ROI with confidence. That gap isn’t a data problem. It’s a structural one.
When value is being generated somewhere in the system but the organization can’t see it, can’t route it, and can’t scale it, the bottleneck is the decision architecture. By the time legal clears a new AI workflow, compliance reviews the data handling, and IT approves the integration, the technology has evolved and the competitive window has closed. What Gartner calls a “committee death spiral” is just organizational design working exactly as intended, in an environment it was never designed for.
This is why 70 to 85% of AI projects fail at the scaling stage. Not at proof of concept. Pilots succeed in controlled environments, insulated from the complexity of the real org. Scaling fails because you have to move intelligence through a system built to manage human coordination overhead. The intelligence hits the org chart and stops.
Where George gets it right, and where the evidence diverges
George’s prescription for Path 1 is structurally sound in principle. Four-person pods. Design, product, and engineering collapsed into one unit. Decisions escalated and resolved in days, not quarters. Everything oriented around eliminating coordination overhead.
But Deloitte’s February 2026 research complicates this in a way the venture playbook hasn’t caught up to. Teams of 10 or more with genuine cognitive diversity were twice as likely to report improvements in both innovation and efficiency compared to teams of four or fewer. The small pod model works for greenfield product discovery, where the task is narrow and the context is controlled. It breaks down on the harder work: integrating AI into complex, multi-system, multi-stakeholder workflows. That work requires cognitive range. It requires people who understand the legal constraint, the data architecture, the customer relationship, and the technical implementation simultaneously.
George’s four-person pod is the right structure for building a new product. It’s the wrong structure for redesigning how an existing organization moves work through its systems. Most companies need the second thing more urgently than the first.
The layoffs are flattening the wrong layer, slowly
The restructuring wave of 2024 and 2025 is real. Over 1.1 million tech roles eliminated. Middle managers now comprising 32% of layoffs, compared to 20% before this cycle. The average supervisor span of control has doubled, from three direct reports to six.
This looks like progress. It is, in the narrowest sense. Amazon explicitly targeted manager roles to increase the ratio of individual contributors by 15%. The direction is correct.
But the pace is organizational cosmetics. You can double the span of control and still have a decision architecture that routes every meaningful choice through four layers of approval before anything ships. Flatter isn’t faster if the remaining structure still optimizes for consensus over velocity.
The LeadDev data makes the human cost visible: 65% of developers now report expanded responsibilities, and team motivation is down 40%. You cut the coordination layers without rebuilding the decision pathways. The work still needs to move somewhere. It moves onto the people who stayed.
The pricing shift is the signal most leaders are still misreading
By 2025, 85% of SaaS leaders have adopted or are actively transitioning away from pure seat-based pricing toward usage or hybrid models. Model API spending moved from $3.5 billion to $8.4 billion in a single year. High-growth companies on hybrid pricing models are showing 21% higher median growth than those holding the seat-based line.
This is not a pricing story. It is an organizational intelligence story.
Seat-based pricing existed because the unit of value was a human doing a task. One seat, one user, one workflow. When 41% of all code written globally is now AI-generated, the unit of value has fundamentally shifted. It is no longer the person clicking the button. It is the output the system produces regardless of who, or what, initiated it.
The org chart is still organized around the former. The budget is visibly moving toward the latter. The companies that feel this most acutely are the ones trying to manage machine output with structures designed to manage human activity.
The actual question
George asks boards to put one question on page one of every deck: which path are we on?
There is a prior question. One that has to be answered before path selection means anything.
Is your organizational architecture capable of moving intelligence through the system at the speed AI requires?
Not AI in a pilot. Not AI in a controlled demo. AI as the actual operating layer of how decisions get made, how work moves, how value gets measured, and how the org responds when the technology changes again in six months.
The companies executing Path 1 or Path 2 successfully aren’t choosing between growth and margins. They are the ones that already solved the structural problem that makes either path executable. Everyone else is selecting a destination without asking whether the vehicle can make the trip.
That is the work that does not appear in a16z essays. It is unglamorous. It does not have a refounding moment. It is the patient, methodical redesign of how an organization moves intelligence through its own structure.
And until that work is done, the path you pick doesn’t much matter.
P.S. The pricing shift is the leading indicator. When your customers stop counting seats and start counting outcomes, your org chart has already fallen behind the market. The question is whether you recognize it before they tell you directly.