If you're building a software company right now, the most important strategic question has a clear answer. Jamin Ball's "Get in the Token Path" names it: the Cloud era's biggest infrastructure winners monetized the core consumption primitive, compute. Snowflake on query compute. Datadog on telemetry. Cloudflare on requests. In the AI era, that primitive is increasingly looking like tokens. Get in the token path. He's right. If you haven't read it, start there.
But a second question sits behind it, equally important and almost entirely undiscussed: once you're in the path, how do you run the company?
That question runs through the work of Eliyahu Goldratt, physicist-turned-management-theorist and the mind behind the Theory of Constraints. His core insight: every system has a constraint at any point in time. Improving anything that isn't the constraint produces local gains, genuine ones, but doesn't lift the ceiling. Only addressing the constraint moves it. The constraint of the Cloud era was coordination cost. And AI is doing more than making companies faster. It's removing a load-bearing constraint. Per Goldratt, the system won't optimize. It will reorganize.
What the Constraint Actually Was
In the Cloud era, complex work required a forced sequence: align first, execute second. Before distributed teams could move, vision had to converge. Before engineers wrote code, product had to align with design. Before design mocked anything up, strategy had to align with leadership.
This wasn't bureaucracy. It was physics. Coordination cost compounded nonlinearly with team size. Each new person multiplied the alignment surface; each new team added another boundary to manage. The org chart was a routing system. Standups were synchronization mechanisms. Design systems, brand guidelines, approval chains: all load-bearing infrastructure for a real problem.
Smart organizations tried to fight it. Amazon's two-pizza team rule tried to cap it directly: keep the team small enough to feed with two pizzas, keep the coordination tax manageable. It worked, up to a point. The constraint didn't disappear. It just got bounded.
The ceiling wasn't talent or ambition. It was how fast you could align before you could act.
The Mechanism: How AI Attacks the Sequence
AI doesn't reduce coordination cost directly. It attacks the sequence, and that attack has a compounding ripple effect across the entire organization.
When you can produce a tangible artifact (a working demo, a pricing model, a repositioned landing page) faster than it takes to schedule the alignment meeting, the right first move changes. You don't align and then build. You build, and the thing you built does the aligning.
A 2025 arXiv paper, "AI as Coordination-Compressing Capital," formalizes this: as AI capital rises, coordination costs fall and the feasible set of organizational tasks expands. Activities previously too expensive to coordinate become viable. The consultancies will have a framework for this in about eighteen months, but the battle is being fought now. The founders who understand the mechanism are restructuring today.
The effect compounds across every role boundary. Design and engineering. Product and design. Strategy and execution. Each boundary had a coordination toll. Each is being renegotiated, not because someone decided to reorganize, but because the cost structure that justified the boundary has shifted. The cost of just trying has dropped below the cost of planning to try.
Those org boundaries were always working solutions, not laws of nature. We built them to solve a coordination problem that AI is now changing. A customer doesn't experience your org chart. They see one product, one company, one experience. They don't care how you've divided the work internally. When the coordination problem changes shape, the structures we built around it can too.
What Actually Reorganizes
Most conversation about AI-native operating models focuses on engineering: agentic stacks, harness design, how to structure a codebase when AI handles significant portions of implementation. These matter.
There is another dimension to this: what AI does to the value of different kinds of work.
AI is eradicating the execution layer: the tasks that coordination structures were built to manage. Writing first drafts. Maintaining visual consistency. Generating on-brand communications. These are becoming commodities.
What rises in value is the architecture layer. Building a custom agent that performs a voice review reliably, one that encodes what a brand means rather than just what a style guide says, and that surfaces that capability seamlessly to every person in the organization from their desktop or their coding environment, requires something rare. It requires deep understanding of the brand, the market, the customer, and how to design a system that can apply that understanding at scale. That isn't traditional software engineering. It's systems engineering with product instinct and organizational sensibility built in. The person who designs and builds that system, enabling everyone around them to work at a higher level, is worth more today than the person who executes the review.
In the Cloud era, you hired people to do the work. In the AI era, you architect systems that do the work. The scarce resource is people who can build, evolve, and orchestrate those systems with a deep understanding of the business, its goals, and how different functions connect. These are people at the intersection of genuine technical depth (code fluency was always a tool, never the skill itself), product thinking, and organizational design. That means a different hiring profile, a different onboarding process, and a different kind of company.
The Warning
The risk is taking AI's local gains, engineers shipping faster and content produced more efficiently, and calling it transformation. Those gains are real. But they are improvements in non-constraints. They don't lift the ceiling.
The harder question is whether the structures built around coordination cost were solving a real problem, or adapting to a constraint that's leaving. The org chart, the approval chain, the handoff ritual: each deserves that question individually. Some will survive it. Teams that find the question threatening are probably the ones most invested in the answer.
The System Is Searching
The collective confusion right now is real: how to hire, how to structure, what roles mean, what culture looks like when coordination assumptions shift. I hear it from founders, operators, investors. Nobody has a clean model yet.
That confusion is expected. It's exactly what Goldratt predicted. When a load-bearing constraint is removed, the system doesn't find its new equilibrium immediately. It searches. Old structures become mismatches. New patterns emerge before they're understood.
Getting in the token path is the right strategic bet. But it answers the where-to-play question. Understanding that the coordination constraint is being removed, and that removing it reorganizes the system rather than just accelerating it, is what tells you how to play.
The confusion won't resolve by waiting. It resolves by understanding what's happening, and building accordingly.
References: Jamin Ball, "Get in the Token Path," Clouded Judgement, March 6, 2026. "AI as Coordination-Compressing Capital," arXiv, 2025. Eliyahu M. Goldratt, The Goal, 1984.