Your Org Structure Is the Bottleneck, Not Your Tools
The machine gun in formation problem.
Someone told me recently that a principal engineer at Google admitted Claude Code replicated a year of her team’s work in one hour. I’ve been thinking about that ever since.
Not the “one hour” part. That tracks with what I experience every day. The “one year” part.
What was the team doing for a year? They weren’t coding for a year. The coding was probably a few weeks of actual implementation. The rest was committee debate. Architecture review boards. Cross-team alignment. Sprint planning. Stakeholder meetings. The organizational overhead that feels essential until you see what happens without it.
The pattern I keep seeing
I talk to a lot of engineering leaders who bought AI tools for their teams. The story is always the same.
They roll out Cursor or Copilot. Engineers start using it. Output goes up a little. PR volume increases. Then something weird happens: review queues back up. PRs sit longer. The overall pipeline doesn’t speed up the way anyone expected. Someone runs the numbers and it’s a 10-15% improvement. Disappointing. They start questioning whether AI is really as transformative as advertised.
What happened is obvious from the outside but invisible from within: they accelerated the smallest part of the pipeline. Coding is maybe 25-35% of the software delivery cycle. They made that part faster and left everything else untouched. The reviews, the approvals, the sprint ceremonies, the deployment processes, the manager check-ins. All still running at human speed.
When AI increases PR volume by almost 100% but your review process stays the same, review times go up, not down. You created more output and then choked on it. The tool exposed the bottleneck instead of removing it.
The machine gun in formation
I keep coming back to this image because it’s the most precise description of what enterprises are doing with AI tools.
Having a machine gun but still fighting in formation because “that’s how wars are fought.” The weapon changed. The tactics didn’t. The people with the weapon get slightly more effective. But the potential of the weapon is wasted on tactics designed for a different era.
Here’s the failed adoption pattern:
Company buys AI tools. Keeps the same team structure. Same management layers. Same sprint cadence. Same PR review requirements. Engineer builds something in two hours that used to take two days. Then waits three days for review. Then waits for the next sprint to deploy. Then waits for the stakeholder meeting to get feedback.
The two-hour build turns into a two-week delivery. The AI accelerated one step. The org slowed down the other nine.
Leadership looks at the numbers, sees 15% improvement, and concludes AI was overhyped.
What restructuring actually looks like
The companies getting transformational results didn’t just buy tools. They changed how the organization operates.
Shopify’s CEO made it a fundamental expectation. Not optional, not experimental. Teams must demonstrate AI cannot do a task before requesting headcount. AI proficiency became part of performance reviews. He described watching his people tackle “implausible tasks, ones we wouldn’t even have chosen to tackle before” and getting 100x the work done.
Klarna cut their headcount nearly in half. Revenue per employee went from $300K to over a million. They cancelled their Salesforce and Workday contracts and rebuilt internally. The CEO of Salesforce publicly asked how they were doing it.
These aren’t stories about better tools. They’re stories about different organizational structures enabled by better tools. Fewer layers. Fewer approvals. Fewer people doing coordination work that AI made unnecessary. The tool didn’t transform the org. The org transformed itself and the tool made it viable.
Why most companies can’t do this
The decision-makers who would need to approve the restructuring are the same people eliminated by it. Middle managers won’t recommend cutting middle management. Directors won’t propose flattening the org. VPs won’t suggest their layer is unnecessary.
Everyone will find reasons their layer is essential. And they’ll have data to back it up, because the data comes from the old system where their layer WAS essential. The proof that they’re no longer needed only appears after they’re gone, and they won’t volunteer for that experiment.
This is the innovator’s dilemma at its most personal. The change that would save the company requires the people running the company to vote themselves out.
What it looks like on the other side
We run our entire company on AI. Two people. Every system, every workflow, every process built in the past three months with AI-first methodology. A decision gets made and executed in the same hour because there’s nobody to ask permission from.
That’s not a flex about team size. It’s a structural observation. The work that needed 100 people now needs a fraction of that. The other 80 aren’t adding capacity. They’re adding coordination overhead. Meetings, alignment, consensus-building, process. A 100-person company isn’t just less efficient at AI adoption. It’s actively paying people to slow things down.
The companies that figure out how to restructure will operate at a fundamentally different cost and speed. The ones that buy tools and keep the old org chart will keep producing reports about how AI underdelivered.