How we moved twice as fast with half the people
Gregory's Substack
SubscribeSign in
How we moved twice as fast with half the people<br>AI requires a new way to organize teams
Gregory Lee<br>Jun 04, 2026
Share
Subscribe
In January, we had a choice, let half our team go, or perish. We expected to slow down. Instead our resource constraints forced us into a new way of work. We embraced how AI is changing organizational structure, and our speed doubled . In short, code used to be the bottleneck, but now the bottleneck is in human communication.
First, there’s a lot of fear right now about AI replacing jobs, which needs to be addressed. My feeling is: clearly the world has changed, and we are living in a new reality. Desiring a different world doesn’t change where we are. Most white collar work will look different in 5 years, some jobs will cease to exist, and new ones will arise. I’m hugely optimistic about the direction. I think AI will replace the mundane parts of our jobs, the parts that no one wanted to do anyway, and leave us the fun parts. No one has aspired to be a data entry specialist.<br>1. Code was the bottleneck
In the old SaaS world, code was the bottleneck. Code was difficult, time-consuming, and expensive to write and maintain. Thus, organizations had to build up process, bureaucracy, and management around writing code. Specs, handoffs, planning documents, detailed tickets, translation layers between domain experts and engineers, all of that made sense in a world where implementation was costly. If code is expensive, it’s most efficient to do a lot of planning before you write it.<br>Now, code is cheap. This shift has far-reaching consequences. It changes how teams should organize. It changes who can contribute, and how they should contribute. And, much of the organizational overhead no longer makes sense.<br>At GTM Engine we communicate via code. We don’t write tech specs at all anymore. We will lightly discuss and outline new big features to get the direction right. Then we just write a draft in code. It’s actually faster and more accurate to build it and review, rather than do a tech spec. If we need to do a full rebuild, that’s fine, because again, code is cheap. A tech spec is just a plan, and execution of that plan can vary. The code is ground truth, so we can review the actual thing and skip the days or weeks of planning discussions.<br>2. The New Bottleneck
What hasn’t gotten cheaper is human communication. Humans still take the same amount of time to think, understand, align, disagree, clarify, and decide. We still need context. We still misinterpret each other. We still have to translate ideas between people with different knowledge, incentives, and vocabularies.<br>When we let people go, we noticed that we had been spending more time explaining than it would have taken to just do the thing. Layer on top of that, team planning meetings, 1:1s, different schedules, and you have a really high bar for what kind of task can get handed off efficiently.<br>3. Lanes and Ownership
So, how did we restructure for this new reality?<br>For us, the answer has been lanes and ownership. Every time work gets handed off, you incur delay, misunderstanding, and context loss. In a world where code was expensive, maybe that cost was acceptable. Nowadays, that communication overhead can become the dominant cost.<br>Thus, each person should be able to drive something end to end as much as possible. Every employee should have a clear lane that they own end to end. Gone are the days were someone contributes a piece, then waits for three other people to do their parts. People now need to be able to take a whole problem, understand it, make decisions, and push it to the finish line.<br>This does not mean working in isolation. People can still pull in others when needed. There is still review, consultation, and collaboration. But the ownership and execution should stay with a single person.<br>We can really live the ideal from the Netflix culture: the person closest to the decision should make the decision. We can enormously empower our employees to do what they believe is right, and use a trust and verify model. This change has drastically improved our productivity and our employee satisfaction.<br>4. Domain Experts Now Have Leverage
A second consequence of cheap code is that AI gives domain experts much more ability to build the things they envision. They no longer have to fully translate every idea through an engineering bottleneck before they can see whether it works. They can go from customer conversation, to product concept, to prototype, to first implementation themselves.<br>That is a huge change.<br>At GTM Engine, we’ve seen this very clearly. Rob is the domain expert. He has built the GTM teams at 5 unicorns, and has lived the problems that we are solving. He also has enough engineering background to understand the shape of software problems. With AI, he can talk to a customer, recognize a need, decide what should exist, and build a...