# Smaller teams first

Author: Daniel Netzer
Published: 2026-10-16
Canonical: https://www.danielnetzer.com/build-log/smaller-teams-first

When AI coding tools arrived, the constraint was the team leads. Shrink teams so leads can learn hands-on, grow again, then remove layers. Velocity rose more than 45%.

When AI coding tools arrived, most of the conversation was about which tools to buy. In my experience the constraint was the team leads.

## The gap

Most team leads are excellent engineers who got promoted. By the time AI became part of daily work, many were managing 8, 10, or 12 people. Their days were full of reviews, planning, and people work, with very little room left to learn a new way of building software: prompting, working with agents, and changing how code gets written and reviewed.

A team rarely adopts a new way of working faster than its lead does.

## What I changed

I made two moves at the same time.

1. A bootcamp, so everyone started from the same baseline on AI-assisted engineering.
2. Teams cut to 4 or 5 engineers, with every team lead spending 60 to 70% of their time hands-on.

Smaller teams gave the leads time back. Hands-on time meant they learned the new tools by shipping with them.

## After a year

After almost a year the team leads were proficient. They knew how to work with agents, how to prompt well, and how to review that kind of work. That is when the teams grew again, because a lead who understands the new way of working can support more people doing it.

Then I went one step further and removed layers of hierarchy. Fewer handoffs between levels meant the business moved faster, and decisions were made closer to the people who understood the work.

Velocity went up by more than 45%, and AI adoption peaked at 98% across the department.

## What I'd tell other leaders

- Team size isn't fixed. Shrink it on purpose while people learn, and grow it again once they can lead the new way.
- Your team leads set the ceiling for adoption, so invest in them first.
- Hands-on time for leads is the training program.
- Once the skills are there, look at your hierarchy. With AI, extra layers cost more than they used to.

Learn first, then grow, then flatten.

## Timeline: how the org changed (static)

| Stage | Team size per lead | Team lead's time | Hierarchy | Results |
| --- | --- | --- | --- | --- |
| 1. AI arrives | 8, 10, or 12 people | Full with reviews, planning, and people work; very little room to learn | Existing layers | — |
| 2. Bootcamp and smaller teams | 4 or 5 engineers | 60 to 70% hands-on | Existing layers | Everyone starts from the same baseline |
| 3. After almost a year | Teams grow again (no figure given) | Not stated | Existing layers | Team leads proficient with agents, prompting, and reviewing that work |
| 4. Fewer layers | Not stated | Not stated | Layers of hierarchy removed | Velocity up more than 45%; AI adoption peaked at 98% across the department |