Skip to content
Agentic Engineering

The 1-Pizza Team: Why AI Makes Smaller Engineering Teams More Effective

Amazon’s “two-pizza team” rule has been gospel for decades: if you need more than two pizzas to feed a team, the team is too big. But something is shifting. Directors at traditional companies are now talking about “one-pizza teams.” The math is changing.

This isn’t about layoffs or doing more with less in a grim, squeeze-the-workers sense. It’s about what individuals can accomplish when they’re managing AI agents alongside human collaborators.

The research backs this up

Section titled “The research backs this up”

A Harvard and Wharton study at P&G found something striking: individuals using AI performed as well as teams without it. And teams with AI significantly outperformed teams without AI on producing top-tier ideas.

Read that again. One person with AI tools matched the output of a traditional team.

Microsoft’s WorkLab research calls this the rise of the “agent boss”—everyone from interns to executives will manage their own constellation of AI agents. The hierarchy isn’t flattening; it’s extending into a new dimension where humans orchestrate machine intelligence.

What this looks like in practice

Section titled “What this looks like in practice”

Kilo’s engineering model is one example. Each engineer owns an entire product area and a WAU metric, not just a codebase. They manage teams of AI agents to parallelize work that would traditionally require multiple people.

One engineer. One product area. One number to own. AI agents handling the parallelizable work.

Anthropic’s internal research shows their engineers now use Claude in 60% of their work, reporting a 50% productivity boost—a 2-3x increase from the previous year. More telling: 27% of their Claude-assisted work is tasks that wouldn’t have been done otherwise. This isn’t just efficiency. It’s expanded capability.

The new mental model: engineers as agent managers

Section titled “The new mental model: engineers as agent managers”

The shift requires thinking about your team differently. It’s not just “how many engineers do I need?” but “what’s the optimal ratio of humans to agents for this work?”

Microsoft is already calling this the “human-agent ratio”—a new metric that will vary by task, process, and industry. Get it wrong, and you miss out on AI’s value or overwhelm your team. Get it right, and you unlock the performance demonstrated in that P&G study.

Your best engineers aren’t just coding anymore. They’re:

  • Decomposing work into agent-appropriate chunks
  • Reviewing agent output for quality and correctness
  • Orchestrating parallel workstreams across multiple agents
  • Making judgment calls agents can’t handle
  • Maintaining context that agents lose between sessions

These are management skills applied to AI systems. The job title stays “engineer,” but the work looks more like coordination.

What this means for team structure

Section titled “What this means for team structure”

Traditional team planning: “This project needs a frontend engineer, two backend engineers, a DevOps person, and a QA engineer. Five people.”

AI-native team planning: “This project needs two senior engineers who can each manage agent workstreams for their domain, plus one engineer focused on integration and quality. Three people, with explicit agent allocation.”

The InsideAI News analysis argues that AI actually makes small autonomous teams more necessary, not less. When individual contributors can have outsized impact through AI leverage, the overhead of large team coordination becomes even more costly.

The skill distribution shifts

Section titled “The skill distribution shifts”

Galileo’s research on AI team dynamics highlights how AI is blurring traditional role boundaries. Everyone shares responsibility for production outcomes, creating cross-functional teams that approach problems holistically.

But this creates a new challenge: engineers must now cultivate expertise in analyzing production data, ensuring system observability, and managing complete software lifecycles—skills that extend beyond writing code. As Charity Majors puts it: “Software engineering is not about writing code. It’s about solving business problems with technology.”

Engineers who can orchestrate AI effectively become force multipliers. Those who can’t risk being outpaced by smaller teams that can.

Practical steps for team leads

Section titled “Practical steps for team leads”

Audit your team structure. Where are you overstaffing because you’re not accounting for AI leverage? A team of 8 doing what 4 people with good AI workflows could handle isn’t sustainable when competitors figure this out.

Define agent allocation explicitly. Don’t let AI usage be ad-hoc. Identify which workstreams benefit from agent parallelization and resource them accordingly.

Measure the human-agent ratio. Start tracking it even informally. How much of your team’s output comes from direct human work vs. agent-assisted work? This will become a key metric.

Train for orchestration, not just coding. Your best engineers need to develop skills in prompt engineering, agent workflow design, and AI output validation. These are trainable skills with compounding returns.

Watch for capability expansion. Anthropic found 27% of AI-assisted work was new work that wouldn’t have happened otherwise. Are your teams using AI to do the same work faster, or to do work that was previously impossible? The latter is where competitive advantage lives.

The uncomfortable truth

Section titled “The uncomfortable truth”

Teams are getting smaller because they can. The organizations that recognize this early gain compounding advantages—they attract engineers who want leverage, they ship faster, and they compound learnings about AI-native workflows.

The question isn’t whether this transition is happening. It’s whether you’re leading it or reacting to it.

Resources

Section titled “Resources”

Essential reading

Section titled “Essential reading”

Deep dives

Section titled “Deep dives”