The recent internal report from Anthropic highlights how agentic workflows are reshaping software teams: engineers report using Claude in 60% of their work and seeing a 50% productivity boost. That finding sits alongside academic field work showing a single person with AI tooling can match the output of a traditional small team, and together these signals are driving a rethink of team size and structure.
Why teams are shrinking — not because of layoffs, but because of leverage
Several organizations are reframing team design around what individual engineers can accomplish with AI. The shift is not simply headcount reduction; instead, it reflects higher per-engineer throughput. Engineers are taking on tasks that previously required coordination across multiple people because AI agents reduce coordination costs and accelerate routine and exploratory work. At the same time, some work that would have been too costly in person-hours is now feasible.
Evidence from research and industry
- A field study conducted at P&G by researchers from Harvard and Wharton found that individuals using AI performed as well as teams without it, while teams with AI outperformed those without. The study is available in the Cybernetic Teammate paper.
- Anthropic’s internal research reports engineers using Claude for a majority of tasks and that 27% of Claude-assisted work comprises tasks that wouldn’t have been done otherwise—efforts unlocked by lower coding cost and faster iteration.
Both findings point to the same mechanics: AI multiplies what an individual can execute, changing the optimal trade-offs between coordination overhead and individual output.
The “agent boss” and the human-agent ratio
The emerging role described by Microsoft’s WorkLab—often called an “agent boss”—captures the new mix of responsibilities: decomposing work for agents, reviewing agent outputs, orchestrating parallel agent streams, and preserving context agents lose between sessions. This turns parts of the engineering role into rapid-cycle orchestration and quality control.
Measuring the human-agent ratio—how much output is direct human effort versus agent-assisted—is becoming a practical management metric. The right balance will vary by project and domain; teams are still experimenting to find it.
Practical questions for engineering leaders
- Where are teams overstaffed under the new AI-enabled math?
- Is AI impact being measured, and can adoption be tied to output or new work performed?
- What valuable tasks become viable when production cost per line of code or analysis falls?
Answering these questions helps clarify whether consolidation is appropriate and which workflows benefit most from agentic augmentation.
Trade-offs and organizational implications
Smaller autonomous teams gain importance as individual contributors gain leverage through AI; coordination overhead for larger teams becomes a greater liability. Organizations designed around older assumptions about per-person output will need to reassess team boundaries, roles, and measurement systems to avoid being slower or overstaffed as competitors adopt agentic engineering.
Original source: https://blog.kilo.ai/p/1-pizza-teams
Related links:
- Anthropic report: https://www.anthropic.com/research/how-ai-is-transforming-work-at-anthropic
- Cybernetic Teammate (P&G study): https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5188231
- Microsoft WorkLab: https://www.microsoft.com/en-us/worklab/ai-at-work-how-human-agent-teams-will-reshape-your-workforce
- InsideAI analysis: https://insideainews.com/2024/04/24/artificial-intelligence-means-smaller-teams-doing-more-with-less-makes-the-small-autonomous-teams-structure-even-more-important/