Product Engineers Over PMs: The AI Reshaping Who Builds Software
Factory's AI coding agent Droid represents a new category of tool—one that changes not just how engineers work, but who does the building.
The shift: from product managers and engineers in separate roles to "product engineers" who direct AI agents to do the work.
What Droid Does
Droid isn't just another coding assistant. It's an enterprise-first AI agent with:
- ROI analytics that prove value
- Multi-surface integrations across workflows
- Rigorous self-validation (linters, tests, screenshots)
The demo: building a complete app from meeting notes. Not prototype quality—production-ready code.
The Two Modes
Factory introduces a useful distinction:
Spec mode: What to build. Defining requirements, features, and outcomes.
Plan mode: How to build. Technical implementation, architecture, code structure.
AI handles plan mode effectively. Humans focus on spec mode—deciding what should get built.
Model-Agnostic Strategy
One practical insight: mixing models strategically.
Use Opus as the planner—strong reasoning for high-level direction. Use GPT-5.2 as the executor—efficient at implementation.
This combination can outperform using a single model for everything.
The Organizational Shift
Here's where it gets interesting: hiring implications.
Companies adopting this approach need fewer traditional roles:
- Product engineers who understand both product and technical direction
- AI-savvy engineers who can direct and validate agent work
Traditional product managers, separated from implementation, become less essential.
The Self-Validation Key
What makes this work: rigorous quality control.
AI agents produce code. But code isn't useful unless it works. Droid validates through:
- Linters checking code quality
- Tests ensuring functionality
- Screenshots verifying UI
Without this validation layer, AI-produced code accumulates technical debt invisible until problems emerge.
The Takeaway
AI coding agents are evolving beyond autocomplete. They're becoming workforce participants that reshape organizational structure.
The winners will be companies that adapt—training product engineers, building validation pipelines, and restructuring teams around human-AI collaboration.
The old model: PMs spec, engineers build. The new model: product engineers direct AI to build, validate, and iterate.
Stay ahead of AI trends. tldl summarizes podcasts from builders and investors in the AI space.