No Priors Ep 106 | With GitHub CEO Thomas Dohmke

GitHub’s Project Padawan vision for 2025 has Copilot taking well-defined issues, planning solutions, and iteratively committing code via pull requests, behaving much more like a human teammate than a passive assistant.

Dohmke stresses that developers will only adopt agents widely if they reliably handle certain tasks, are easy to guide, produce outputs that can be inspected and rolled back, and genuinely save time instead of burning cycles.

Agents still struggle to decompose vague, high-level ideas into coherent plans without constant clarification, so developers’ roles will increasingly center on problem framing, architecture, and trade-off decisions rather than raw typing.

Beyond writing code, GitHub is investing in code review agents, security and dependency backlog reduction, and tooling to help specify, plan, and iterate on features—where much of real engineering time is actually spent.

Just as no single language or framework “won,” Dohmke expects a mosaic of models, agents, and tools; GitHub’s model catalog and extension system reflects a strategy of integrating many options rather than betting on one.

Projects like DeepSeek and FLUX show how open weights and open tooling spur rapid innovation and let students, hobbyists, and researchers experiment locally without relying solely on proprietary APIs.

At ~$20/user/month with measured productivity gains of 25%+ on end-to-end tasks, Copilot delivers strong ROI; Dohmke expects pricing to ultimately tie to compute and usage rather than to “rent-a-programmer” human-equivalent pricing.