Future of Software Development
summary
Panel claims agents now write and review most code, turning human PR review into a cost/scale problem and pushing teams to agent-led PR review. Tools mentioned include Fire Hose (okthink) for orchestrating fleets, Apex/dedicated models for technical work, and local vs cloud tradeoffs under local vs cloud models. Surprising metrics: code output jumps from thousands to hundreds of thousands of lines per week and a single-line change yields a 90% endpoint speedup in one case. Junior apprenticeship gaps and governance in regulated sectors remain unresolved.
AI-written summary · about
chapters (24)
- 0:00Introduction and panelists
- 2:22Managing agents versus developers
- 4:05Individual productivity versus organizational output
- 5:50How client expectations are changing
- 8:00When models write and review the code
- 9:40Firehose and agent-led review
- 10:45Junior developers and the future of apprenticeship
- 18:10How new engineers learn from models
- 20:04Local, cloud, and specialized models
- 23:23Choosing models for different jobs
- 25:02Governance in regulated industries
- 27:22Code quality, craft, and AI slop
- 29:14The boundary between agent output and human writing
- 30:20Evaluating founders in a world of abundant output
- 35:40Practical advice for experienced engineers and leaders
- 40:25Predictions for the next six and 36 months
- 42:00OpenAI Dev Day
- 43:15Rapid-fire round
- 44:20Overrated AI advice and the phrase vibe coding
- 45:50Agent-built work that surprised the panel
- 46:50Unpopular opinions about agentic development
- 47:40One model and tool on a desert island
- 49:20Is product taste still a competitive advantage?
- 50:00Closing
related talks
20mSoftware Engineering Is Becoming Plan and Review
25mPanel The State of Web Development in 2026
19mIf AI Writes the Code, Who Runs Production?
50mPanel Software Engineering in the age of AI
33mThe Autonomous Enterprise
50mPanel: Beyond The Vibe: Code Quality First
26mYour pipeline is the bottleneck
01h 05mPanel: Agentic SE & Open Source in the FMware Era – Toward Trustworthy, Context-Aware FMware
