If I (only) Had A Hammer
summary
Copilot trains on public GitHub and often suggests code verbatim, raising copyright and supply‑chain risks where adversaries could poison popular repos; it helps spot SQL issues and recommends indexes in Azure but also hallucinates incorrect logic. Legacy tools reshaped practices: ORMs like NHibernate/EF encourage expensive full object load patterns, ReSharper drives proliferation of interfaces, and mocking frameworks can produce tests with only mocks. Empirical reports show roughly 10–30% productivity gain for small, well‑scoped tasks, but rigorous review and mentoring remain essential.
AI-written summary · about
chapters (20)
- 0:00Introduction
- 2:14AI tools limitations
- 5:21History of software tooling
- 8:54Code quality over speed
- 12:30Conway's law tool impact
- 13:35Mocking frameworks pitfalls
- 16:04Resharper IDE evolution
- 20:39ORMs and SQL decline
- 23:30GitHub Copilot SQL AI
- 25:55Copilot vs Stack Overflow
- 26:37LLMs code generation
- 28:04Copilot contextual limits
- 31:33Copilot security concerns
- 32:34AI coding productivity
- 36:22Copilot self-criticism
- 38:06AI software engineers future
- 41:16Slower coding better software
- 44:04Mentoring next developers
- 47:18Developer time and AI
- 48:23Q&A
related talks
39mPower of AI and Code Quality and Performance
34mNavigating AI assistant development
47mCode Is Cheap. Software Isn’t.
49mCopilot, Cursor, and Custom LLMs: Navigating the New .NET Developer Experience
42mRevisit Old Problems with New Eyes!
15mThe AI-Augmented Engineer [...]
45mAI Assistance for Software Teams: The State of Play
31mModern Architecture for AI-Augmented Development Teams