My fireside chat about agentic engineering at the Pragmatic Summit
a day ago
- #Agentic Engineering
- #AI Programming
- #Software Development
- Stages of AI adoption in programming: from using ChatGPT to coding agents writing more code than the developer.
- Trusting AI output: comparing AI to professional teams and gaining confidence in AI like Opus 4.5 for familiar tasks.
- Test-driven development with agents: emphasizing the importance of tests and how agents can efficiently handle TDD.
- Manual testing and Showboat: using tools to document manual tests and ensure functionality beyond automated tests.
- Conformance-driven development: leveraging AI to build test suites based on multiple implementations for standardization.
- Code quality: context-dependent importance, with agents capable of producing high-quality code if guided properly.
- Codebase patterns and templates: maintaining consistency and quality in codebases to guide AI agents effectively.
- Prompt injection and the lethal trifecta: risks of outsourcing decisions to gullible language models with access to sensitive data.
- Sandboxing: running coding agents in secure environments to limit potential damage from malicious instructions.
- Safe testing with user data: avoiding sensitive data in tests and using mocking for edge cases.
- Evolution of AI models: key inflection points like GPT-4 and Claude Code leading to reliable AI coding agents.
- Exploring model boundaries: continuously testing new models to discover their capabilities and limitations.
- Mental exhaustion and career advice: managing multiple projects with AI agents and expanding skills ambitiously.
- Impact on open source: challenges and opportunities for open source projects in the era of AI-assisted programming.