Observations from Thoughtworks' Future of Software Engineering retreat on how agentic AI development changes software design priorities. Key debates covered whether low-level code quality still matters (conclusion: yes, because LLMs reinforce bad code, context windows are limited, and humans must debug what agents can't), whether specifications should replace code as the source of truth, and the role of domain models and bounded contexts in agentic workflows. The author argues for maintaining high test coverage, sound domain models, and code quality as a hedge against AI's risks and potential reversibility to human engineering. Rigour doesn't disappear with agents — it migrates upstream into specifications and test suites.
Table of contents
The retreat’s participantsLow level code qualityAchieving code qualityHigh level software designSpecifications or code?Placing your bets1.1K Impressions1 Comment