Rewilding Software Engineering - AI, Tools and Human Decisions
summary
Claims: large language models extract about 60–80% of an architecture from code but miss 20–40% you cannot predict, so automated diagrams require recoding to verify. Example: a codebase of 1.5 million lines with 194 components and 396 dependencies yields a deterministic architecture tool that LLMs can invoke. An agent swarm builds a testing engine that turns out to be a simulation not real tests, so LLMs work well for hypotheses and small deterministic tool scaffolds but not for end-to-end trustworthy decisions.
AI-written summary · about
chapters (9)
- 0:00Rewilding software engineering
- 1:51Technological landscape
- 5:42Need for software engineers
- 11:47Adjacent unexplored
- 18:07Cloud computing and Red Queen
- 21:46Architecture in code
- 25:22Decision making as engineering
- 29:39Legacy systems and AI limits
- 33:13Key takeaways and future
related talks
21mAI's impact on software architecture
30mUsing AI to Create Better Requirements and Designs
45mThe AI Software Architect: Revolution or Illusion?
09mLessons Learned in AI Engineering
14mSAG Expert Interview #03: Impact of AI in Software Architecture
44mAutomate your software design How far we can get and how the AI chan
16mBuilding Applications with AI Agents
43mThe Evolution of Design Patterns in the Age of AI