What I believe about the future of software development

Thorsten Ball argues that model-generated software will change what matters in development. Code review and unit tests will matter less, code-writing as a craft will fade, and the harder problems will be choosing the right thing to build, shipping it, and getting feedback. He also says many current assumptions about code quality, open source, and team structure will lose force as models get faster and cheaper.

Reading notes

#

• Code review will stop being a useful way for humans to inspect model-written code line by line; review will move to the system and its composition.
• Unit tests may matter less if models can produce large amounts of correct code without errors.
• Writing code as a craft will disappear, while building software as problem solving will matter more.
• More bugs will come from asking for the wrong thing than from code mistakes.
• Open source in its current form may make less sense when artificial eyeballs can inspect code.
• Human performance-critical optimization will remain an edge case in most software.
• The terminal, shells, text editors, and CLI tooling will be replaced by token-based workflows.
• Tokens are presented as the new computing paradigm, with software rebuilt on top of them.
• The PM/Design/Eng triad, along with Agile and Scrum, is expected to fade.
• The gap between large-company and small-company software work will widen.
• The idea of “good code” is questioned because humans will modify less of the codebase.
• Some people will be priced out of software production if they cannot get enough tokens.
• Cheaper models may be less attractive if smarter models reduce mistakes and interaction turns.
• Faster models will enable UI to be generated on demand, reducing the need for many menus and dashboards.
• The transition will take years, and people will still be employed writing code during the change.