Why AI Velocity Is Becoming a Debt Accelerator

6 days ago

The Thoughtworks Future of Software Development Retreat discussed AI's impact on software development, noting that current practices are breaking under AI-assisted work.
Key themes included the 'supervisory engineering middle loop', 'risk tiering', and 'TDD as the strongest form of prompt engineering'.
Participants found comfort in shared uncertainty about AI's role in productivity, team restructuring, and role shifts.
AI is seen as an accelerator, amplifying existing practices, with concerns about it becoming a 'debt accelerator' without best practices.
LLMs may reduce the need for specialist developers, raising questions about the future role of 'Expert Generalists'.
Cost of LLM tokens is uncertain, with potential implications for how freely they can be used in the future.
There's debate on whether LLMs will encourage waterfall-style development or enhance evolutionary design.
Security concerns were noted, with a call for platform teams to create safe paths for AI use in application building.
The retreat highlighted the value of open, respectful dialogue and the importance of inclusivity in tech discussions.
Stephen O'Grady and Grady Booch view AI as another step in technology's evolution, emphasizing calm adaptation.
Research shows LLMs perform better in healthy codebases, with higher defect risks in less-maintainable code.
TDD is reported as essential for effective LLM usage, providing clear tests and cycles for driving LLMs.

Hasty Briefsbeta