Upgrade to Pro — share decks privately, control downloads, hide ads and more …

The role of developer skills in today's AI-assi...

The role of developer skills in today's AI-assisted world

Deck for a 15 mins talk I gave at O'Reilly's "Coding with AI" online conference on May 8, 2025.

https://learning.oreilly.com/live-events/coding-with-ai-the-end-of-software-development-as-we-know-it/0642572171612/

Avatar for Birgitta Boeckeler

Birgitta Boeckeler

June 04, 2025
Tweet

More Decks by Birgitta Boeckeler

Other Decks in Technology

Transcript

  1. The reality We change more than we create from scratch

    Big and sprawling codebases Integrated systems Antiquated tech stacks Lack of fast feedback loops …
  2. Your AI team mate Now with super admin access! eager

    to help stubborn very well-read, but inexperienced won’t admit when they don’t know something
  3. Context of my usage ~8K JS ~8K Python ~3.6K CSS

    4 active maintainers, live for internal audience ~27K JS ~400 CSS ~500 HTML I’m the only maintainer and almost only user Codebases Agentic assistants Model Claude-Sonnet 3.5 (and 3.7) - Terminal command execution - Lint error detection - Web research - Custom rules - MCP servers
  4. We’re hitting an out of memory error, let’s increase the

    memory limit But why the memory error? We’re running npm install three times in our build process, and it includes dev dependencies
  5. Tests are a whole thing… Verbose or redundant tests Not

    enough tests Tests in the wrong places Too much mocking “Fixing” the code instead of the test
  6. 12 Impact radius of AI blunders Iteration Codebase lifetime Commit

    Friction for the team flow soon after commit Slower time to commit, instead of faster Negative impacts on long-term maintainability https://martinfowler.com/articles/exploring-gen-ai.html#memo-13
  7. Are AI-assisted codebases getting bigger faster? https://www.gitclear.com/ai_assistant_code_quality_2025_research Added Churn!! Moved!!

    2024 data ↑ Duplicate code ↑ Corrections within 2 weeks ↓ Refactoring Copy&Pasted
  8. Iteration Codebase lifetime Commit No working code Misdiagnosis of problems

    Too much up front work Brute force fixes instead of root cause analysis Misunderstood requirements Verbose or redundant tests Lack of reuse Overly complex implementations Complicating the developer workflow Negatively impacts long-term maintainability of the code Creates friction for the team flow soon after commit Slows down time to commit, instead of boosting it Antidotes Team alignment / knowing the requirements Simplicity Understanding what good feedback loops are Logical and analytical thinking https://martinfowler.com/articles/exploring-gen-ai.html#memo-13
  9. Working with AI agents today Think about your feedback loops

    Know when to quit Fight complacency and sunk cost fallacy Review, review, review Culture where both experimentation AND AI skepticism get rewarded Code Quality Monitoring Tools Small sessions, memory, custom instructions