David Foster
davidfstr@mastodon.world
<p>Software sculptor. Writer. Teacher. Inquiring mind. Bringing world-class computer science education to K-12 schools, teachers, and students at <a href="https://techsmart.codes" target="_blank" rel="nofollow noopener" translate="no"><span class="invisible">https://</span><span class="">techsmart.codes</span><span class="invisible"></span></a></p>
Posts
-
Post #2206278
If you&#39;re spending more than $100/month/engineer on AI tokens, you&#39;re doing it wrong. Use a subscription. I currently subscribe to Claude Pro ($20/mo) and GitHub Copilot Pro ($20/mo), totalling $40/month. https://newsletter.pragmaticengineer.com/p/the-pulse-token-spend-breaks-budgets #aicoding
-
Post #1963908
I created a Revise Skill (for Claude Code, Codex, VS Code, etc) that helps AI coding agents write code the way I want, optimized for readability &amp; maintainability: https://dafoster.net/projects/revise-skill/ #aicoding #refactoring
-
Post #1963907
gvc (git-visual-compare) is a new visual git diff tool that I&#39;m releasing today! Try it out at: https://dafoster.net/projects/gvc/ #coding #programming #diff #git I&#39;ve been using the diff mode of the venerable yet unmaintained GitX tool for a decade. But now I need my diffs to support dark mode, to avoid waking the baby at night. Astonishingly, existing GUI diff tools only support diffing *individual* commits, not ranges . So I finally bit the bullet and wrote my own. Enjoy!
-
Post #1963906
I wrote up how I&#39;ve been attributing usage of AI models &amp; coding agent harnesses when drafting code commits: https://dafoster.net/articles/2026/04/17/ai-attribution-in-git/ #aicoding #git
-
Post #1963905
RE: https://mastodon.gamedev.place/@eniko/116426136253346140 Even with today&#39;s most advanced AI coding tools/models I consistently find their output - although functional - to lack in readability/ maintainability/ elegance. I always have to revise AI code to meet my standards. The current trend of some prominent folks having no human review at all for AI code I think is a shame and sometimes even harmful. Consider: Security bugs in AI drafted/unreviewed code. #aicoding
-
Post #1963904
Oops, I wrote a database. I discuss some of challenges I faced getting full ACID guarantees working in my latest post: https://dafoster.net/articles/2026/02/23/oops-i-wrote-a-database/ #databases #warstories #programming
-
Post #1963903
I actually like Claude Code’s 5-hour session limits on usage. It enforces that I take a break occasionally and trains me to prompt more cost-efficiently, which will be useful whenever token prices stop being subsidized. #aicoding
-
Post #1963902
4 GB of space remaining. Restart computer. 22 GB of space remaining. -- Wild to see such a large fluctuation. I still remember when my total drive capacity was 0.5 GB (500 MB)...
-
Post #1963899
GitHub Copilot Agent (GCA) is moving from per-request/interaction billing to token-usage based. Shucks. I&#39;ve been enjoying the ability to give 1 well-specified request to GCA and have it churn for 15 minutes, while only charging me 1 &quot;premium request&quot;. https://github.blog/news-insights/company-news/github-copilot-is-moving-to-usage-based-billing/ #aicoding