๐ ๐๐ ๐๐ฑ๐ฉ๐๐ซ๐ญ๐ฌ ๐ฐ๐๐ง๐ญ ๐ ๐ง๐๐ฐ ๐ง๐๐ฆ๐ ๐๐จ๐ซ ๐ฐ๐ก๐๐ญ ๐ฐ๐ ๐๐จ. ๐'๐ฏ๐ ๐ ๐จ๐ญ ๐จ๐ง๐. ๐ ๐๐ฆ ๐๐๐ค @oreilly.bsky.social ๐ ๐๐๐๐ ๐๐๐๐๐ ๐ก๐๐๐๐ ๐๐ ๐กโ๐ โ๐๐๐๐๐ ๐ก ๐๐๐๐๐๐๐ ๐๐ ๐๐๐๐๐ข๐ก๐๐ ๐ ๐๐๐๐๐๐: ๐๐๐๐๐๐ ๐กโ๐๐๐๐ Boris Cherny, who runs Claude Code, told Business Insider that the phrase "vibe coding" had started to annoy himโฆ
Andrew Stellman ๐พ
@andrewstellman.bsky.social
Author, developer, team lead, musician. Author of O'Reilly books including Head First C#, Learning Agile, and Head First PMP. Solving complexity with simplicity.
๐ ๐๐ซ๐๐๐ญ ๐ฒ๐จ๐ฎ๐ซ ๐๐ ๐๐ ๐๐ง๐ญ ๐ฅ๐ข๐ค๐ ๐ ๐ฉ๐ข๐ฉ๐, ๐ง๐จ๐ญ ๐ ๐๐๐ญ๐๐๐๐ฌ๐ ๐ ๐ ๐ข๐ ๐๐๐๐ฆ ๐ โ๐๐๐ก ๐ ๐๐ ๐ ๐๐๐๐ ๐กโ๐๐ก ๐๐๐๐ ๐๐๐๐ โ ๐๐๐๐ ๐๐๐ ๐, ๐๐๐ก ๐๐๐ ๐๐๐๐ ๐๐๐ ๐กโ๐๐ก โ๐๐ ๐ก๐ ๐๐๐๐๐๐๐๐ The technique from my latest @OReilly.bsky.social Radar article that I lean on most is also one of the simplest:
๐ ๐๐ก๐ข๐ฅ๐๐๐๐ฅ๐ฉ๐ก๐ข๐ ๐ก๐๐ฌ ๐๐๐๐ง ๐ญ๐๐๐๐ก๐ข๐ง๐ ๐ฉ๐ซ๐จ๐ฆ๐ฉ๐ญ ๐๐ง๐ ๐ข๐ง๐๐๐ซ๐ข๐ง๐ ๐๐จ๐ซ ๐๐๐๐๐๐๐ฌ ๐ I'm teaching my @OReilly.bsky.social course on AI-assisted coding from Philadelphia this week to celebrate America's big 250th birthday, and the prompt engineering curriculum here is a lot older than the tools.
๐ ๐๐ฆ๐๐ซ๐ข๐๐ ๐ญ๐ฎ๐ซ๐ง๐ฌ ๐๐๐ ๐ญ๐ก๐ข๐ฌ ๐ฐ๐๐๐ค, ๐ฌ๐จ ๐'๐ฆ ๐๐ซ๐ข๐ง๐ ๐ข๐ง๐ ๐ฆ๐ฒ ๐๐จ๐ฎ๐ซ๐ฌ๐ ๐ญ๐จ ๐๐ก๐ข๐ฅ๐๐๐๐ฅ๐ฉ๐ก๐ข๐ ๐ It's America's 250th birthday, and Philadelphia is where it all started! So to celebrate, I'm bringing my next @OReilly.bsky.social live training course to Philly!
๐ ๐๐จ๐ฎ๐ซ ๐๐ ๐๐๐ง ๐ก๐๐ฏ๐ ๐ญ๐ก๐ ๐๐ง๐ฌ๐ฐ๐๐ซ ๐ฌ๐ข๐ญ๐ญ๐ข๐ง๐ ๐ข๐ง ๐ข๐ญ๐ฌ ๐๐จ๐ง๐ญ๐๐ฑ๐ญ ๐๐ง๐ ๐ฌ๐ญ๐ข๐ฅ๐ฅ ๐ฆ๐ข๐ฌ๐ฌ ๐ข๐ญ ๐ ๐โ๐ ๐๐๐ข๐๐กโ ๐๐๐ก๐๐๐๐ ๐๐ ๐๐ฆ ๐๐๐๐ก๐๐ฅ๐ก ๐๐๐๐๐๐๐๐๐๐ก ๐ก๐๐๐๐๐๐ฆ Sometimes I wonder if my AI is as smart as I think it is. I've definitely had some experiences lately that made me question that.
๐ ๐๐จ๐ฎ๐ซ ๐๐ ๐๐ ๐๐ง๐ญ ๐๐จ๐๐ฌ๐ง'๐ญ ๐ค๐ง๐จ๐ฐ ๐ฐ๐ก๐๐ญ ๐ข๐ญ ๐๐จ๐ซ๐ ๐จ๐ญ ๐ ๐ด ๐๐๐ก๐๐๐๐๐๐๐ ๐ก๐๐ ๐โ๐๐๐ ๐๐๐ก๐โ๐๐ ๐๐๐๐ก๐๐ฅ๐ก ๐๐๐ ๐ ๐กโ๐๐ก ๐กโ๐ ๐๐๐๐๐ก ๐๐๐'๐ก ๐๐๐๐ I cover a lot of ground in my new @oreilly.bsky.social Radar article on AI and learning about contextโฆ
๐ ๐๐๐ฏ๐ ๐ฒ๐จ๐ฎ๐ซ ๐๐ ๐ฌ๐๐ฌ๐ฌ๐ข๐จ๐ง'๐ฌ ๐ฌ๐ญ๐๐ญ๐ ๐ญ๐จ ๐๐ข๐ฌ๐ค ๐๐๐๐จ๐ซ๐ ๐ฒ๐จ๐ฎ ๐ง๐๐๐ ๐ข๐ญ ๐ ๐ด ๐๐๐๐ โ๐๐ ๐ ๐๐ ๐ ๐๐๐ ๐๐ ๐ ๐๐๐๐๐ ๐๐๐๐๐ฆ๐๐๐๐ ๐คโ๐๐ ๐กโ๐ ๐๐๐๐๐๐ก๐๐๐ก ๐ ๐ก๐ข๐๐ ๐๐๐ฃ๐๐ ๐๐ ๐ ๐๐๐๐ My new @oreilly.bsky.social Radar article just went up, and it's the one I'd been looking forward to writing.
"The agentโs working memory fills up, older information gets silently dropped or compressed, and the agent keeps going without realizing itโs forgotten something." @andrewstellman.bsky.social talks about a pattern he has been refining for detecting and recovering from that problem: bit.ly/4uxjoWr
When Context Collapses: Teaching Agents to Detect and Recover from Lost Memory
Or, how I learned to stop worrying about compaction and love the file system
bit.ly
๐ ๐๐ฒ ๐ง๐๐ฐ @Oreilly.bsky.social ๐๐๐๐๐ซ ๐๐ซ๐ญ๐ข๐๐ฅ๐ ๐๐ซ๐จ๐ฉ๐ฉ๐๐: ๐๐ก๐๐ง ๐๐จ๐ง๐ญ๐๐ฑ๐ญ ๐๐จ๐ฅ๐ฅ๐๐ฉ๐ฌ๐๐ฌ ๐ ๐ป๐๐ค ๐ก๐ ๐ก๐๐๐โ ๐ฆ๐๐ข๐ ๐๐๐๐๐ก๐ ๐ก๐ ๐๐๐ก๐๐๐ก ๐๐๐ ๐๐๐๐๐ฃ๐๐ ๐๐๐๐ ๐๐๐ ๐ก ๐๐๐๐๐๐ฆ A few weeks ago I was deep into a long session with Copilot, working through a multi-phase code review.
๐ ๐๐ก๐ ๐๐ ๐ญ๐๐๐ก๐ง๐ข๐ช๐ฎ๐๐ฌ ๐ ๐ฎ๐ฌ๐ ๐๐จ๐ซ ๐๐จ๐๐ข๐ง๐ ๐ฐ๐จ๐ซ๐ค ๐ญ๐ฎ๐ซ๐ง ๐จ๐ฎ๐ญ ๐ญ๐จ ๐ฐ๐จ๐ซ๐ค ๐๐จ๐ซ ๐ฐ๐ซ๐ข๐ญ๐ข๐ง๐ ๐ญ๐จ๐จ ๐ ๐น๐๐ข๐ ๐๐๐๐๐ก๐๐๐๐ ๐๐ ๐ก๐ค๐ ๐๐๐๐๐๐ก๐ , ๐๐ ๐๐๐๐๐๐๐๐๐ ๐๐๐๐ข๐๐๐๐ AI is generally a pretty lousy writer. You can usually tell when an AI wrote something, and honestly, as an @oreilly.bsky.social author, I'm pretty happy about that.
And if you havenโt been reading Andrewโs posts about agent orchestration, go read those too.
We'll build a real project together in Python, Java, and C#. The goal is to walk out with the muscle memory to use Cursor on real code the same day. Here's a link to the course. I hope to see you there! www.oreilly.com/live-events/... ๐โ๐ข๐๐ ๐๐๐ฆ, ๐ฝ๐ข๐๐ 4 ๐๐ ๐'๐ ๐๐๐๐๐ฆ ๐ฟ๐๐ฃ๐ ๐ธ๐ฃ๐๐๐ก๐ .
๐ ๐๐จ๐ฆ๐จ๐ซ๐ซ๐จ๐ฐ! ๐๐ฒ ๐'๐๐๐ข๐ฅ๐ฅ๐ฒ ๐ฅ๐ข๐ฏ๐ ๐๐จ๐ฎ๐ซ๐ฌ๐ ๐ข๐ฌ ๐๐-๐๐ฌ๐ฌ๐ข๐ฌ๐ญ๐๐ ๐๐จ๐๐ข๐ง๐ ๐ฐ๐ข๐ญ๐ก ๐๐ฅ๐๐ฎ๐๐ ๐๐ง๐ ๐๐ฎ๐ซ๐ฌ๐จ๐ซ, ๐๐ง๐ ๐ฌ๐ฅ๐จ๐ญ๐ฌ ๐๐ซ๐ ๐๐ข๐ฅ๐ฅ๐ข๐ง๐ ๐ฎ๐ฉ ๐ ๐ป๐๐๐๐ -๐๐ ๐ด๐ผ ๐๐๐๐๐๐ ๐๐๐ ๐ค๐๐๐๐๐๐ ๐๐๐ฃ๐๐๐๐๐๐๐
๐ ๐๐จ๐ฎ๐ซ ๐๐ ๐๐จ๐๐ข๐ง๐ ๐ฌ๐๐ฌ๐ฌ๐ข๐จ๐ง ๐ข๐ฌ ๐ ๐๐ญ๐ญ๐ข๐ง๐ ๐๐ฎ๐ฆ๐๐๐ซ ๐๐ฒ ๐ญ๐ก๐ ๐ฉ๐ซ๐จ๐ฆ๐ฉ๐ญ ๐๐ง๐ ๐ฒ๐จ๐ฎ ๐๐๐ง ๐๐๐๐ฅ ๐ข๐ญ ๐ ๐ป๐๐๐'๐ ๐กโ๐ ๐๐๐ฃ๐ ๐๐๐๐๐๐ ๐ฆ๐๐ข ๐๐๐๐โ ๐๐๐ "๐๐๐ค ๐โ๐๐ก" There's a really frustrating problem that everyone coding for AI runs into eventually. If you're a developer who uses AI, I bet you've seen it.
๐ ๐๐ง๐ ๐ญ๐ก๐ข๐ง๐ ๐'๐ฆ ๐ญ๐๐๐๐ก๐ข๐ง๐ ๐ข๐ง ๐ฆ๐ฒ ๐'๐๐๐ข๐ฅ๐ฅ๐ฒ ๐ฅ๐ข๐ฏ๐ ๐๐จ๐ฎ๐ซ๐ฌ๐ ๐๐ก๐ฎ๐ซ๐ฌ๐๐๐ฒ: ๐ ๐๐ญ๐ญ๐ข๐ง๐ ๐ฆ๐จ๐ซ๐ ๐จ๐ฎ๐ญ ๐จ๐ ๐๐ฅ๐๐ฎ๐๐ ๐ ๐โ๐ ๐ถ๐๐๐ข๐๐ ๐๐๐๐๐๐ก๐๐๐ ๐ก๐๐โ๐๐๐๐ข๐๐ ๐กโ๐๐ก ๐๐๐ก๐ข๐๐๐๐ฆ ๐๐๐ฆ ๐๐๐ Slots are still available for my ๐๐-๐๐ฌ๐ฌ๐ข๐ฌ๐ญ๐๐ ๐๐จ๐๐ข๐ง๐ ๐ฐ๐ข๐ญ๐ก ๐๐ฅ๐๐ฎ๐๐ ๐๐ง๐ ๐๐ฎ๐ซ๐ฌ๐จ๐ซ @oreilly.bsky.social live training course on Thursday!
๐ ๐๐ฒ ๐ง๐๐ฑ๐ญ @oreilly.bsky.social ๐ฅ๐ข๐ฏ๐ ๐๐จ๐ฎ๐ซ๐ฌ๐ ๐ข๐ฌ ๐ง๐๐ฑ๐ญ ๐๐ก๐ฎ๐ซ๐ฌ๐๐๐ฒ: ๐๐-๐๐ฌ๐ฌ๐ข๐ฌ๐ญ๐๐ ๐๐จ๐๐ข๐ง๐ ๐ฐ๐ข๐ญ๐ก ๐๐ฅ๐๐ฎ๐๐ ๐๐ง๐ ๐๐ฎ๐ซ๐ฌ๐จ๐ซ ๐ ๐๐ค๐ โ๐๐ข๐๐ , โ๐๐๐๐ -๐๐, ๐ฝ๐๐ฃ๐/๐ถ# ๐๐ฅ๐๐๐๐๐๐ I'm running a new live event on O'Reilly next Thursday, June 4: AI-Assisted Coding with Claude and Cursor.
๐ ๐๐ฒ ๐'๐๐๐ข๐ฅ๐ฅ๐ฒ ๐๐๐๐๐ซ ๐๐ซ๐ญ๐ข๐๐ฅ๐ ๐ ๐จ๐ญ ๐ฉ๐ข๐๐ค๐๐ ๐ฎ๐ฉ ๐๐ฒ ๐ญ๐ก๐ ๐๐ญ๐๐๐ค ๐๐ฏ๐๐ซ๐๐ฅ๐จ๐ฐ ๐๐ฅ๐จ๐ ๐ ๐โ๐ ๐ด๐๐๐๐๐๐๐ก๐๐ ๐๐๐โ๐๐ ๐ก๐๐๐ก๐๐, ๐๐๐๐ข๐๐๐๐ โ๐๐ ๐๐ ๐กโ๐ ๐๐๐๐๐๐๐ ๐๐ ๐ ๐๐๐ค ๐๐๐๐กโ๐๐ฆ ๐๐๐๐ข๐๐
๐ ๐๐ฒ ๐ง๐๐ฐ ๐'๐๐๐ข๐ฅ๐ฅ๐ฒ ๐๐๐๐๐ซ ๐๐ซ๐ญ๐ข๐๐ฅ๐ ๐๐ซ๐จ๐ฉ๐ฉ๐๐: ๐๐จ๐ฎ๐ซ ๐๐ ๐๐ ๐๐ง๐ญ ๐๐ฅ๐ซ๐๐๐๐ฒ ๐ ๐จ๐ซ๐ ๐จ๐ญ ๐๐๐ฅ๐ ๐จ๐ ๐๐ก๐๐ญ ๐๐จ๐ฎ ๐๐จ๐ฅ๐ ๐๐ญ ๐ ๐ป๐๐ค ๐ก๐ ๐๐๐๐ ๐๐๐๐๐ก๐ ๐๐๐ ๐ ๐๐๐๐๐ ๐๐๐๐ ๐๐๐ ๐๐๐ ๐ก๐๐๐๐ ๐๐๐-๐ค๐๐๐๐๐๐๐ค This is the latest article in my O'Reilly Radar series on AI-driven development and agentic engineeringโฆ
๐ ๐๐ฌ ๐๐จ๐ง๐ญ๐๐ฑ๐ญ ๐๐จ๐ฆ๐ฉ๐๐๐ญ๐ข๐จ๐ง ๐ฃ๐ฎ๐ฌ๐ญ ๐ ๐๐ซ๐๐๐ ๐ ๐๐จ๐ฅ๐ฅ๐๐๐ญ๐ข๐จ๐ง ๐๐จ๐ซ ๐๐๐ฌ? ๐ ๐ผ๐ ๐ฆ๐๐ข'๐ฃ๐ ๐ค๐๐๐ก๐ก๐๐ ๐บ๐ถ-๐๐๐๐๐๐๐๐ฆ ๐ฝ๐๐ฃ๐, ๐ฆ๐๐ข'๐๐ ๐๐๐๐๐๐๐๐ฆ ๐๐๐๐๐๐๐ฆ ๐กโ๐๐๐๐๐๐ ๐กโ๐๐ A lot of people talk about the context window as AI's short-term memory and persisted context as long-term. The analogy to human memory doesn't really work for me.
๐ ๐๐ง๐๐ฅ๐ฎ๐๐ ๐ญ๐ก๐ "๐ฐ๐ก๐ฒ" ๐ข๐ง ๐ฒ๐จ๐ฎ๐ซ ๐๐จ๐ง๐ญ๐๐ฑ๐ญ ๐๐ข๐ฅ๐๐ฌ, ๐จ๐ซ ๐ญ๐ก๐ ๐๐ ๐ฐ๐ข๐ฅ๐ฅ ๐ฎ๐ง๐๐จ ๐ฒ๐จ๐ฎ๐ซ ๐๐๐๐ข๐ฌ๐ข๐จ๐ง๐ฌ ๐ ๐โ๐ "๐๐๐๐๐๐๐" ๐๐๐๐ ๐กโ๐๐ก ๐๐๐๐๐ ๐๐๐๐๐๐ ๐๐๐๐ For the last year I've been running an ongoing experiment in AI-driven development: I build production-quality applications and AI skills where I let the AI write all the codeโฆ
๐ ๐๐ญ๐๐ซ๐ญ๐ข๐ง๐ ๐ ๐๐ซ๐๐ฌ๐ก ๐ฌ๐๐ฌ๐ฌ๐ข๐จ๐ง ๐๐๐ง ๐๐๐ญ๐ฎ๐๐ฅ๐ฅ๐ฒ ๐ฆ๐๐ค๐ ๐ฒ๐จ๐ฎ๐ซ ๐๐'๐ฌ ๐๐จ๐๐ ๐ฐ๐จ๐ซ๐ฌ๐ ๐ ๐๐๐ข'๐๐ ๐ก๐๐๐๐๐๐ ๐๐๐๐๐๐๐ก๐๐๐ ๐๐๐ ๐๐๐๐๐ ๐๐ There's something I see developers do all the time that they absolutely shouldn't. You're deep into an AI coding session, and the AI has built up a real understanding of your codebase.
"I think context management is one of the most important skills in AI-driven development, and itโs weird that compared to other AI-related topics, almost nobody talks about it." Check out new #Radar article by @andrewstellman.bsky.social to learn more: bit.ly/4tzWwFg
Why Doesn't Anyone Teach Developers About Context Management?
We overestimate what AI can remember and underestimate what it can orchestrate.
bit.ly
๐ ๐๐ฒ ๐ง๐๐ฐ @oreilly.bsky.social ๐๐๐๐๐ซ ๐๐ซ๐ญ๐ข๐๐ฅ๐ ๐๐ซ๐จ๐ฉ๐ฉ๐๐: ๐๐ก๐ฒ ๐๐จ๐๐ฌ๐ง'๐ญ ๐๐ง๐ฒ๐จ๐ง๐ ๐๐๐๐๐ก ๐๐๐ฏ๐๐ฅ๐จ๐ฉ๐๐ซ๐ฌ ๐๐๐จ๐ฎ๐ญ ๐๐จ๐ง๐ญ๐๐ฑ๐ญ ๐๐๐ง๐๐ ๐๐ฆ๐๐ง๐ญ? ๐ ๐๐ ๐๐ฃ๐๐๐๐ ๐ก๐๐๐๐ก๐ ๐คโ๐๐ก ๐ด๐ผ ๐๐๐ ๐๐๐๐๐๐๐๐ ๐๐๐ ๐ข๐๐๐๐๐๐ ๐ก๐๐๐๐ก๐ ๐คโ๐๐ก ๐๐ก ๐๐๐ ๐๐๐โ๐๐ ๐ก๐๐๐ก๐ www.oreilly.com/radar/why-do...
๐ ๐๐ฒ ๐ง๐๐ฐ @oreilly.bsky.social ๐๐๐๐๐ซ ๐๐ซ๐ญ๐ข๐๐ฅ๐ ๐๐ซ๐จ๐ฉ๐ฉ๐๐: ๐๐ ๐๐จ๐๐ ๐๐๐ฏ๐ข๐๐ฐ ๐๐ง๐ฅ๐ฒ ๐๐๐ญ๐๐ก๐๐ฌ ๐๐๐ฅ๐ ๐จ๐ ๐๐จ๐ฎ๐ซ ๐๐ฎ๐ ๐ฌ ๐ ๐ฟ๐ข๐๐๐๐๐ฆ, ๐กโ๐๐๐'๐ ๐ ๐ค๐๐ฆ ๐ก๐ ๐๐๐ก๐โ ๐กโ๐ ๐๐๐๐ ๐กโ๐๐ก ๐๐ ๐ ๐ก๐๐ข๐๐ก๐ข๐๐๐ ๐๐๐๐๐ฆ๐ ๐๐ ๐๐ฃ๐๐ ๐ค๐๐๐ I really like how this one came out. It starts with a story about how I vibe coded a bus tracker app here in Brooklyn.
"One reason many of us donโt trust AI-generated choice is because thereโs a growing gap between how fast AI can generate code and how well we can verify that the code actually does what we intended," said @andrewstellman.bsky.social in his new #Radar article. Check it out: bit.ly/4ctCG8f
AI Is Writing Our Code Faster Than We Can Verify It
What if the answer to AIโs biggest problem has been sitting on the shelf for fifty years?
bit.ly
๐ ๐๐ ๐ถ๐ ๐๐ฟ๐ถ๐๐ถ๐ป๐ด ๐ผ๐๐ฟ ๐ฐ๐ผ๐ฑ๐ฒ ๐ณ๐ฎ๐๐๐ฒ๐ฟ ๐๐ต๐ฎ๐ป ๐๐ฒ ๐ฐ๐ฎ๐ป ๐๐ฒ๐ฟ๐ถ๐ณ๐ ๐ถ๐ ๐ ๐๐บ ๐ฏ๐ฆ๐ธ @oreilly.bsky.social ๐๐ข๐ฅ๐ข๐ณ ๐ฑ๐ช๐ฆ๐ค๐ฆ ๐ซ๐ถ๐ด๐ต ๐ฅ๐ณ๐ฐ๐ฑ๐ฑ๐ฆ๐ฅ โ ๐ข๐ฏ๐ฅ ๐ด๐ฐ ๐ฅ๐ช๐ฅ ๐ฎ๐บ ๐ฏ๐ฆ๐ธ ๐๐ ๐ด๐ฌ๐ช๐ญ๐ญ I'm really excited about this new O'Reilly Radar piece. It's the first in a few pieces I'm writing about how AI is causing serious problems with code qualityโฆ
๐ ๐ ๐๐๐ฒ๐ฑ ๐๐น๐ฎ๐๐ฑ๐ฒ ๐ฎ๐ป๐ฑ ๐๐ฒ๐บ๐ถ๐ป๐ถ ๐๐ผ๐ด๐ฒ๐๐ต๐ฒ๐ฟ ๐ณ๐ผ๐ฟ ๐ฎ๐ป ๐ฒ๐ป๐๐ถ๐ฟ๐ฒ ๐ฝ๐ฟ๐ผ๐ท๐ฒ๐ฐ๐. ๐๐ฒ๐ฟ๐ฒ'๐ ๐๐ต๐ฎ๐ ๐ต๐ฎ๐ฝ๐ฝ๐ฒ๐ป๐ฒ๐ฑ. ๐ ๐๐ง ๐บ๐ฐ๐ถ'๐ณ๐ฆ ๐ฐ๐ฏ๐ญ๐บ ๐ถ๐ด๐ช๐ฏ๐จ ๐ฐ๐ฏ๐ฆ ๐ฎ๐ฐ๐ฅ๐ฆ๐ญ, ๐บ๐ฐ๐ถ'๐ณ๐ฆ ๐ฎ๐ช๐ด๐ด๐ช๐ฏ๐จ ๐ข ๐ญ๐ฐ๐ต My latest @oreilly.bsky.social Radar piece on the Toolkit Pattern just came out last week, about writing documentation that AI reads instead of your users.
๐ ๐ ๐ ๐ป๐ฒ๐ @oreilly.bsky.social ๐ฅ๐ฎ๐ฑ๐ฎ๐ฟ ๐ฝ๐ถ๐ฒ๐ฐ๐ฒ ๐ผ๐ป ๐๐ต๐ฒ ๐ง๐ผ๐ผ๐น๐ธ๐ถ๐ ๐ฃ๐ฎ๐๐๐ฒ๐ฟ๐ป ๐ท๐๐๐ ๐ฑ๐ฟ๐ผ๐ฝ๐ฝ๐ฒ๐ฑ ๐ ๐๐ฉ๐บ ๐บ๐ฐ๐ถ๐ณ ๐ฑ๐ณ๐ฐ๐ซ๐ฆ๐ค๐ต'๐ด ๐ฃ๐ฆ๐ด๐ต ๐ฅ๐ฐ๐ค๐ถ๐ฎ๐ฆ๐ฏ๐ต๐ข๐ต๐ช๐ฐ๐ฏ ๐ช๐ด ๐ข ๐ง๐ช๐ญ๐ฆ ๐ฐ๐ฏ๐ญ๐บ ๐๐ ๐ธ๐ช๐ญ๐ญ ๐ณ๐ฆ๐ข๐ฅ I'm really excited about this one! It's the third piece in my O'Reilly Radar series on AI-driven development.
The Toolkit Pattern
Why your project's best documentation is a file only AI will read
oreilly.com
๐ ๐ฌ๐ผ๐๐ฟ ๐๐ ๐ฝ๐ถ๐ฝ๐ฒ๐น๐ถ๐ป๐ฒ ๐บ๐ถ๐ด๐ต๐ ๐ฏ๐ฒ ๐น๐๐ถ๐ป๐ด ๐๐ผ ๐๐ผ๐ ๐ฎ๐ฏ๐ผ๐๐ ๐ถ๐๐ ๐ผ๐๐ป ๐ฟ๐ฒ๐๐๐น๐๐. ๐ ๐๐ช๐ฏ๐ฆ ๐ธ๐ข๐ด. ๐๐ฏ๐ฅ ๐ต๐ฉ๐ณ๐ฆ๐ฆ ๐๐ ๐ข๐ฅ๐ท๐ช๐ด๐ฐ๐ณ๐ด ๐ฎ๐ช๐ด๐ด๐ฆ๐ฅ ๐ช๐ต. This one really got me. My AI blackjack simulation that ran hundreds of hands to compare different strategies was reporting great results.
Keep Deterministic Work Deterministic
How eight iterations of a blackjack simulation earned each nine the hard way
oreillyradar.substack.com
๐ ๐ง๐ฟ๐ ๐๐ต๐ถ๐: ๐ฝ๐ฎ๐๐๐ฒ ๐ด ๐ฝ๐ฟ๐ผ๐บ๐ฝ๐๐ ๐ถ๐ป๐๐ผ ๐ฎ๐ป๐ ๐ฐ๐ต๐ฎ๐๐ฏ๐ผ๐ ๐ฎ๐ป๐ฑ ๐๐ฎ๐๐ฐ๐ต ๐ถ๐ ๐ณ๐ฎ๐ถ๐น ๐ ๐๐ฉ๐ฆ๐ฏ ๐ข๐ด๐ฌ ๐ช๐ต ๐ต๐ฐ ๐ธ๐ณ๐ช๐ต๐ฆ ๐ค๐ฐ๐ฅ๐ฆ ๐ต๐ฉ๐ข๐ต ๐ฅ๐ฐ๐ฆ๐ด ๐ต๐ฉ๐ฆ ๐ด๐ข๐ฎ๐ฆ ๐ต๐ฉ๐ช๐ฏ๐จ. ๐๐ต ๐จ๐ฆ๐ต๐ด ๐ฆ๐ท๐ฆ๐ณ๐บ ๐ข๐ฏ๐ด๐ธ๐ฆ๐ณ ๐ณ๐ช๐จ๐ฉ๐ต. I built an exercise for my new O'Reilly Radar article that I had a lot of fun designing.
Keep Deterministic Work Deterministic
How eight iterations of a blackjack simulation earned each nine the hard way
oreillyradar.substack.com
๐ ๐ ๐ต๐ฎ๐ฑ ๐ฎ๐ป ๐๐ ๐ฝ๐น๐ฎ๐ ๐ต๐ฌ๐ฌ ๐ต๐ฎ๐ป๐ฑ๐ ๐ผ๐ณ ๐ฏ๐น๐ฎ๐ฐ๐ธ๐ท๐ฎ๐ฐ๐ธ. ๐๐ฒ๐ฟ๐ฒ'๐ ๐๐ต๐ฎ๐ ๐ต๐ฎ๐ฝ๐ฝ๐ฒ๐ป๐ฒ๐ฑ. ๐ ๐๐ฉ๐ฆ ๐๐ ๐ณ๐ฆ๐ง๐ถ๐ด๐ฆ๐ฅ ๐ต๐ฐ ๐ฎ๐ข๐ฌ๐ฆ ๐ฃ๐ข๐ฅ ๐ฑ๐ญ๐ข๐บ๐ด, ๐ฆ๐ท๐ฆ๐ฏ ๐ธ๐ฉ๐ฆ๐ฏ ๐ ๐ต๐ฐ๐ญ๐ฅ ๐ช๐ต ๐ต๐ฐ. ๐๐ฉ๐ฆ ๐ข๐ณ๐ต๐ช๐ค๐ญ๐ฆ ๐ฆ๐น๐ฑ๐ญ๐ข๐ช๐ฏ๐ด ๐ธ๐ฉ๐บ. I built a pipeline that compares three blackjack strategies: "The Pro" plays strict basic strategy. โฆ
Keep Deterministic Work Deterministic
How eight iterations of a blackjack simulation earned each nine the hard way
oreillyradar.substack.com