TIL the Codex VSCode extension doesn't work well with profiles in config.toml - if you have a profile set, it won't let you control the model and reasoning effort, Remove the profile and the extension will set the model correctly. It's clearly WIP but mostly works well.

When everyone zigs, zag. While everyone is obsessing with the risks of having their likeness cloned, Sam Altman made himself cloneable. Now he's in your face non-stop.

Leading VS Code through the AI revolution! 🚀 Conversation with @pierceboggan on open-source AI tools, rapid iterations, dev empathy, and keeping coding joyful. Must-watch for @code and @GitHubCopilot fans, and for all devs embracing AI. 💻 open.substack.com/pub/eliteai...

ChatGPT got a lot of bad press for routing (fair - it doesn't work), but I don't think it's getting enough praise for making it so straightforward to set reasoning budget. You can now explicitly choose between no thinking, light, standard, extended, heavy, or pro. That's great!

"AI Work-Slop" — derogatory term used by your lazy colleagues as a put-down when they wake up from their slumber and realise you've become 10X more productive.

To my neighbour with the ukelele: may you be happy, may you be well, may you be peaceful, may you know no suffering, may your ukelele never break or burn

The {structure collapsing → I'm falling} sequence is a frequent dream / hallucination I experience which is at the same time unpleasant and fascinating (unfortunately it often disturbs my sleep). I wonder if that's common and this is the reason it's such a hit in AI gen videos.

When people think about John Searle, they usually think the Chinese Room Experiment. But Searle's most important contribution, IMO, is Speech Act Theory. Framing language as a tool for taking action in the world and on other humans. Still revolutionary. RIP

A few people asked me how I turn my messy stream-of-conciousness dictated text into neat prompts. So there you go: gist.github.com/intellectro...

👐 Can't wait to start welcoming students into the Discord server for our course - Elite AI-Assisted Coding (with @isaac_flath and myself). 🤝 Nothing in the world beats a community for learning.

I switched to @usemonologue for dictation a couple of days ago, and it's really good. If you're looking for an easy to use and affordable AI dictation package, check it out. And if you aren't yet in the habit of talking to your computer, give it a try. It's a huge improvement.

Sonnet 4.5 early vibes: not sure I like it any better than I liked the previous Claudes for coding — it still got that teenage spirit smell — but WOW it's good for agentic workflows.

One of my weirdest psychological quirks is how I agonise in deep pain every time I consider signing up for a new AI subscription.

Many people take the increasing compression of the benchmark tables to mean that LLMs are becoming commoditised. I think that's the wrong conclusion. In fact, the opposite is true. 1/8

The benchmarks are becoming meaningless. There are now so many excellent agentic models with subtle differences (speed, cost, tool use, reasoning quality and effort). Welcome Sonnet 4.5. Looks and feels very good. As do GPT-5/-codex, Grok Fast, Gemini, and the many open models.

Debating an AI pair programmer can sharpen your code! 💻🤖 Uncovers new tools, refines designs, hones prompts. Sometimes a teacher, sometimes a better rubber duck — both make @isaac_flath elite. elite-ai-assisted-coding.dev/p/i-argue-w...

How I like to be in contact: Discord > Signal > Email > WhatsApp > Telegram > X > LinkedIn > Slack > SMS > Paper Post > Phone Call

💡 Instruct your AI coding agent to use `git commit --author="AI <[email protected]>"` to attribute commits clearly. Filter AI changes with `git log --author="AI"`. Keeps your repo history tidy, easily identify agentic commits! elite-ai-assisted-coding.dev/p/attribute...

Many of the most prominent LLM AI critics sound like they're looking for Élan Vital.

Defending against prompt injection in AI coding? Scale security to your risk: Treat model output as untrusted, isolate contexts, sandbox agents with RBAC, and red-team relentlessly. No one-size-fits-all — adapt to stakes! 🛡️💻 open.substack.com/pub/eliteai...

My Anki decks: Mixture of Excerpts

My issue with the calls for "continuously learning" AI is that such a system would be an absolute nightmare. Impossible to reliably evaluate, manage, and control. And unlike current systems, inherently unsafe. Why would anyone want to create something like that?

How do AI coding agents even work? LLMs with codebase access & tools for planning, searching, and editing. Core loop? Perceive → Plan → Act → Observe. Simple yet powerful! 🚀 Join our free workshop to build one live. elite-ai-assisted-coding.dev/p/how-ai-co...

📐 Discover Ruler: @intellectronica's open-source tool for centralizing context & configs across AI coding agents. 🔧🚀 To learn more about AI-assisted software development, join our course Elite AI-Assisted Coding. 🔗 in 🧵⤵️

🧑‍✈️ @GitHubCopilot CLI drops! Terminal agent for your copilot sub — install via npm, prompt or automate tasks. Powered by Copilot models, billed per request for crystal-clear costs. Fills the CLI gap perfectly! Full review: elite-ai-assisted-coding.dev/p/github-co...

When @isaac_flath and I launched Elite AI-assisted Coding, publishing every day was not an explicit goal. We just wanted to share everything we know and are excited about in this space. Two months later ...

🧰 @isaac_flath’s toolbox: Claude Code + Amp and Cursor. Gemini CLI + WhisperFlow for blogs. CLI: LLM for easy piping. Gemini/Claude web projects for context. To learn more about AI-assisted software development, join our course Elite AI-Assisted Coding. 🔗 in 🧵⤵️

🇬🇧 en route to London; if you're around and want to meet give me a shout

something comes and pins me to the sky www.youtube.com/watch?v=Vpo...

🚀 2025 AI Coding Survey: @claudeai 4 Sonnet top model, @code rules IDEs, @GitHubCopilot leads extensions. CLI agents? Claude Code, followed by long tail of terminal agents. 📊 Full insights on dev tool adoption 👇 open.substack.com/pub/eliteai...

A lot of old media is now boomers and Xers complaining about zoomers being too smart, sophisticated, hardworking, not drinking and drugging and fucking around enough. Older generations always complain about the youngsters ... but complaining that the youngsters are too good?!

🧰 @intellectronica's advice for devs: Treat AI coding tools like portfolio management. Don't chase every new feature—pick a few, master them deeply, and supercharge your workflow. 💻🚀 To learn more about AI-assisted software development, join our course Elite AI-Assisted Coding. 🔗 in 🧵⤵️

After using Grok 4 Fast for a bit longer (mostly in @raycast, as the driver for mundane utility manoeuvres) I still like it a lot, but there's one problem: it _really_ likes tool calling. It binges on tool calls. Like, it will do 20 different calls for a simple question.

🎉 GPT-5-Codex in @code / @GitHubCopilot ... I think this is the beginning of a beautiful friendship. 😍

Top English editing trick I learnt: before you're done, search for "very" in the text and remove ~90% of occurances.

"Human in the Loop" is getting old. Let's adopt a new term: "Human in the Pool". Our AI future will only arrive when we no longer need to babysit the agents and instead can sip piña coladas by the pool. To understand / design / build the future, think autonomouns AI.

שנה טובה ומטוגן www.youtube.com/watch?v=-FU...

For every dumb and irresponsible human activity we're not going to get news headlines for "someone did something dumb and dangerous ... with AI".

The solution for merge conflicts isn't a merge agent that can resolve merge conflicts. It's a coding agent and VCS that operates on syntax trees. Just sayin'... 🤷‍♀️

it annoys me profoundly when people shorted "podcast" to "pod". all my brain can think of when is hear it is that "podcast isa broadcast".

"$COMPANY just dropped {PDF}!" is the mansplaining of social media.

it take years to find the nerve www.youtube.com/watch?v=8Eu...

apparently it's rosh-hashana (the jewish new year, literally "head of the year" ) today. grandma called to remind me. so to all who celebrate, have a fab year! and please try to be nice to one another. 🫶

My default AI chat model and many of my AI commands in @raycast, which previously used a combination of Gemini Pro, GPT-5, and Sonnet, are now set to ** @grok 4 Fast (Reasoning) **. It's about as good, and the speed makes everything better. Give it a try, xAI really cooked.

My guess: "fixing vibe-coded apps" is really codename for "take my vibe-coded prototype and remake it as useful software I can deploy to prod". DFY pretending to be DWY, helps people maintain the illusion that what they're doing is software development.

AI gave you a giant, messy function? Don’t ship it. Ask it to split into atomic pieces: validate, process, format, handle errors—each with one job. Faster tests, easier debugging, real reuse. Turn monoliths into modules 💡 open.substack.com/pub/eliteai...

The reason you are confused about whether AI is better for "juniors" or for "seniors" is that it's a bogus distinction. It was bogus all along, the availability of AI just helped nudge you into rethinking it. Use the opportunity.

Grok Code Fast 1 + Artisanal Homemade Agent — Screen Saver

Best new feature of the new 🍎 iEcosystem is podcasts pausing when you fall asleep. No longer need to browse the history and try to figure out at which point I became unconcious and stopped listening. Major quality of life improvement.