The versatility and flexibility of (Claude) Skills consistently blow my mind. I'm currently using them a lot for knowledge base building/management and podcast/technical content production but you can do absolutely anything with them, as they're just markdown with a bit of

🔮 2026 prediction >>> we're going to shift back hard to browser-based everything. Smartphones like apps, agents love browsers. 🫵 Start the year learning with @isaac_flath and me how to cusomise your browser with agentic coding! 🐵

I'm long @code and @opencode. When everyone is done hyperventilating over Claude Code (which tbh they have every good reason to be excited about - I love it too) they'll realise that there are excellent open and versatile agents they can pick up. 2026 is going to be a blast! 💥

https://t.co/tGVkxVSh4u

Posting about "What I did with Claude Code today" is becoming the lunch photo instagram of techies. Like, yeah, it's _a little bit_ interesting what you've had for lunch / manoeuvred Claude to do today. But not a lot. And keep in mind that everyone eats lunch / drives Claude.

🤔 Sometimes the best way to teach is by showing, not telling. In fact, that's one of the best ways to teach an agent a new skill. See how in E03 of Introduction to Agent Skills.

My love letter to Agent Skills

@nikunj @GeminiApp Totally. Flash is 90% the quality at 10% the cost and 10X the speed. In so many cases that's such a great tradeoff to make. I use it as my main driver in @code and @opencode, going to Opus and GPT-5.2 when I really need that extra quality.

Claude Code gang is going to realize how goated @GeminiApp Flash 3.0 is in a few months.. For the price, size of the context window, speed AND capabilities, I really don’t think anything comes close. Google absolutely cooked on that model 🔥

Judging by my feed, Opus would start complaining how sycophantic humans are towards it

My biggest worry for this year is that once the journos hear about @opencode the stock market will crash.

My love letter to Agent Skills

The "Ralph" thing is: 1. A brilliant (and I guess accidental?) exercise in branding. 2. Proof that even the best agentic models still aren't good enough to decide when to stop and when to continue without external poking.

Confidence is cheap now. Taste is still expensive.

Vibe Coders of the world, unite! Let's make our voice heard, and convince @AnthropicAI to support AGENTS . md in @claudeai. 📝 Sign the petition: https://t.co/WTslp2ic6f

🗓️ In a short time, all jobs that happen in front of a computer will be achieveable better, faster, more reliably, and much cheaper using AI. Soon after, jobs requiring physical movement will follow the same path. 🧘‍♀️ Learn to meditate. Get to know your mind. You'll need it.

Starting the year strong ... Happy 2026, everyone!!

@intellectronica @isaac_flath full post here: https://t.co/Ef1U9FWevS watch the workshop on YT here: https://t.co/HJbO0C4avY join our free livestreamed podcast about Effective AI-Assisted Coding next week (or register for the recording):

Serious hacks to level up your AI-Assisted Coding from two of the best: @intellectronica & @isaac_flath. The Living Documentation is such a useful practice because it’s is the closest we’ve got and the simplest possible implementation of a learning agent that gets better over

One cool advantage of @GitHubCopilot CLI is that it can authenticate with a GITHUB_TOKEN — that's perfect for async background jobs (AKA "Continuous AI"). Claude and Codex want you to switch to API keys and per-token billing when you move from interactive to headless.