So you want to know what uses energy in your house? £16 of tools and a few hours free time is all you need. Here's how to find out. A 🧵

What are y'all using to convert Claude Code and Codex session logs to Markdown to index with qmd? /cc @tobi

Yet again @plotlygraphs saves the day. @ApacheECharts is good, but sometimes your Sankey diagram needs DAGs and layout control to show how the process really is!

“Why did it all seem to matter so much? And what do we do now with our memories, dreams, and hurts, our hope and weariness, in the Church or out of it?” Thanks to @LucySixsmith for speaking to me about her v good book on the ‘History-Maker generation’ churchtimes.co.uk/articles/2…

Really liking GPT-5.5, but boy it burns tokens at Opus 4.7 rates. Exhausted my 5h Plus plan window in 44 minutes. Never managed to hit a limit before @thsottiaux :)

This is a helpful thread on the new features and how to use them. Already used 2 of these today!

So *that's* how I get it to work!

I had a conversation recently with someone who was desperate to get AI into their workflow. They'd seen what it could do and wanted that productivity boost. But when I looked at how they were actually working, I had to be honest with them: the AI wasn't the problem they needed to solve. Their task management was a mess. Priorities shifted daily. Nobody was clear on what good output actually looked like. Adding AI to that situation wouldn't make things better, it would just make things worse faster. There's a version of AI adoption that looks like progress but is really just an expensive way to produce mediocre work more quickly. Before you can prompt an AI to write a report, you need to know what a good report looks like. Before you can ask it to prioritize your work, you need a system for knowing what matters. I've spent a long time building an AI setup that I actually find useful, and the reason it works is that the underlying processes were already solid. The AI references real procedures, not guesswork. The groundwork came first, and then the AI became genuinely helpful rather than just fast. If you're finding that AI tools aren't giving you the results you expected, it might be worth asking whether the foundations are there before blaming the tools. #AI #AgencyAdvice #Productivity

Due to popular demand, I am proud to announce my brand new course! Mastering Anthropic's Terms of Service: A 12-Module Masterclass "Finally understand what you're allowed to build." -------- Here's a quick peek into the course and what you can expect: Module 1: Is Claude Code a Product or a Suggestion? Module 2: The -p Flag: Personal, Professional, or Please Don't Sue Me Module 3: ACP: What It Means and Why Even Anthropic Isn't Sure Module 4: Can You Ship an App or Just Stare at One Lovingly? Module 5: Advanced Interpretive Reading of the 1,300-Page SDK EULA Module 6: Case Studies from Developers Who Tried and Disappeared Bonus Module: How to tag Anthropic employees on @X for Clarification ------ Join 43,000 confused developers. Lifetime access. No refunds. Link in comments 👇

Opus is being ever so dumb this morning. GPT-5.{2,3} has now sorted 2 tasks it couldn't

Love this experience by one of the basic income pilot participants in Wales. They got £1,290 a month after tax for two years (about $1,862/mo), and at first they planned to not get a job, but then after getting into the pilot, they changed their mind, because why not do more to get more money?

Surprised Byobu doesnt get more mentions. Resizable tabs, persistence, nice scrollable windows all driven by keyboard shortcuts Been using it for the last 10 years and never looked back.

Hi @augmentcode your web page talks about using a Claude Code subscription inside Intent. Is your app allowed by their license?

What's a good, tiny local ML model for contextual spellings which I can continually finetune? E.g. for: "She will meet linden under then linden tree" => "She will meet Lyndon under the linden tree"

How can I help @claudeai or @OpenAi efficiently reason about the @ChromiumDev or @firefox codebases? They are massive and asking questions like "Where is. ___?" or "How does ___ work?" needs a̶ ̶l̶o̶t̶ ̶o̶f̶ too many tokens just to get started.

Nice to see a bit of realism about AI and software development at companies.

If anyone from the @ChromeDevTools team is reading this, for accessibility testing it would be amazingly useful to be able to control browser zoom via the MCP. github.com/ChromeDevTools/ch… has more details

this is huge. all agents have to do is send "Accept: text/markdown" and we’ll automatically convert the response to Markdown on-the-fly for any enabled site. way fewer tokens. no need to handle it on your own.

Debugging drag-n-drop today and discovered Playwriter MCP from @__morse has full support: page.mouse.move(), down(), up(). Nice work, can watch the animations as AI tests it! @ChromiumDev MCP can't - it jumps from start to end locations.

What happens when you combine obligations without the means to meet them