📹 https://nitter.gq/t.co/XrvfUwXjBt (Agentic coding videos) 💻 https://nitter.gq/t.co/WDfX6R7Mqm (Agent sandboxes) ⚛️ Prev Physics @ @cambridge_uni
Tokyo, Japan
Joined August 2019
- Tweets530
- Following1.5K
- Followers1.2K
- Likes4.8K
Opus 5.5 seems to really like "receipts" language when it comes to talking about evidence, proof, logs, etc.
- "Comes with a receipt"
- "That's the receipt"
- "That's a big drop on the receipt"
After analysing 4,500 past sessions, turns out it says "receipt" more than 2x as much as Opus 4.8
5
386
Super bullish on explainer videos!
Two weeks ago, my favourite use case of Opus 5.5 was to quickly understand the rules of a board game.
My prompt was literally:
"Make a video explaining how to play the game with ElevenLabs CLI, 3D animations, lots of diagrams, no text."
I was shocked by how well it turned out!
Now, I spend close to an hour each day watching AI explainer by Opus 5.5 about pretty much everything.
We'll be spending a lot more time trying to understand the outputs of language models. A few thoughts, tips & tricks:
Writing. Something I've had success with: Ask your LLM to explain something in ASD-STE100, it's a controlled language specification originally developed for aerospace maintenance documentation. LLMs well-versed in this language and it comes with heavy constraints on clean writing style that I often find a lot more readable. Sometimes I've tried to soften it a bit e.g. ask for "80% of the way to ASD-STE100" because the spec is quite stringent. But even better:
Diagrams / images. Instead of writing, ask your LLM to create a diagram. These can be a lot easier to process, parse, and understand. But even better:
Web pages. Ask for output "in HTML" to get a beautiful, interactive webpage. LLMs are getting really good at frontend and can create beautiful experiences, animations, etc. But even better:
Explainer videos. The output format I am most bullish on is fully custom / bespoke explainer videos generated on any arbitrary topic. Experiment with things like "Create a 3b1b style video explainer on X. Use my ElevenLabs API key for audio narration". (you'd need an API key for the latter or you can ask your LLM to find you decent free alternatives that use your local compute). This is actually starting to work!
In summary:
- As LLMs get better, they will do more and more of the legwork autonomously, and a lot more of our work will rise up the abstractions into oversight and understanding.
- Luckily, LLMs can help here too because as intelligence and code are increasingly abundant, you can ask for large, custom, discardable software artifacts (e.g. web apps, video explainers) that would have never made sense to create before. Push the boundaries here and you'll be surprised.
My favourite mod is one to show an overview of what Claude did after a long turn
As agents work for (a) longer or (b) parallelise more. It's more important than ever to use the right visualisations or artefacts to quickly understand what was done rather than reading walls of text
You can now mod Claude Code:
- Change how it behaves
- Customize the UI
- Swap in your own features
Write one with a few lines of TypeScript, or have Claude build it for you. Mods ship inside plugins, so you install them with /plugin in the CLI or desktop app.
A few examples:
3
11
711
I shared this mod and other mods I made in my GitHub repo here: github.com/ray-amjad/claude-…
2
1
239
Whenever I upgraded machines, I always worried about whether to (1) sell the old one or (2) keep it around in case I needed it later.
Now I just connect it to Tailscale, tell my agents it exists (along with its specs), and they find better ways to put it to use than I ever could.
6
369
Just had Claude Code launch 100 sandboxed agents running gpt-6-luna-xhigh with Playwright to find problems on my site from as many angles as it can think of
Gotta make the most of these resets before they expire lol
1
6
425
Set up a cool daily routine where I have an agent try to “hack” into my apps.
It opens up an issue if it succeeded. Then another agent creates the patch.
After a few weeks of this, the “hacker” agent is running for longer and longer trying to find a new exploit.
1
7
623
GPT-6 Astra seems really good at helping me rewrite messy abstractions that older Claude models wrote.
Fable seems less willing to just rewrite and felt too "biased" by the existing implementation.
1
5
340
Yesterday, @ClaudeDevs released a new Claude Code feature: function hooks.
It's perhaps the most powerful feature released in Claude Code to date.
I explained it with a whole bunch of examples in this video.
My favorite new feature in Claude Code released yesterday: function hooks.
After hours of playing around and being inspired by @trq212's articles, I made my own.
2
5
626
My favorite new feature in Claude Code released yesterday: function hooks.
After hours of playing around and being inspired by @trq212's articles, I made my own.
