If you're already using mise to manage tool versions and environment variables, then consider using mise tasks as well — it's a great way to wrap up common workflow commands into short, easy-to-access tasks.
www.visualmode.dev/add-mise-tas...
I’ll be curious to know how this differs from their /code-review plugin. github.com/anthropics/c...
One major difference seems to be that it is directly integrated as a GitHub app that runs relative to the PR lifecycle.
I should have said “well before the November 2025 Inflection”. bsky.app/profile/jbra...
I feel so inspired to improve my own (micro)blog tooling every time I spend even 5 minutes on @simonwillison.net’s site.
E.g., here is a tag that was linked to in the body of a post. It has its own description and then a bunch of posts that were tagged with it -> simonwillison.net/tags/novembe...
"Whatever AI you are using right now is going to be the worst AI you will ever use."
That's from @emollick.bsky.social in mid-2023 while writing Co-Intelligence, well before models like Opus 4.5/6, Gemini 3, etc. were released.
However, I was talking to a guy at this wedding recently whose "watch me restore vintage audio equipment" youtube channel is modestly, but surely growing. I think there is a huge moat around people doing content on tangible and tactile real-world areas of interest.
In a lot of ways, this feels like an awful time to be trying to break into content creation, both due to over-saturation and the ways that AI is changing discoverability and consumption habits. Not to mention the way AI flattens the effort required to produce "content".
I’m caught somewhere between my prestige TV era and my reality TV era
yeah Claude Code, Cursor, and I'm sure other LLM software tooling, have had this for a while now.
![A screenshot from claude.app where before responding to a prompt, the LLM is prompting me with a specific question "Should notes persist across essions (i.e. saved to a database), or is this a single-session tool where notes live only in memory/local state?" And then presents four options, 1) Persisted (user accounts + database), 2) Local/session only (no auth needed), 3) Start local, plan for persistence later, and 4) [text area to type something else]](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:rceroyjz5cpubdf6brys2fsm/bafkreidbhooilcxqp73f3rgxengforihoaltglhpddxuhl5yvkj6cq6itm)
A screenshot from claude.app where before responding to a prompt, the LLM is prompting me with a specific question "Should notes persist across essions (i.e. saved to a database), or is this a single-session tool where notes live only in memory/local state?" And then presents four options, 1) Persisted (user accounts + database), 2) Local/session only (no auth needed), 3) Start local, plan for persistence later, and 4) [text area to type something else]
ooh, the Claude app got first-class support for follow-up questions from the LLM
Very nice and very subtle Linear feature I noticed this week:
When I'm delivering the feature described by an issue, I often post a quick video as a comment on the issue, demonstrating how it works.
The subtle feature is that Linear *automatically subtitles the videos*.
I'm amused by the serendipity of finding myself writing a TIL about timezone handling on the day of the DST cutover.
github.com/jbranchaud/t...
I'm finding "AI-exposure" to be a useful term and metric for thinking about what kinds of jobs will be more or less impacted by AI.
Software has lots of AI-exposure, bartending very little.
I'm sure it originates somewhere else, but I heard it in this study --> www.anthropic.com/research/lab...
Linear feature request: another `copy` option for issues that creates a markdown link with the issue title and URL and then puts it on your clipboard
Incredible name for an editor tho
The cover of Ethan Mollick’s Co-Intelligence: Living and Working with AI, which has the distressed look of an old oil painting with a hand reaching up to pluck an apple from a tree — garden of Eden, original sin vibes.
I’m reading @emollick.bsky.social’s Co-Intelligence: Living and Working with AI.
A necessary condition of being able to experiment with building my own pieces of LLM and neural network tools is having a sizeable (but not too big) and real-world text to train on.
I decided to use my 1700+ TIL repo. Here is how I rolled up all the files.
www.visualmode.dev/build-a-smal...
3/
Silent Parade (2022)
Keigo Higashino
bsky.app/profile/jbra...
Whether it is for myself when I grab lunch or go for a walk, or for LLM agents, this @scriptkit.com script prompts me for a duration and then invokes caffeinate to keep my computer awake for a set time.
Super handy!
github.com/jbranchaud/d...
, ') ' (End of parenthetical), '.\n\n```' (End of a sentence, two new lines, then opening of a code block), 'ation' (Very common suffix for a lot of english language words).](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:rceroyjz5cpubdf6brys2fsm/bafkreidpn5gaici3ffvawoc6wmyf63bsxs6x26pqmxylphxv7navriq24i)
A list of multi-character tokens that are part of this Byte Pair Encoding's vocabulary including 'you ' (I speak to 'you' a lot in my TIL writing), '.md)\n- [' (Tons of links to other markdown files in a list format), '](https://' (Markdown lists to the public internet), ') ' (End of parenthetical), '.\n\n```' (End of a sentence, two new lines, then opening of a code block), 'ation' (Very common suffix for a lot of english language words).
The LLM book I'm reading hand-waved over how Byte Pair Encoding works, so I decided to implement my own naive version to understand it better.
I just ran it against a rollup of all (1700+) markdown files in my TIL repo and its cool to see real-world tokens it added to the extended vocabulary.
oh wow, not the answer I was expecting. that's impressive.
Love these!
What do you use to create slides / visuals like this?
@simonwillison.net what are your favorite newsletters and who are your favorite people to follow in the AI, LLM, and agentic coding space?
I'm interested to find more high-signal stuff to complement what I'm reading from you.
Slowly, painfully becoming a Twilio A2P 10DLC compliance and registration expert.
Did you know your REPL (probably) supports popping open your editor to write or edit longer / more-complex statements?
I wrote about it in my latest blog post www.visualmode.dev/an-editor-in...
I'm going to use Sonnet 4.6 over Opus 4.6 as my daily driver this week and see how it does. Will report back.
screenshot of the linked blogmark which shows the title "Tiktokenizer: visualize LLM prompt tokenization" followed by some tags, a paragraph, then an image of the token visualizer, and then another paragraph
I'm experimenting with using Cloudflare's R2 to host and serve images for my blogmarks, as well as my actual blog.
Cloudflare made this very easy to set up.
still.visualmode.dev/blogmarks/235
I've being doing Rails my entire career and I've never seen someone refer to Rails' scaffolding generators as capital-S "Scaffold" as if it is a stand-alone tool.
I've been back on a kick of posting regularly to my blogmarks page 📑
still.visualmode.dev/blogmarks
Woke up at 6:30am to make coffee and watch the gold medal game of olympic men's hockey. Absolutely thrilling game.
