Moment of Zen youtu.be/jYiexkNTgKY?...
Posts by Ben Tucker
Still very much a work in progress, but it’s this: github.com/btucker/espa...
Not an original concept, but essentially it automatically gives you terminal tabs based on got worktrees. And soon it’ll have a web UI to go with it.
I accidentally started building a little Mac app today. Claude + Swift/SwiftUI makes this so much fun! What a difference from Obj-C when I last tried this 15 years ago.
/experiment lets you run A/B tests without actually deploying anything
/interview lets you poll the synthetic users with any question as you go
I’ve been getting a lot of value using /interview to help answer questions that come up with brainstorming
I’ve been using this pattern of synthetic users running in a loop while building with Claude code. I just packaged it up as a plugin.
/dogfood checks whether your app does what it should
/user-trial has users try to use it without knowing how it should work
github.com/btucker/blin...
Cursor employees sit in front of shelves with various objects introducing Cursor 3
Cursor gets in on the action:
Am I missing something or is there very little done in the space of benchmarks of coding agent harnesses together with models? SWE-Bench is just the model as it runs it's own harness.
This is the only thing I've found:
marginlab.ai/trackers/cla... marginlab.ai/trackers/codex
I like the idea of github.com/steveyegge/beads, but in practice I kept running into problems.
So Claude Code & I built a much simpler version: github.com/btucker/micr...
No SQLite, no daemon, merge conflicts. Fully works in Claude Code Web.
I also dislike the term because all LLM output is a “hallucination.” It just so happens these hallucinations line up with reality with high frequency.
Someone on HN posted about a common hallucination & then someone else noticed that the HN post caused Gemini to doubt itself. I'm thinking we should call these "Heisenjections"
news.ycombinator.com/reply?id=460...
THREAD: Judge Ellis is the first federal judge to review extensive body cam video of DHS's actions in Chicago. She finds that DHS *repeatedly* misled the public and made claims that were disproven by agents' own videos.
I'll go through some of the most egregious ones here.
US citizen Maria Greeley, a Latina who was adopted, was walking home from work downtown when masked agents zip tied her. They said she “doesn’t look like a Greeley.” This is how the government treats people under current deportation policies.
www.chicagotribune.com/2025/11/15/l...
svg of a pelican riding a bicycle on a white background
Here's what the SVG looks like on a white background.
Ha! This just reminded me of my experiment from March after OpenAI launched their new image generation model of having ChatGPT generate an image, then trace it to SVG. I just tried this again & it's still pretty similar: chatgpt.com/share/69161f...
Took two prompts still.
bsky.app/profile/btuc...
I think you could also do that using claude itself. You don’t necessarily need another LLM.
Certainly possible, but I’m not sure the value would be high. Another approach is you can use Claude Code in headless mode from a Claude Code plugin/skill. This allows using Claude as the LLM from those.
What I find useful about local models is when you don’t want the content to leave the machine.
What sorts of use cases are you thinking about?
it turns out PyPI doesn't yet support macOS 26 wheels. I opened a PR to add this: github.com/pypi/warehou...
Then I wanted to make it easier to play with, so another hour with Claude Code and I had a plugin for @simonwillison.net's llm: github.com/btucker/llm-...
What's cool this is you don't have to install anything other than some python packages & you have full access to a reasonably capable LLM.
I've been wanting to explore Apple's local LLM ("Foundation Models") that's now there on every macOS 26 install. I was surprised I couldn't find any python bindings. A few hours with Claude Code (still mind blowing to me this is possible!) and I have an initial version: github.com/btucker/appl...
The comments on the Slate launch video are quite something. They seem to be making someone people want.
youtu.be/jKVwEg4ZToI?...
Hopefully fatally funny
Painting saying STOP COMRADE TRAITOR TRUMP
Protesters in Daley Plaza, Chicago
Protesters in Daley Plaza, Chicago
Protestor holding sign saying: WE ARENT GETTING PAID TO BE HERE. WE HATE YOU FOR FREE!
Great turnout today for the #50501 protest in Chicago
An extract from the judgement: It is difficult in some cases to get to the very heart of the matter. But in this case, it is not hard at all. The government is asserting a right to stash away residents of this country in foreign prisons without the semblance of due process that is the foundation of our constitutional order. Further, it claims in essence that because it has rid itself of custody that there is nothing that can be done. This should be shocking not only to judges, but to the intuitive sense of liberty that Americans far removed from courthouses still hold dear.
🚨BREAKING: 4th Circuit Court of Appeal throws out Trump’s request to halt proceedings in the case of wrongfully imprisoned Kilmar Abrego Garcia.
It is a devastating judgement.
Just read this extract.👇
He was afraid people would believe in evolution.
This is so fucked up. ICE agents in Massachusetts smashed the window of a car to grab a Guatemalan guy who was with his wife. They were waiting for their lawyer to arrive because he has a pending asylum case.
Story via @wcvb5.bsky.social: www.wcvb.com/article/ice-...
I’m all for folks going on the bropods & getting in front of these audiences. BUT, get it in writing that you get edit approval for any clips.
Case in point, this edited clip of a “debate”
youtube.com/shorts/uRVOy...
We need to stop saying “wrongly deported.” That means we expel someone out of the country.
No, we sent him to a US-funded prison.
Abrego Garcia is being “wrongly imprisoned.”