Jonny Buchanan (@jbscript) header
Jonny Buchanan (@jbscript) profile picture

Programmer/web developer from Norn Iron living in Australia. Making Control Panel for Twitter and Control Panel for YouTube at https://www.teapawt.org/t.co/ZlvkRfy6xS

Brisbane, Australia
Joined August 2012
Photo by Jonny Buchanan (@jbscript)Photo by Jonny Buchanan (@jbscript)Photo by Jonny Buchanan (@jbscript)Photo by Jonny Buchanan (@jbscript)Photo by Jonny Buchanan (@jbscript)Photo by Jonny Buchanan (@jbscript)Photo by Jonny Buchanan (@jbscript)Photo by Jonny Buchanan (@jbscript)Photo by Jonny Buchanan (@jbscript)
Men will literally cold-open a 4,500 line PR for something you're in the middle of working on instead of going to therapy
106
Memory is for trying to make your LLM thing sticky for people who (can't be bothered/don't know that you can/don't know how) to put things they want it to "know" about them in their personal instructions It's terrible for literally any other use case
1
1
146
I've observed that it pleases a certain group of users (who think of themselves as power users) *so much* when this happens
34
Remember when VS Code release notes used to be something other than wall-to-wall AI stuff which only unfortunates have to actually use?
93
If the purpose of a thing is what it does, then the purpose of Sonar (in an enterprise environment, at least) is to _completely incorrectly_ make every PR have a red X next to it
I finally watched Tariq Shaukat's keynote from AI Engineer World Fair, and there's one line that I've been noodling on: "code is abundant, confidence is scarce." We're writing code faster than ever before, but we're not necessarily that much faster at shipping it. Agentic coding has pushed lines of code up 2,450% and commits up 240%, but production releases are only up 30%. The bottlenecks still exist later in the SDLC. Human reviews aren't the silver bullet (even though I admittedly wish they were) to fix this. The research in the talk shows reviewers agree with AI 93% of the time when it's right, and 80% of the time when it's wrong. Which is... not good enough. LLM-as-a-judge has the same problem, in that identical prompts get different verdicts 7-17% of the time, and asking "are you sure?" flips the answer 7-34% of the time. There's a confidence formula framing that I like, which is intent × guide rails × verification × evidence. If any one of those is zero, the whole thing is zero. Verification is deterministic where it can be. That's what @SonarSource is pitching here, Sonar Vortex keeping agents anchored on intent, and SonarQube as the deterministic check on the other side. If your team is shipping agent-written code faster than you can review it, worth a watch: fandf.co/4A8zZ6h
88
Asking me if I'm gonna need a Jira ticket for that
88
astro-lazy-youtube-embed v0.6.0 now emulates the current YouTube embed style, and supports Astro 7 (whoops) github.com/insin/astro-lazy-…
1
120
Feels a bit pointless publishing these now 🤔
41
One by one, the lore of all the doppelgangers I accidentally receive emails for is revealed
2
86
Fucksake OFX
1
122
Hello fellow Django and JSX enjoyers, there are dozens of us bsky.app/profile/jbscript.de…
109
The AI (LLM) Mandates to AI (LLM) Cost Controls pipeline is real
112
Guy Montgomery's Guy-Mont Spelling Bee Live! in Brisbane next weekend 🎉
74
TIL a trailing & in nested CSS adds an ancestor selector developer.mozilla.org/en-US/…
103
Control Panel for YouTube v1.37.0 adds an EXPERIMENTAL option to disable the new watch page layout on desktop, a new option to disable AI-generated chapters and much more… Release notes: github.com/insin/control-pan… If you have the new layout, please give it a go and let us know
1
22
1
117
3,776
Somehow missed UH OH! by @waaxband getting a video, skill issue youtube.com/watch?v=BlXvLS-A…
108
Oh no! Not my personalised recommendations!
111