Soon it’ll be tokens for enterprises and open weights for everyone else

Chaos Agent Engineering: swapping known agents in an orchestrated workflow with “do the very bad thing” agent

Didn’t ever think I’d be saying “as a human” as much as I am nowadays

More experiments with Tone of Voice / LLM as a Judge. `brolint`: detects formulaic LinkedIn builder-bro writing patterns — the closer the match, the higher (worse) the score

dont let the bastards grind you down www.vam.ac.uk/shop/prints/...

``` Write a summary of the work you have done in this session. This document will be used in a fresh context to continue work. Save it to `fresh-context.md` ```

`Update this file to use the gate/goal metaphor/verb`

`write a document so that a Human or Agent would know how to <DO-SOMETHING> in another repository`

bot hr

Opu$

This for me is the next level of a parasocial relationship

"in the transitionary period between now and then things are probably very bad"

Digging Grace Ives tonight

Checking out @transrights.northsky.social and trying to run the AppView locally. Got the Web and React Native versions running ~~AND (don't know if this is a good idea)~~ Storybook running for both Web & React Native

Goal + Gate wins again

For the people who keep showing me a lighthouse accessibility score. www.matuzo.at/blog/buildin...

Blacksky is excited to announce Acorn, a platform for growing and managing communities on decentralized infrastructure. blackskyweb.xyz/introducing-...

Got a version of that Agentic experimentation system working that Shopify used to make performance optimizations. My version is tailored to a React based SPA and it’s working towards a Lighthouse Performance score of 100

Ok it is opening this up to the public for *commentary and suggestions* this is not a final form its a request for comments. If you have good ideas please submit them in a github issue or in replies here. github.com/Navigators-G...

Hot damn: docs.fabro.sh/getting-star...

Really interesting approach to software development described here: simonwillison.net/2026/Feb/7/s... StrongDM, a security company, apparently has no humans writing or reading code. Maybe humans read/write in the harness/validator side of things?

Got access to Copilot CLI at work this week. Experimented with a TDD style multi agent orchestrator using `git worktree`s for each sub-agent. Seems like a very interesting pattern with lots of potential.

A 20% increase in cost from following a best practice seems like a big deal > Meanwhile, the context files increase the # steps in every setting on average by 2.45 and 3.92 steps, respectively, which leads to a cost increase of 20% and 23% on average, respectively

Are the invariants holding?

This looks really good: github.com/supercent-io... It's a collection of approaches to aisdlc Nice links to research, resources Skills are useful, eg: github.com/supercent-io... Feels like a roadmap for personal experimentation

Thinking about: - Compound Engineering retros - A promotion path for instructions/learnings: IC > Team > Org - Getting fancy with storage/distribution: APM, Lexicon/PDS every.to/guides/compo...

Made an ‘iterative optimizer’ agent at work today. Interested to see if this is a useful approach. The API is a bit awkward at the moment. You pass in a goal, measure_command, success_condition, max_rounds, change_scope, and off_limits Just set it to get lighthouse performance to 100. We’ll see…

`rm -rf src/` addresses the “World Series Team” problem

@chadfowler.com i am supposed to be writing right now, but i've spent the last 2.5 hours reading the last few months of your writing. i can't believe i am just discovering it now! there is SO much good stuff here! (yes, i've been under a rock since november ✍️📖☠️)

+1 ‘One day I looked up and realized I hadn’t written a line of code “by hand” in weeks. Everything goes through Copilot Chat, and my building looks a lot more like “alright, do this part next,” “here’s a screenshot of the design, match this,” “ah, no, I don’t like this, try this instead.”’

Enrich my neighbours

@rude1.blacksky.team just want to say I appreciate you. Keep on killing it

“What would WordPress look like if it were invented today?” www.pootlepress.com/2026/03/are-...

Is there any way to search what you’ve seen on the Discover feed or in a thread? I saw a comment recently about how in an agentic world valuable skills are the ability to drive results, effective communication, etc.

this post about killing human code review is obviously written to piss everyone off but this is a great name for what I already do by hand, namely ping-pong between claude and codex for code review. low false positive rate, quite helpful www.latent.space/p/reviews-dead

Happy #DeLaSoulDay! 3/3 🌼🌼🌼 Celebrate with us at the @npr.org Tiny Desk ☁️🏠⛅️ youtu.be/5AVYDHTOixU?... #TinyDesk #NPRTinyDeskConcert

Got a spec based multi agent orchestration setup that follows TDD (test, implement, refactor). Seems to work well.

it's almost time for SRECon26, which means it's been less than a year since @ferd.ca and i were delivering the closing keynote at SRECon25. my perspective has changed A LOT since then, and the advice i would give has changed too. i wrote it up here. charitydotwtf.substack.com/p/my-hypothe...

Generated Mermaid diagrams of all unit test cases for a function. It really helped to clarify the behaviour of the function. Easier to understand than mentally running the tests or console logging the inputs/outputs

Just started reading asdlc.io and like what I see Thanks @villetakanen.com

“Nirvanna the Band the Show the Movie” is the funniest movie I’ve seen in a long time

“But what if enough is enough? What if all we need is [good old-fashioned AI]? What if we don't want to pollute the environment, or steal the labor of artists, or cause a global shortage in hardware”

hell yes, scot friends; www.tartanregister.gov.uk/tartanDetail... did you know there's a transgender tartan officially reigstered with the scottish government?

Looking into PROSE at work: danielmeppiel.github.io/awesome-ai-n... It claims to make software development with LLMs reliable and scalable Seems not bad. It’s a collection of practices & techniques instead of a library Would love to hear if anyone has experiences working with this or similar

“Many critics suggest that LLMs are useless. I disagree. It’s debatable if they are worth their costs—economically, environmentally, socially—but users of an LLM do something both powerful and concerning: they engage with an alienated abstraction of their own thoughts.” - @eryk.bsky.social

So what will they do? They will blindly feed your review comments straight to their agentic coding model because that's the only thing they know how to do. By doing so the vibe coder has put the maintainer in the weird position where the maintainer is doing agentic coding through a shitty UX

demo

Just got everything lined up to release `ed3d-plugins`, the Claude Code harness I use in my day-to-day. While they're yelling about gas towns, I present an e-bike, just don't drive it on the sidewalk. I'm biased, but I think it's good. I've tried a few similar, and I keep going back.

By now, asteroid 2026 Dreadstone has impacted the Indian Ocean, sparking a firestorm not seen on this planet since the K–Pg Extinction Event. Most people just want to be with their families. Here are three ways effective business leaders can leverage targeted KPIs to turn tragedy 😔 into triumph 💪