I joked about the "QA engineer for my agents" thing earlier but it's honestly so peaceful to just sit and polish software with the help of agents.
You can really fight a lot of slop with intentionality and attention to detail.
I joked about the "QA engineer for my agents" thing earlier but it's honestly so peaceful to just sit and polish software with the help of agents.
You can really fight a lot of slop with intentionality and attention to detail.
An illustration of two playing cards, bearing Aljoscha and Sammy styled as face cards.
worm-blossom.org#y2026w8
This week! Part 5 of βwhy is worm-blossom called thatβ heads to the Pacific Ocean. Sammy feels like a frog. Aljoscha is heads down (but has provided a good tune nonetheless).
use the filesystem to describe your software (screaming architecture). prefer many small, scoped files rather than sprawling ones. turns out this is good for agents too!
> A file called ./billing/invoices/compute.ts communicates much more than ./utils/helpers.ts
bits.logic.inc/p/ai-is-forc...
donβt bloat your AGENTS.md - use it as a βtable of contentsβ for your repo to progressively disclose info and instructions
> A short AGENTS.md (roughly 100 lines) is injected into context and serves primarily as a map, with pointers to deeper sources of truth elsewhere
openai.com/index/harnes...
really like the idea of continuous agents for reviewing integrated changes, updating docs, surfacing risks
Definitely a builder at heart and feel very energised using AI. I feel like Iβm actually able to spend more time (and enjoyment) on the crafting now - especially docs, tests and design. Things that used to get in the way of building
Is the penny proxy itself serverless?
here's a guide to setting it up! gist.github.com/lukehedger/5...
Claude Code session in Ghostty terminal, showing a custom status line with model name, context % used and $ cost
got a nice little custom status line in Claude Code!
model name, context % used and $ cost
Nice! Same here! I feel like my job is exciting again
I think youβre right here Pete. Going to be interesting to see the techniques we develop for optimising context.
What measurements/signs do we have right now to tell us the agentβs performance is suffering due to bloated context?
Now on YouTube, Monday's CMU Database Seminar on Aurora DSQL. Thanks again to the awesome CMU folks (especially Andy Pavlo) for inviting me to do this talk.
www.youtube.com/watch?v=mK2h...
Really? Whatβs your job?
What does it look like?
AI-generated code shifts the real burden of development onto *validation*, yes.
But humans can't validate your code by reading it any more than tests can validate it by turning green.
"Is it gonna work? Is it gonna work?? IS IT???" Bub, you don't get to know til you put it in production.
Wow! Thanks for the link. This is pretty crazy
Oh really!? In what way?
Very cool research on a CodeBuild misconfiguration which could have had significant consequences. Iβm a bit disappointed that there wasnβt more done to secure the supply chain after the Q Developer incident.
www.wiz.io/blog/wiz-res...
There arenβt yet many things in cloud computing that have the exact shape Sprites do: Instant creation No time limits Persistent disk Auto-sleep to a cheap inactive state
One of the interesting things about sprites to me is how similar they are to Durable Objects / Virtual Actors fly.io/blog/design-...
Off to the pub
status.claude.com
I understand the devs who mourn the loss of their identities because agents write code pretty well now. This is your craft. You trained for this.
But no one said you have to stop writing code to use agents.
Chefs still cook. They just use microwaves and air fryers too.
Ooo I like the theming on your Claude terminal UI⦠have you customised it?
This is a nice intro to Skills if you havenβt dug into them yet claude.com/blog/buildin...
> Skills give Claude Code access to your procedural knowledge through progressive disclosure, revealing information in layers only when needed, rather than flooding the context window
What sort of thing are you using Claude Skills for?
So far Iβve got skills for:
- writing ADRs
- creating Jira tickets (via acli jira)
- developing Lambda functions
- developing CDK stacks
- enforcing typescript best practices
- teaching patterns (tests, design patterns, DI, composition etc)
Monitoring Third-Party Webhook Delays with AWS Durable Functions
Using Lambda durable functions to monitor delivery and performance of third party webhooks
dev.to/aws-builders...
"Why is Claude Code better/worse than [other coding agent] - theyβre both using the same models under the hood?"
A reasonable question, so I spent some time reverse-engineering Claude Code's interactions with the model to illustrate why coding agent matters
blog.thepete.net/blog/2025/12...
π§΅...
First of all, the code base is under 3 weeks old. On a scale of "polished diamond" to "uncut rough" to "I just smuggled it 400 miles upriver in my ass," I'm going to characterize Gas Town as "You probably don't want to use it yet." It needs some Lysol. It's also 100% vibe coded. I've never seen the code, and I never care to, which might give you pause. 'Course, I've never looked at Beads either, and it's 225k lines of Go code that tens of thousands of people are using every day. I just created it in October. If that makes you uncomfortable, get out now.
Gas Town is also expensive as hell. You won't like Gas Town if you ever have to think, even for a moment, about where money comes from. I had to get my second Claude Code account, finally; they don't let you siphon unlimite dollars from a single account, so you need multiple emails and siphons, it's all very silly. My calculations show that now that Gas Town has finally achieved liftoff, I will need a third Claude Code account by the end of next week. It is a cash guzzler.
If you're not at least Stage 7, or maybe Stage 6 and very brave, then you will not be able to use Gas Town. You aren't ready yet. Gas Town is an industrialized coding factory manned by superintelligent chimpanzees, and when they feel like it, they can wreck your shit in an instant. They will wreck the other chimps, the workstations, the customers. They'll rip your face off it you aren't already an experienced chimp-wrangler. So no. If you have any doubt whatsoever, then you can't use it.
Gas Town needs a lot of fuel. It both consumes and produces guzzoline, or work molecules. Aside from just keeping Gas Town on the rails, probably the hardest problem is keeping it fed. It churns through implementation plans so quickly that you have to do a LOT of design and planning to keep the engine fed.
This is utterly unhinged in the best possible way. High praise for running a tour of the plane while it's a quarter-built and mid-flight.
steve-yegge.medium.com/welcome-to-g...
Some choice excerpts.
Perfect advice for blending a coding assistant into your workflow in 2026. If youβve been put off or overwhelmed so far, Rizelβs approach could work for you!