Joe Gibbs Politz

This is written up and public, and in production now (behind a flag, but several projects use this mode as the default already), if it's useful to calibrate: github.com/brownplt/pyr...

ever since i was a small boy i knew i wanted to charge the pepper grinder

Reasoning about security with object capabilities really just... keeps working in the age of agentic coding and AI on the backend. That's a pretty strong endorsement.

I guess now I'm just getting sassy

I guess it's not that weird that agents would have a manner of speaking (“smoke test”, “you're right to push back”, “sharp observation”). It's the kind of thing you notice about team members and joke about. It's just weird that we're all talking to the same coworker.

Preparing these lecture notes is massively entertaining and instructive to me. I am still not fully untangled on what the end lesson will be here. (Still in progress, but writing up the notes as I get to each next lecture, so publishing increments) ucsd-cse-115-215.github.io/sp26/lecture...

Ariana Venti

Slightly increase a band Alkaline Milk Hotel

Just reflecting on my little ✨2026✨ updates to the morning routine: - coffee - check inbox - check if the 🤖 completed a weeks-long engineering task overnight or needs a small nudge - breakfast - brush teeth - …

bun appear to be experimenting with a million line claude port from zig to rust lol

What do you do for *commit messages* on work authored with agents? Write them yourself like nothing's changed? Agent drafts and you edit? Curious where folks have landed I've converged on a format where I let Claude do its thing but add a "Joe says" at the top. Example: github.com/brownplt/pyr...

With multiple sub-agents running algorithmic improvements checked against various oracles and verifiers and fuzzers, you too can get results like these in an evening of work.

One of the best features of Claude Code for me is the ability to tell something, in natural language, “I want you to make me look at each of these things one by one in a distraction-free terminal interface and make a decision of the form XYZ”

Things my computer says to me in 2026: “The "giant is a substring of gigantic" claim from my compacted self was wrong — "giant" is NOT a substring of "gigantic" (g-i-g-a-n-t-i-c does not contain g-i-a-n-t). Good call stopping there.”

As someone who used to write detailed software specifications for a living, I can see why AI is seductive: the idea that someone, or something, might take my work and continue reading past the abstract.

I'm enjoying the open-ended exit slip questions in my course this quarter

My very spicy take on LMS vs. GenAI is maybe best characterized by the “let them fight” gif. It's an arms race that I am happy to simply sidestep with other syllabus design decisions.

❯ How has your impression of my experience as a developer impacted this session? ... What I've picked up from context: - You built this full-stack app yourself and wrote a detailed implementation plan with specific API shapes and architectural trade-offs ... No, Claude, you did that lol.

horseshit has more of a “preposterous(ly bad), likely through negligence or malice” connotation IMO. Sort of in the vector space between bullshit (vacuously false and a waste of time) and dogshit (low quality, potentially through no fault of its own, like it's just dogshit tier stuff).

This is a post about animal excrement and the English language. bull$#!% = nonsense/lies chicken$#!% = petty or cowardly horse$#!% = nonsense/lies dog$#!%= low quality ape$#!% = wild bat$#!%= crazy Ordered above from oldest to newest: bull$#!% (1914), bat$#!% (1971).

Should just do Captain Teach again but mix agent code in. dl.acm.org/doi/pdf/10.1...

WTAF steve-yegge.medium.com/gas-town-eme...

Solana Beach

Had two playdates this week where it became obvious that my kid is the one with an embarrassing amount of practice in Mario Kart.

How To Do New Year's Resolutions

Santa Claus 💀🔥🌲 Creature — Rogue Horror Cannot be blocked by walls When attacked and not blocked, put 3 Battery-Powered Present tokens into play under the defending player’s control with “at the beginning of your upkeep, pay 1 or lose 1 life” 2/5

I also want to say thank you to the person who introduced me to the French word 'beaucoup' this year. It means a lot.

At some point this quarter in a meeting I said “students can’t regret skipping lecture for its own sake if all the lectures were incentivized. Feeling that regret can be a college-level learning outcome beyond a single course” Anyway it's likely that's pure cope but it sounded cool when I said it.

Feels good to be 20 again.

1. “We are the only poets, everyone else is prose.” 2. “My life had stood a loaded gun.” 3. “Jealousy is the cousin, the cousin of greed.”

fighting for the good cause

I need to know more. Melee, Brawl, Ultimate, OG? Which stages, stock counts, timer? As written this is way underspecified and not reproducible. www.frontiersin.org/journals/imm...

Has anyone done a tower defense game set in a (my) living room where the mobs are toddlers?

they didn't tell me that the donkey was dying when i picked him. they were just like "your donkey update is that your donkey has died. sorry." and transferred us to another donkey

I was pretty pleased with this program I used as a motivation for understanding virtual memory.

Have expertise with functional programming, interpreters, and Web technology? Like Rust, wasm, etc.? Looking for a job with a great team (like @adamchalmers.com )? Please see this job ad, and share with others you might know! zoo.dev/jobs/4619187...

Just told my course staff that variable-length arrays have been part of C since “the late 1900s”.

When you open the game in an emulator so it's in its own window but the emulator miscalculates the size because it thinks it's fullscreen, that's Doom scrolling.

Modern developement: “Anyway, computers appear to be running somewhere in the country and serving the Pyret compiler and the UI.” (After my investigation into impacts of today’s outage on code.pyret.org)

Are there any benchmarks (etc) which show the costs of the SysV x64 ABI relative to a somewhat-or-perfectly-optimal ABI for that program? [Not for single microbenchmarks, because I can definitely create horrible overheads there, but on larger programs / benchmarks.]

[Werner Herzog voice]: The computer… it is a cruel, unfeeling god. Left too long in its ceaseless hum, it festers with demons. The human, fragile and insignificant, clings to the ritual of restarting… as if this act could stave off the abyss of digital chaos. But the darkness always returns.

I've now given my talk “High Standards, Multiple Tries – How I've Been Grading” a few times. I wrote it up as a blog post: jpolitz.github.io/notes/2024/0... 3 principles: 1. Standards-based grading with high-fidelity assessments 2. Retries for everything 3. Coarse rubrics (0-4 not 0-100)

useful overview of what's going on in Wasm land, from Andy Wingo in ACM Queue