Andrew Healey@healeycodes
Oct 31
revisiting some of my pre-AI blog posts

first up is my sokoban solver

currently trying to hillclimb my way to a SOTA solver with the latest models πŸ§—
0
0
5
333
Andrew Healey@healeycodes
Oct 28
really glad y'all like this one

I'll go decompile some more of my fav games!
Screenshot of Hackernews pointing to my post in the 9th position > 9. * Decompiling Fast Food Tycoon 2: Drawing the World (healeycodes.com) 60 points by healeycodes 5 hours ago | hide | 11 comments
0
0
8
237
Andrew Healey@healeycodes
Oct 24
I spent a few days decompiling my favorite childhood game, Fast Food Tycoon 2, and figuring out:

β€£ world rendering
β€£ sprite animations/character movement
β€£ light mechanics

so much to dig through and discover the dev team were on fire here: healeycodes.com/decompiling-fa…
1
1
15
836
Andrew Healey@healeycodes
Oct 24
Replying to @healeycodes
pizzeria pipeline!
1
0
0
93
Andrew Healey@healeycodes
Oct 24
Replying to @healeycodes
plus how the game AI scores your pizza designs!
0
0
0
50
Andrew Healey@healeycodes
Sep 3
the stronger the searcher, the more dangerous metadata about the existence of an answer becomes

in security: "just a rumour of a bug is enough to find an exploit these days"

in chess: Carlsen would be MUCH stronger with just one bit per game: "this position is critical"
0
0
1
194
Andrew Healey@healeycodes
Sep 2
I wrote about the performance of LLM tokenizers

like most things you want to go fast, it's better to use compact contiguous state instead of complex data structures!

healeycodes.com/what-makes-llm…
Screenshot of blog post: Below are the first four merges from the real goldshire trace. GPT-2 chooses the lowest-ranked valid adjacent pair (not the leftmost pair or the longest token). So r+e, which has rank 4, wins before the leading-space pair, even though it is near the end of the match. The important bit to take away is that each merge can expose another candidate. ...
0
1
15
650
Andrew Healey@healeycodes
Aug 16
I wanted to write a blog post titled (provocatively) "We Could Have Had Coding Agents in 2019"

I used the GPT‑2 XL weights (no post-training) and backported several later agent-design breakthroughs (action/observation loops, filesystem tools, etc.)
screenshot of terminal: β†’ ./run.sh GPT-2 XL CODING AGENT Model: openai-community/gpt2-xl Device: mps Seed: 419 INITIAL PROMPT β†’ GPT-2 ──────────────────────── You are a programmer controlling a tiny project. Choose one action at a time. Use exact filenames from Computer. After a FAIL names a file, READ that file next. Actions: ACTION: LIST ACTION: READ filename.py ACTION: WRITE filename.py CONTENT: complete file contents END ACTION: RUN COMMAND: python check.py ACTION: FINISH Example: Task: Fix the project so all checks pass. Programmer: ACTION: LIST Computer: check.py message.py names.py
1
0
9
441
Andrew Healey@healeycodes
Aug 16
Replying to @healeycodes
but GPT-2 can't reliably emit JSON, or call tools in a meaningful way.

the only hint of agency came from leaning into text completion: Programmer β†’ ACTION, Computer β†’ observation, etc.

to get it to fix tests, the examples were so close to the answer that it was kinda useless
2
0
2
193
Andrew Healey@healeycodes
Aug 16
Replying to @healeycodes
however, I haven't given up and have high hopes for "We Could Have Had Coding Agents in 2Μ΄0Μ΄1Μ΄9Μ΄ 2021"
0
0
3
95
Andrew Healey@healeycodes
Aug 2
I added πšπšŽπšπšŽπš› to the typescript compiler

but along the way, I realised that what I want is different syntax for πšžπšœπš’πš—πš

I just want to be able to write:

πšžπšœπš’πš—πš πšŠπš πšŠπš’πš πšœπšŽπš–πšŠ.πšŠπšŒπššπšžπš’πš›πšŽ()

but you can't

↧

healeycodes.com/adding-defer-t…
Screenshot of blog post: So the goal is to be able to write TypeScript code like this: async function readFile(path: string) { await sema.acquire(); defer sema.release(); // New! // ... use resource } The TypeScript Compiler The TypeScript compiler (tsc) is mostly a static analysis engine. Its complexity lies in type-checking a fundamentally dynamic language, and supporting extremely incremental compilation to meet latency expectations in an IDE. Lucky for us, we don't need to worry too much about types or other analysis in order to add our defer statement. tsc already has the machinery for "recognize syntax X, replace it with equivalent syntax Y."
0
1
13
954
Andrew Healey@healeycodes
Jun 27
two sum in lisp is actually really cute
(defun two-sum (xs target) (let ((seen (make-hash-table))) (loop for x in xs for i from 0 for need = (- target x) do (multiple-value-bind (j ok) (gethash need seen) (when ok (return (list j i)))) (setf (gethash x seen) i))))
3
0
7
364
Andrew Healey@healeycodes
Jun 25
new post!

I wrote a tiny compiler that lowers simple kernel loops into explicit data-parallel code (lanes! masks!)

ofc I made up a language to make the input/output code easier to read
Screenshot of post: A Tiny Compiler for Data-Parallel Kernels Jun 2026 A lot of fast code starts as a boring loop. Modern hardware can perform the same operation on multiple values at once (e.g. SIMD and SIMT), and sometimes we write code directly for those execution models but other times, a compiler starts with regular-looking code and rewrites it so multiple loop iterations can run together. I built a tiny compiler (~180LOC of Python) to understand what that transformation looks like. My compiler lowers kernels (rewrites them into a simpler, more explicit form where data parallelism is visible). The input is a small hand-written AST, and the output is a lowered IR that I print as Python-like code. Rather than going all the way from source code to instructions, think of this compiler as an intermediate step in a larger compiler. Let's take a look at an example. Scaling audio is easy to parallelize, but it is still common to write non-explicitly parallel code like this:
1
0
4
633
Andrew Healey@healeycodes
Jun 25
Replying to @healeycodes
healeycodes.com/a-tiny-compile…

read all about it ⇩
0
0
1
94
Andrew Healey@healeycodes
Mar 26
quickjs is GOATed

delightful API surface

I was able to hack together a tiny runtime and event loop v quickly with it
Screenshot of: Building a Runtime with QuickJS Mar 2026 A JavaScript engine (e.g. V8, JavaScriptCore) executes JavaScript code. It doesn't know about things like files, HTTP requests, or timers. On the other hand, a JavaScript runtime (e.g. Node.js, Bun) is a more complete environment where JavaScript runs. It contains a JavaScript engine, extra APIs, an event loop and task queues, and platform-specific features. That's what I'm hacking on today: a tiny runtime with console.log, process.uptime(), setTimeout and clearTimeout, fs.readFileSync and fs.readFile, as well as an event loop and worker pool for file I/O. Built on top of QuickJS. Here's an example program it can run: const startedAt = process.uptime();
1
1
24
1,670
Andrew Healey@healeycodes
Mar 26
Replying to @healeycodes
healeycodes.com/building-a-run…

hope you've been enjoying the C posts
0
0
2
228
Andrew Healey@healeycodes
Mar 17
I built a tiny shell with C so I could learn more about syscalls and pipes

great fun

I found C to be quite suited for the task (fork, execvp) ... until I had to manage some string lifetime stuff πŸ₯²
Screenshot of: To recap a bit, I'll step through what happens when ls $HOME | grep foo is entered. It's tokenized into ["ls", "$HOME", "|", "grep", "foo"]. Then expanded into ["ls", "/Users/andrew", "|", "grep", "foo"]. The flat token list is separated into structured pipeline commands: ["ls", "/Users/andrew"] ["grep", "foo"] The shell creates a pipe to connect the output of ls to the input of grep. The child commands start via fork, and execvp swaps them into the target programs. By the wonderful design of Unix: if grep reads faster than ls writes, it blocks waiting for more data; if ls writes faster than grep reads, the pipe buffer fills and ls temporarily blocks. This synchronization happens automatically through the pipe, while the shell simply waits for both child processes to finish. The output of grep isn't connected to a pipe as it's the last command, and any results are displayed to the user.
2
4
49
3,787
Andrew Healey@healeycodes
Mar 17
Replying to @healeycodes
read my notes and code snippets and thoughts
healeycodes.com/building-a-she…
0
0
6
301
Andrew Healey@healeycodes
Mar 17
glad y'all like this one 🫑
0
0
18
821
Load more