How This Ex-Meta L8 Engineer Ships 40 PRs a Day with AI Agents
Kun Chen quit big tech and now ships more code in a day than most engineers ship in a month — by building three tools that move him almost entirely out of the loop.
June 7thA 47-minute conversation with Matt Van Horn, who ships viral open source projects without reading code or even the plans his agents write.
The bottleneck to shipping is not coding skill — it is the discipline to give an agent a structured plan first, trust it to execute without babysitting, and stress-test the output with a single adversarial question instead of reading every line.
Matt Van Horn ships production-quality software without reading code or even the plans his agents write. His core loop — /ce plan then /ce work via Compound Engineering — forces the agent to produce a structured plan before executing, which defeats agents' natural tendency toward lazy shortcuts. He runs 6 agent windows simultaneously and asks one adversarial question per session to verify scope. His most viral project, the Printing Press, auto-generates a CLI + agent skill + MCP for any website by HAR-sniffing hidden APIs and absorbing community GitHub projects — no official API required. He has merged PRs into Python, Vercel, OpenCV, and 200+ other repos, mostly via automated overnight systems, and credits the resulting network as the most valuable asset he has built.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →
Matt states he does not read code; Peter frames the interview premise.

Compound Engineering's ce plan / ce work cycle; the real morning Agent Cookie example; why the plan is for the agent, not you.

Matt's background as a self-described nerd with no CS degree; arriving in SF in 2007; BCAC inflection point Thanksgiving 2024.

Failed 'build a programming language' idea -> Printing Press; Peter Steinberger's SQLite-backed CLI pattern; power-user persona design.

Live Papa John's CLI print; CLI vs MCP auth reliability; what gets printed (CLI + agent skill + MCP for every tool).

How Matt started contributing; the adrenaline of a merged PR from Peter Steinberger; automated overnight contribution systems.

last30days 20K stars, Printing Press 3K stars, Python CPython cross-language method suggestion feature; banned from Rust.

Two finished projects Matt fears launching; the founder simulation game; the build-for-yourself floor.
The entire shift in agentic engineering is realizing the structured plan is not your checklist to review -- it is the agent's instruction to stop cutting corners.
“Agents have a propensity to be lazy. They wanna make you as happy as possible as fast as possible with as few tokens spent as possible.”
“What if anyone could print their own CLI about anything?”
“Every API has a secret identity.”
“Even if I had no users of agent cookie, like, if no one used it, I get value out of it. Like, it helps my agents be so much better.”
“I'd say I'm the most incompetent person that I've met in terms of actual skills to success ratio.”
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
The guest opens the interview confessing he has not written code since high school -- and then spends the next 47 minutes explaining how he became one of the most prolific open source contributors in the AI tools ecosystem.
00:01
00:47
01:29
02:05
02:41
03:16
03:52
04:28
05:17
05:51
06:05
06:56
07:40
07:48
08:39
09:01
09:36
10:26
10:44
11:38
12:14
12:49
13:25
13:53
14:37
15:13
15:49
16:24
17:00
17:36
18:12
18:48
19:33
19:59
20:35
21:11
21:47
22:35
22:44
23:45
24:10
24:49
25:22
26:10
26:24
27:09
27:34
28:06
28:40
29:24
29:52
30:44
31:27
31:44
32:19
33:07
33:43
34:19
34:55
35:30
36:13
36:46
37:18
37:54
38:44
38:49
39:25
40:01
40:53
41:33
42:11
42:56
43:05
44:03
44:11
45:14
45:25
46:24
46:51
47:43Kun Chen quit big tech and now ships more code in a day than most engineers ship in a month — by building three tools that move him almost entirely out of the loop.
June 7thA 13-minute walkthrough of the four plain-text files that give an AI enough context to provide genuinely useful life and career advice.
June 17thA 19-minute live build showing how to make Claude Code skills that grade their own output, remember past sessions, and get better every time you run them.
June 3rdJosh Pigford built and sold Baremetrics, now runs five AI products solo — and his Claude Code skill stack is the most systematic one on record.
May 31stA 4-minute demo of two open-source skills that turn Claude Code plan output into interactive MDX wireframes, API specs, and diffs — and the argument that the plan layer is where engineers will live next.
June 16thAndrew Warner and Peter Cooper run through 10+ GitHub repos that give AI agents cheaper web access, less token bloat, and a design taste system — all free and ownable.
June 12th