davepeck

A friend of mine sometimes jokes that code is “fractally busted” when it’s wrong at every level, from single lines all the way up to the design of the system’s main interfaces.

That’s also roughly what I find after a long English-only session with a coding agent. Our conversation covers a lot of ground... and leaves out details at every level. When I look at what got built, the misunderstandings are fractal too.

I agree it’s important to design it yourself, but UML is only one altitude. Lately I’ve been using a scratch Python file (or even a scratch repo) that Claude and I both edit. Never mind that I’m writing Rust. Python reads like pseudocode, and pass and ... (Ellipsis) let me skip whatever I don’t care about. At any level where I do care, I can say exactly what a method’s signature should look like, what fields a domain model should have, or how it should maintain key invariants. Everywhere else I leave an ellipsis, maybe a docstring, and let the agent decide.

It works both ways: I ask the agent to sketch the current system in Python, with some details elided and others spelled out. Then I edit the sketch and have the agent push the changes back into the much bigger Rust+Tauri app.

Time will tell whether this approach has legs, but when I want to be precise, I find it sometimes pays to just write some code.

> It makes me sad that I missed the boat on startups.

You didn't.

I started my career and my first company during the dot-com boom, before YC even existed and the word "startup" took its current narrow shape.

Things were ugly right after the bubble burst, but the years that followed were some of the best for building I can remember.

The tools change, and not necessarily for the better, particularly if you care about the craft for its own sake. But the hard part was never the build. Even back then, before AI, anyone could copy you fast. The hard part was knowing what to build, for whom, and how to reach them.

What anyone calls a startup doesn't matter. Build the things!

Some of my fave C64 memories as a kid: listening to cassette tapes with stuff stored on them, punching in COMPUTE! listings with MLX, having my uncle explain 6510 and learning the Merlin assembler, long runs in Moon Patrol, making sprites with Gary Kitchen’s GameMaker, trying to understand how GEOS could possibly work, wearing out the spiral binding of the Programmer’s Reference Guide.

A wise troll once said:

> best weapon against complexity spirit demon is magic word: "no"

In counterpoint, I believe small teams can remain small. Small teams can ship simple monoliths with high velocity, commit count, and quality. Service orientation didn’t suddenly become low-cost because of agents; the boundaries between multiple services that version and deploy independently are still tricky beasts to wrangle. And it’s not clear why “running more agents” is inherently desirable or impactful; my small team’s (admittedly anecdotal) experience is that the value quickly saturates.

It may be an age thing. 15-ish years ago, TDD, BDD, Red/Green Testing, and a whole bunch of others were hot topics. They’re still useful approaches to know about, even if some (like, IMO, BDD) didn’t really stand the test of time.

My strong feeling while reading this “philosopher CEO”’s manifesto: consumer VPNs primarily help customers fight for their right to stream Netflix. I’m not sure how effective an advertisement this is.

> the entire point of the ipad is that it's a handicapped tool

amusing nonsense

This thread is pretty quiet for what strikes me as a substantial set of changes with, presumably, more substantial changes still to come for anyone not grandfathered into a Pro plan.

I get the impression that the intersection of HN posters and Copilot users is quite small in practice; that Claude Code and Codex suck up all the oxygen in this room. But it seems plausible we’ll see similar “true costs greatly exceed our current subscription pricing” from Anthropic and OpenAI someday soon…

My parents bought one for the house when I was in elementary school. I still remember the sound of the Speech Synthesizer, discovering 20 GOTO 10, and playing Hunt the Wumpus.

CEOs have many audiences; great CEOs communicate capably with each.

FWIW it's not entirely clear to me who Entire's long-term customer is, but the (interesting!) CLI that shipped today is very much for developers who are busy building with agents.

Dohmke never spoke clearly to developers when he was GitHub's CEO.

Watching their public roadmap to see what happens. Right now, it looks about the same as it has for a while: useful new features and expected maintenance, moving along at a reasonable if not blistering clip.

https://github.com/orgs/heroku/projects/130

Oh, that's very interesting work. And, yes, I'd also be surprised if (today's) agentic tools were at all helpful for that: it's way outside of distribution, and conceptual correctness truly matters.

I made some small contributions to cpython during the 3.14 cycle. The codebase is an interesting mix of modern and “90s style” C code.

I found that agentic coding tools were quite good at answering my architectural questions; even when their answers were only half correct, they usually pointed me in the right direction. (I didn’t use AI to write code and I wonder if agentic tools would struggle with certain aspects of the codebase like, for instance, the Cambrian explosion of utility macros used throughout.)

Per PEP 744, cpython shipped with an experimental JIT (default disabled) in 3.13. It remains experimental in 3.14.

See https://docs.python.org/3/whatsnew/3.13.html#an-experimental...

This is self-destructive defeatism. It is also flat wrong on its substantive points.

Give https://www.paper2audio.com/ a try; it is targeted at just this use case. It’s a Seattle-local startup.

If you’re curious about or playing with t-strings, see https://t-strings.help/

I'd never heard of it, alas. Luckily, they live in pretty different language ecosystems.

Hi! Thanks for your deeply non-silly reply; it's nice to (virtually) meet a cofounder.

If you have time, I'd love to hear your thoughts on Mullvad's campaign here in Seattle.

For what it's worth, I suppose my perspective boils down to: the first three issues aren't issues here in town, or can be addressed in more direct ways (we have a wide choice of providers; 1st party browsers and services cover the gamut of tracking concerns; etc). Circumventing geographical restrictions is useful, but -- perhaps understandably! -- doesn't appear to be what Mullvad is advertising on the trains I ride.

Long ago, in the era of Firesheep and exploding prevalence of coffee-shop Wi-Fi, consumer VPN services were definitely valuable.

But that was long ago. Now, HTTPS is the norm. The only use cases for consumer VPNs today seem to be (1) "pretend I'm in a different geography so I can stream that show I wanted to see" and (2) "torrent with slightly greater impunity".

I live in Seattle and Mullvad VPN seems to have bought approximately all of the ad space on public transit over the past couple months. Their messaging is all about "freeing the internet" and fighting the power. It's deeply silly and, I worry, probably quite good at attracting new customers who have no need for (or understanding of) VPNs whatsoever.

It's not unreasonable to ask but I think it probably is unreasonable to expect a strictly technical solution. Instead, we're in the realm of politics, policy, and law.

Okay. Which, specifically, are the "AI scum" companies you're speaking of?

There are plenty of non-AI companies that also use dubiously sourced IPs and hide behind fake User-Agents.

> Everyone loves the dream of a free for all and open web. But the reality is how can someone small protect their blog or content from AI training bots?

I'm old enough to remember when people asked the same questions of Hotbot, Lycos, Altavista, Ask Jeeves, and -- eventually -- Google.

Then, as now, it never felt like the right way to frame the question. If you want your content freely available, make it freely available... including to the bots. If you want your content restricted, make it restricted... including to the humans.

It's also not clear to me that AI materially changes the equation, since Google has for many years tried to cut out links to the small sites anyway in favor of instant answers.

(FWIW, the big companies typically do honor robots.txt. It's everyone else that does what they please.)

There are many ways that "I used AI to do it all and now I've got $REAL ARR" sniffs wrong to me. To name just two:

1. I code with LLMs (Copilot, Claude Code). Like anyone who has done so, I know a lot about where these tools are useful and where they're hopeless. They can't do it all, claims to the contrary aside.

2. I've built a couple businesses (and failed tragicomically at building a couple more). Like anyone who has done so, I know the hard parts of startups are rarely the tech itself: sales, marketing, building a team with values, actually listening to customers and responding to their needs, making forward progress in a sea of uncertainty, getting anyone to care at all... sheesh, those are hard! Last I checked, Copilot doesn't exactly solve any of that.

Which is not to say LLMs are useless; on the contrary, used well and aimed at the right tasks, my experience is that they can be real accelerants. They've undoubtedly changed the way I approach my own new projects. But "LLMs did it all and I've got a profitable startup"... I mean, if that's true, link to it because we should all be celebrating the achievement.

Thanks, I’ll take a look. Everyone uses these tools differently, so I find AI-generated repos (and AI live-coding streams) to be useful learning material.

FWIW: “Infuriating/wonderful” is exactly how I feel about LLM copilots, too! Like you, I also use them extensively. But nothing I’ve built (yet?) has crossed the threshold into salable web services and every time someone makes the claim that they’ve primarily used AI to launch a new business with paid customers, links are curiously absent from the discussion… too bad, since they’d be great learning material too!

Ive never once successfully gotten a usable sprite sheet out of ChatGPT. The concept seems foreign to it and no matter how hard I try to steer it it’ll find a way to do something hopeless (inconsistent frame sizes; incoherent animations; no sense of consistent pixel sizes or what distinguishes (say) 8-bit from 16-bit era sprites; it’ll draw graph paper in the background for some reason; etc etc.). If anyone has a set of magic prompts for this, I’d love to learn about it. But my suspicion is that it’s just fundamentally the wrong tool for the job.