HNHacker News
TopNewBestAskShowJobs

gordonhart

1,171 karma · joined November 26, 2016

GitHub: https://github.com/gordonhart

Co-founder: https://kolena.com

Atlas of Space: https://atlasof.space

submissionscomments
gordonhart··on I'm not worried about AI job loss
Honestly curious, have you seen agents succeed at this sort of long-trajectory wide breadth task, or is it theoretical? Because I haven't seen them come close (and not for lack of trying)
gordonhart··on I'm not worried about AI job loss
Codex with their flagship model (currently GPT-5.3-Codex) is my daily driver. I still end up doing a lot of steering!
gordonhart··on I'm not worried about AI job loss
Curious to hear if you've seen this work with 100k+ LoC codebases (i.e. what you could expect at a job). I've had some good experiences with high autonomy agents in smaller codebases and simpler systems but the coherency starts to fizzle out when the system gets complicated enough that thinking it through is the hard part as opposed to hammering out the code.
gordonhart··on I'm not worried about AI job loss
Here's an example ticket that I'll probably work on next week:

    Live stream validation results as they come in
The body doesn't give much other than the high-level motivation from the person who filed the ticket. In order to implement this, you need to have a lot of context, some of which can be discovered by grepping through the code base and some of which can't:

- What is the validation system and how does it work today?

- What sort of UX do we want? What are the specific deficiencies in the current UX that we're trying to fix?

- What prior art exists on the backend and frontend, and how much of that can/should be reused?

- Are there any scaling or load considerations that need to be accounted for?

I'll probably implement this as 2-3 PRs in a chain touching different parts of the codebase. GPT via Codex will write 80% of the code, and I'll cover the last 20% of polish. Throughout the process I'll prompt it in the right direction when it runs up against questions it can't answer, and check its assumptions about the right way to push this out. I'll make sure that the tests cover what we need them to and that the resultant UX feels good. I'll own the responsibility for covering load considerations and be on the line if anything falls over.

Does it look like software engineering from 3 years ago? Absolutely not. But it's software engineering all the same even if I'm not writing most of the code anymore.

gordonhart··on I'm not worried about AI job loss
Whenever I get worried about this I comb through our ticket tracker and see that ~0% of them can be implemented by AI as it exists today. Once somebody cracks the memory problem and ships an agent that progressively understands the business and the codebase, then I'll start worrying. But context limitation is fundamental to the technology in its current form and the value of SWEs is to turn the bigger picture into a functioning product.
gordonhart··on Good Riddance, 4o
I kind of agree with GP more than the author here. OpenAI got these people hooked and pulling the plug is potentially more harmful than letting them continue to chat with it until they move on naturally (assuming that they eventually will)
gordonhart··on Good Riddance, 4o
Releasing the weights is an easy and low-cost way for OpenAI to fix this problem.
gordonhart··on Good Riddance, 4o
That's... a strategy. Matter of time before an AI companion company succeeds with this by finetuning one of the open-source offerings. Cynically I'm sure there are at least a few VC backed startups already trying this
gordonhart··on Good Riddance, 4o
Anecdotally, 4o's sycophancy was higher than any other model I've used. It was aggressively "chat-tuned" to say what it thought the user wanted to hear. The latest crop of frontier models from OpenAI and others seems to have significantly improved on this front — does anybody know of a sycophancy benchmark attempting to quantify this?
gordonhart··on GPT‑5.3‑Codex‑Spark
Yeah, given that frontier model training has shrunk down to a handful of labs it seems like a very solvable problem to just build the stack directly without CUDA. LLMs are mechanically simple and these labs have access to as much engineering muscle as they need. Pretty small price to pay to access cheaper hardware given that model runs cost on the order of $100M and every lab is paying Nvidia many multiples over that to fill up their new datacenters.
gordonhart··on ai;dr
The main coding agent failure modes I've seen:

- Proliferation of utils/helpers when there are already ones defined in the codebase. Particularly a problem for larger codebases

- Tests with bad mocks and bail-outs due to missing things in the agent's runtime environment ("I see that X isn't available, let me just stub around that...")

- Overly defensive off-happy-path handling, returning null or the semantic "empty" response when the correct behavior is to throw an exception that will be properly handled somewhere up the call chain

- Locally optimal design choices with very little "thought" given to ownership or separation of concerns

All of these can pretty quickly turn into a maintainability problem if you aren't keeping a close eye on things. But broadly I agree that line-per-line frontier LLM code is generally better than what humans write and miles better than what a stressed-out human developer with a short deadline usually produces.

gordonhart··on ai;dr
For better or worse, a lot of people seem to disagree with this, and believe that humans reading code is only necessary at the margins, similarly to debugging compiler outputs. Personally I don't believe we're there yet (and may not get there for some time) but this is where comments like GP's come from: human legibility is a secondary or tertiary concern and it's fine to give it up if the code meets its requirements and can be maintained effectively by LLMs.
gordonhart··on Officials Claim Drone Incursion Led to Shutdown of El Paso Airport
Mass panic? Think of how wildly it would be misrepresented in the media and how disruptive it would be to all air travel in the country. People aren't rational actors and the most sensationalized headline is what ends up spreading
gordonhart··on GitHub is down again
Great project, thanks for building and sharing!
gordonhart··on Claude’s C Compiler vs. GCC
The difference I see is that, after "get to orbit", the goalposts for SpaceX are things that have never been done before, whereas for LLMs the goalposts are all things that skilled humans have been able to do for decades.
gordonhart··on Claude’s C Compiler vs. GCC
He responds to a thoughtful and detailed 600-word comment from a maintainer with a dismissive "Here's the AI-written copyright analysis..." + thousands of words of slop.

The effort asymmetry is what's rude. The maintainers take their project very seriously (as they should) and are generous enough with their time to invite contribution from outsiders. Showing up and dropping 13k lines of code, posting comments copy+pasted from a chat window, and insisting that your contribution is trustworthy not because you thought it through but because you fed it through a few LLMs shows that you don't respect the maintainers' time. In other words: you are being rude. They would have to put in more upfront effort to review your contribution than you put in to create it! Then they have to maintain it in perpetuity.

gordonhart··on We mourn our craft
Personally I’m less bullish on oil as the metaphor given how much of the modern world is underpinned by cheap and ubiquitous oil. If oil disappeared tomorrow, society would collapse — if tobacco disappeared tomorrow, it would make some subset of the population very unhappy for a few weeks.

Software engineering AI API dependence seems to have already screamed past the tobacco mark but we’re still a long ways away from oil. Though I would bet that we hit it sometime in the next few decades once the bulk of the industry has never written code in any serious capacity.

gordonhart··on We mourn our craft
On a scale that would make big tobacco blush.
gordonhart··on U.S. jobs disappear at fastest January pace since great recession
Populations grow logistically as they hit carrying capacity, it just looks exponential when it’s far from K
gordonhart··on OpenClaw is what Apple intelligence should have been
Clawdbot/Moltbot/OpenClaw is so far from figuring out the “trust” element for agents that it’s baffling the OP even chose to bring it up in his argument
gordonhart··on Bunny Database
Same, it's nice to use a no-BS CDN for personal projects (e.g. https://atlasof.space/). Their pricing is good and I actually appreciate that they have no free tier so that there's no "oh shit" moment when you suddenly exceed it and owe real $$$ (looking at you, Netlify). I probably won't use their database feature but I'll for sure keep using their CDN if they can keep things as straightforward as they currently are.
gordonhart··on Qwen3-Coder-Next
By the time that happens, Opus 5 and GPT-5.5 will be out. At that point will a GPT-5.2 tier open-weights model feel "good enough"? Based on my experience with frontier models, once you get a taste of the latest and greatest it's very hard to go back to a less capable model, even if that less capable model would have been SOTA 9 months ago.
gordonhart··on xAI joins SpaceX
I'm not really interested in getting in a political mudslinging contest; you asked what's to stop a future president from doing so and I gave a practical answer.
gordonhart··on xAI joins SpaceX
SpaceX launches about 60% of National Security Space Launch payloads. The only other active launcher cleared for these missions is ULA's Vulcan Centaur, which has launched 3 times total, ~once every 8 months.
gordonhart··on Hacking Moltbook
That’s been my reaction to every Slate Star Codex/Astral Codex Ten piece I’ve read.
gordonhart··on xAI joins SpaceX
Shocked to see SpaceX buy the datacenter in space meme. Where does the power come from? Where does the heat go? Why add (high) launch costs to your buildout capex? Why add radiation as another risk factor to your already-unreliable GPUs? Am I missing something fundamental here...?
gordonhart··on How AI assistance impacts the formation of coding skills
Hinges on whether this new high level -> low level transformation becomes reliable enough to build watertight abstractions on top of it. If AI code becomes good enough that you don't have to worry about the low-level representation 99.9...% of the time, absolutely. But we're pretty far from that at the moment and it's impossible to say where things will be in another 50 years.
gordonhart··on Show HN: Moltbook – A social network for moltbots (clawdbots) to hang out
As far as I can tell, it’s a viral marketing scheme with a shitcoin attached to it. Hoping 2026 isn’t going to be an AI repeat of 2021’s NFTs…
gordonhart··on Ask HN: Notification Overload
If you’re like me and get inundated with spam texts from politicians begging for money, you can set up an app to filter texts from unknown contacts matching content patterns. I recently set up SpamHound but there are many options, and configured filters for the telltale begging signs like “text stop to quit”, “end2end”, and about 2 dozen other variants they come up with to sneak past these filters.
gordonhart··on Claude Code daily benchmarks for degradation tracking
GP was making a joke, but Anthropic could implement this if they wanted to. Not a bad metric actually if you can measure it cheaply enough.
← PreviousPage 4 of 6Next →