Yes. Three complex websites in production. Basically 99,99% of code written by AI.
One of them is a google-like vertical web crawler, parser and indexer, with all the complexity and challenges that come with it.
AI coding works.
Ask HN: Do you have any evidence that agentic coding works?
351–360 of 478 posts
Re: Ask HN: Do you have any evidence that agentic coding works?
#352Earlier quoted context omitted.
A principal engineer at Google made a public post on the World Wide Web and assumed some shared Google/Claude-context. Do you hear yourself?
Working in a large scale org gets you accustomed to general problems in decision making that aren’t that obvious. Like I totally understood what she means and in my head nodded with “yeah that tracks”.
Maybe it helps them sleep at night.
Re: Ask HN: Do you have any evidence that agentic coding works?
#353Does it work? Absolutely. It just depends how you use it and what for.
I had 12 legacy node apps running node 4, with jQuery front ends. Outdated dependencies and best practices all over. No build pipeline. No tests. Express 3. All of it worked but it's aging to the point of no return. And the upgrade work is boring, with very little ROI.
In a month, without writing any code, I've got them all upgraded to Node 22, with updated dependencies, removed jQuery completely, newer version of express, better logging, improved UI.
It's work that would have taken me a year of my free time and been soul crushing and annoying.
Did it with codex as a way of better learning the tooling. It felt more like playing a resource sim game than coding. Pretty enjoyable. Was able to work on multiple tasks at once while doing some other work.
It worked really well for that.
Re: Ask HN: Do you have any evidence that agentic coding works?
#354I use Augment with Claud Opus 4.5 every day at my job. I barely ever write code by hand anymore. I don't blindly accept the code that it writes, I iterate with it. We review code at my work. I have absolutely found a lot of benefit from my tools. I've implemented several medium-scale projects that I anticipate would have taken 1-2 weeks manually, and took a day or so using agentic tools. A few very concrete advantage…
I wouldn't consider the proposed workflow agentic. When you review each step, give feedback after each step, it's simply development with LLMs.
Interesting. What would make the workflow "agentic" in your mind? The AI implementing the task fully autonomously, never getting any human feedback?
To me "agentic" in this context essentially that the LLM has the ability to operate autonomously, so execute tools on my behalf, etc. So for example my coding agents will often run unit tests, run code generation tools, etc. I've even used my agents to fix issues with git pre-commit hooks, in which case they've operated in a loop, repeatedly trying to check in code and fixing errors they see in the output.
So in that sense they are theoretically capable of one-shot implementing any task I set them to, their quality is just not good enough yet to trust them to. But maybe you mean something different?
Re: Ask HN: Do you have any evidence that agentic coding works?
#355I use Augment with Claud Opus 4.5 every day at my job. I barely ever write code by hand anymore. I don't blindly accept the code that it writes, I iterate with it. We review code at my work. I have absolutely found a lot of benefit from my tools. I've implemented several medium-scale projects that I anticipate would have taken 1-2 weeks manually, and took a day or so using agentic tools. A few very concrete advantage…
1. And 2. I.e. creating a spec which is the source of truth (or spec driven development) is key to getting anything production grade from our experience.
Yes. This was the key thing I learned that let me set the agents loose on larger tasks. Before I started iterating on specs with them, I mostly had them doing very small scale, refactor-this-function style tasks.
The other advice I've read that I haven't yet internalized as much is to use an "adversarial" approach with the LLMs: i.e. give them a rigid framework that they have to code against. So, e.g., generate tests that the code has to work against, or sample output that the code has to perfectly match. My agents do write tests as part of their work, and I use them to verify correctness, but I haven't updated my flow to emphasize that the agents should start with those, and iterate on them before working on the main implementation.
Re: Ask HN: Do you have any evidence that agentic coding works?
#356well I made a machine to help my machine's work on their machine building through my knowledge management machine while being prioritized from another machine
but the machine's keep going down, and we haven't cleaned enough of the bugs in the meta machine to open source it yet.
so...kinda?
Re: Ask HN: Do you have any evidence that agentic coding works?
#357Earlier quoted context omitted.
> which obviously is too little for me to compromise my ethics and morals What would be enough to compromise your ethics and morals? I'm sure they can accommodate.
Hah, after submitting my comment, I actually though about it because I knew someone would eventually ask :) I'm fortunate enough to live a very comfortable life after working myself to death, so I think for 20,000,000 USD I'd do it, happily so. 2,000,000 would be too little. So probably between those sit the real price to purchase my morals and ethics :)
It wasn't a shot at you personally but the point of this was that AI companies are flush with money and desperate to show any kind of growth and willing to spend money to do that. I'm sure they are finding people that have some social following and will happily pocket couple extra green bills to present AI products in a positive light with little to no actual proof.
Re: Ask HN: Do you have any evidence that agentic coding works?
#358Earlier quoted context omitted.
JavaScript isn't the only programming language around. I'm not the strongest around with JS either but I can figure it out as necessary -- knowing C/C++/Java/whatever means you can still grok "this looks better than that" for most cases.
I don't think so. Imagine it was vice versa, someone saying they knew JS and were weak at C/C++/Java.
This doesn't sound right to me. If someone who were expert in JS looked at a relatively simple C++ program, I think they could reasonably well tell if the quality of code were good or not. They wouldn't be able to, e.g., detect bugs from default value initialization, memory leaks, etc. But so long as the code didn't do any crazy templating stuff they'd be able to analyze it at a rough "this algorithm seems sensible" level".
Analogously I'm quite proficient at C++, and I can easily look at a small JS program and tell if it's sensible. But if you give me even a simple React app I wouldn't be able to understand it without a lot of effort (I've had this experience...)
I agree with your broad point: C/C++/Java are certainly much more complex than JS and I would expect someone expert in them to have a much easier time picking up JS than the reverse. But given very high overlap in syntax between the four I think anyone who's proficient in one can grok the basics of the others.
Re: Ask HN: Do you have any evidence that agentic coding works?
#359A principal engineer at Google posted on Twitter that Claude Code did in an hour what the team couldn’t do in a year. Two days later, after people freaked out, context was added. The team built multiple versions in that year, each had its trade offs. All that context was given to the AI and it was able to produce a “toy” version. I can only assume it had similar trade offs. https://xcancel.com/rakyll/status/200765974…
[flagged]
I’ve mostly used the 20 a month cursor plan and I’ve gotten to the point I can code huge things with rarely the need to do anything manually
Re: Ask HN: Do you have any evidence that agentic coding works?
#360A principal engineer at Google posted on Twitter that Claude Code did in an hour what the team couldn’t do in a year. Two days later, after people freaked out, context was added. The team built multiple versions in that year, each had its trade offs. All that context was given to the AI and it was able to produce a “toy” version. I can only assume it had similar trade offs. https://xcancel.com/rakyll/status/200765974…
[flagged]
I’ve mostly used the 20 a month cursor plan and I’ve gotten to the point I can code huge things with rarely the need to do anything manually