Developer of 20+ years here, can't give you an accurate multiplier but I am faster.
Because spotting holes in specs has never been one of my strengths. And working without technical colleagues much of the time, it's a boon to be able to "rubber-duck" my ideas with something that is at least more intelligent than plastic.
Grabbing multipliers from thin air, the coding bit may only be 2x faster with a poorer-quality outcome, but working out what's needed is a good 5x faster.
And yes, I'm using the same adversarial AI MO as @wood_spirit, combined with Matt Pocock's excellent /grill-me and /grill-with-docs skills [1] and Plannotator [2] to review the plans.
I actually use LLMs a lot to rubber duck my problems and help develop plans. Then I manually code, to ensure my skills don't deteriorate. I feel like I'm a lot faster, with few of the downsides. Do you have any thoughts on this process?
If you can type code fast and accurately, it sounds a great process to use. You're using LLMs for the bit where they bring great value, and yourself as a higher quality coding agent :)
So that's what it is. Reading its README I thought it was another harness like Pi [1], but with built-in memory so it remembers what it learns, and gets more capable the longer it runs.
Like Letta [2], Dirac [3][4] and the other "more experimental harnesses that look interesting but I haven't had time to try out".
What's the privacy/data security like? I can't find that on that page.
Edit: found it.
> We may use your Content to operate, maintain, improve, and develop the Services, to comply with legal obligations, to enforce our policies, and to ensure security. You may opt out of allowing your Content to be used for model improvement and research purposes by contacting us at membership@moonshot.ai. We will honor your choice in accordance with applicable law.
Yup, they train on your inputs and OpenRouter is complicit by claiming that Moonshot's ToS says that they don't. Contacted OpenRouter about this a while ago and was met with silence because it's bad for their business to stop lying about it.
> Stripe APIs being simple and easy is a meme from the 2010s. It isn't anymore.
I'm working with Stripe subscriptions at the moment for a charity taking donations via their website. The subtle differences between subscriptions done through Stripe checkout and subscriptions set up yourself using Stripe elements are by turn infuriating and frustrating.
The documentation is geared towards people using checkout. Stripe's own AI help could find us a bit of information which going through the documentation didn't give us, and it even struggled to find the reference in the docs for it.
One product, two different ways to use it, and slightly diverging feature sets between the two. Argh!
Or more cynically they reach their level of competence, go one level further and stay there to keep them from ruining the productivity of the people doing the work...
Dealing with Google is a nightmare. I'm one of the volunteer sysadmins for https://forum.buildhub.org.uk/, a DIY and self-build forum. For 10 years it ranked very well on Google, particularly in the UK, and then on 28 December 2025 it disappeared from Google's index.
Nothing has helped, the Google forums are tumbleweed and there's no one to reach out to for what could be an algorithm change or something gone wrong. I'm a paying Workspace customer and it's made me think I need a backup plan in case I'm ever suspended. Reports like this don't encourage.
> Nothing has helped, the Google forums are tumbleweed and there's no one to reach out to for what could be an algorithm change or something gone wrong.
The own-brand forum (Google, Microsoft, Apple) seem to be infested by netizens from lower-income countries trying to build online customer support portfolios by providing utterly useless answers.
That, or trying to game the system and getting shortlisted for a free trip to Google HQ for one of their contributor summits.
I am genuinely curious if anybody knows of a non-trivial problem being solved on one of these forums, at least for a huge company that’s palming off customer support. It just feels like screaming in to the void, only for someone to (deliberately?) misinterpret your question and give you some generic advice.
It depends on what you call "non-trivial". I found answers on how to circumvent dumb macos bugs on Apple forums at least twice in the last 6 months. One related to displays, I was about to return a new USB-C monitor which wouldn't turn on. A silly issue, but it's a bug on my book, I wouldn't find the answer on the docs.
That counts! I suppose I’m lucky enough to know of more reliable resources (macadmins.org Slack is an excellent community), and so I turn to them after reading more than a couple of threads on the Apple Support Community. Perhaps it has improved or I never dig deep enough.
I’d be at a complete loss for any obscure Windows issue though.
Might be something specific to my and my colleagues' systems, but it breaks the TUI. It needs git authentication, which fails, and the TUI stops accepting input reliably
Can anyone enlighten me how having a coding harness when for most customers you say "we won't train on your code" helps you do RL? What's the data that they rely on? Is it the prompts and their responses?
It doesn't matter what your privacy setting is, with any savvy vendor. Your data is used to train by paraphrasing it, and the paraphrasing makes it impossible to prove it was your data (it is stored at rest paraphrased). Of course the paraphrasing stores all the salient information, like your goals and guidance to the bot to the answer, even if it has no PII.
That's an interesting accusation there! You're essentially accusing every "savvy vendor" of large-scale fraud... DOn't suppose you'd have any actual citations or evidence to back that up?
Eg, When a prompt had a bad result and was edited, or had lots of back and forth to correct tool usage that information can be distilled and used to improve models.
And now imagine if you are focused on this for weeks you can likely come up with other ideas to leverage the metadata to improve model performance.
Because spotting holes in specs has never been one of my strengths. And working without technical colleagues much of the time, it's a boon to be able to "rubber-duck" my ideas with something that is at least more intelligent than plastic.
Grabbing multipliers from thin air, the coding bit may only be 2x faster with a poorer-quality outcome, but working out what's needed is a good 5x faster.
And yes, I'm using the same adversarial AI MO as @wood_spirit, combined with Matt Pocock's excellent /grill-me and /grill-with-docs skills [1] and Plannotator [2] to review the plans.
1. https://github.com/mattpocock/skills
2. https://github.com/backnotprop/plannotator
reply