We just did this the other week and it's such a great setup using AI. Monorepos in general are better for coding agents since it's a single location to search. But now we have the ability to say "Add xyz optional param to our API" and claude adds the code + updates the documentation. I was also able to quickly ask "look at our API and our docs, find anything out of date".
Our set up is:
packages/
↳ server
↳ app
↳ docs
Using mintlify for the docs, just points to the markdown files in the docs folder. And then a line in the claude.md to always check /docs for updates after adding new code.
The one thing I hate about monorepos is nothing ever gets versioned, packaged, and shipped.
Polyrepos are workable, the way to do it is to actually version, ship, and document every subcomponent. When I mean ship, I really mean ship, as in a .deb package or python wheel with a version number, not a commit hash. AI can work with this as well, as long as it has access to the docs (which can also be AI-generated).
The monorepo make it easier to ship the overall product but harder to ship parts of it.
I've used a monorepo for the past 13 years and I got all shared packages with version 0.0.0 and I still haven't figured out a simple way to share just some parts of it like a CLI.
Does anyone have a monorepo and publishes NPM packages with source code of only that folder? Sub-gits required to pull in multiple places...
The best thing about monorepos is nothing ever gets versioned and packaged.
That means, a subcomponent can just make a needed change in the supercomponent as well, and test and the ship the subcomponent without excess ceremonies and releases.
That strategy sucks when you have 15 supercomponents depending on 1 subcomponent, you want to rewrite the subcomponent, and now you have to deal with all 15 of them and get codeowners approvals from all 15 of those teams.
In a polyrepo org you just release version 2.0 of the subcomponent and upgrade the 1-2 of the 15 teams where the 2.0 features are actually needed now. The rest of the users can upgrade on their own schedule.
I've got about ~15 repos for a project and I just start Claude Code in the parent directory of all of them, so it has clear visibility everything and cross-reference whatever it needs.... super handy.
Yes it's awesome! I'm creating a lot of CLIs with Claude Code to interact with external services. Yesterday made a CLI for the Google Search Console so I can prompt "get all problems from indexing in Google Search Console and fix them".
Same with Sentry bugs. Same with the customer support "Use the the customer support cli skill to get recent conversations from customers and rank bug reporting and features requests and suggest things to work on"
Sentry MCP is great, “find out top 10 issues by users affected, check what it would take to fix and if you think it’s a low risk fix, apply it. Open a PR that links to the issues and explain the issue and the fix in three sentences max”.
In terms of "[XYZ] for agents", I think CRM is a big one that people haven't talked about as much. It becomes super relevant as soon as people start using an agent for anything customer related.
And the design principals are already pretty well established (accounts, contacts, leads, opportunities, custom object model, stages, etc.). It just needs to be turned into a database boilerplate with a bunch of agent tools. Excited to try this out.
We’re building the AI-powered infrastructure layer for small-business lending. Banks and fintechs use Omni to automate document collection, financial modeling, public-record research, and ongoing borrower communication.
I understand the pitch, and the extra value add of having some existing relationship with the hotels / etc.
But the quality of the actual AI response is just worse than GPT 5.2. Which makes it feel like a tacked on thing and more of a gpt wrapper.
I asked about a retreat with our US team that could also include one engineer in Pakistan that needs a visa. And the response was something to the effect of: "Assuming your engineer has a US visa, you can go to Puerto Rico".
Whereas chatgpt gave a much more well researched answer.
This is interesting thank you for reporting that. I am goi t o investigate that use case deeper. The models at core are Gemini / ChatGPT / claude so you should get a similar level answer or better (because of our data) not worse. I think in the VISA, we may have to do a specific training and setup of the system prompt.
Just so I know, you asked the exact same thing in both ChatGPT and our product? no changes?
Maybe the algorithm was too pushy on venue selection and not enough focused on the understanding of your problem.
Similar for me. I asked for a retreat in Lima, Peru, and got one low star option in Lima and then places in Ohio, Mexico and Colombia... ChatGPT got it on first try
Yes you are right, we have a lack of supply there, and we do not show places that we are not able to Book. Chat GPT is showing data, we are showing bookable venue. But I agree that in Peru our supply is too limited so the experience is not ideal.
Thank you for reporting, I appreciate the time spent
> If it were me, I would have avoided involving third party regulators in the initial contact at least.
I'm surprised to see this take only mentioned once in this thread. I think people here are not aware of the sheer amount of fraud in the "bug bounty" space. As soon as you have a public product you get at least 1 of these attempts per week of someone trying to shake you down for a disclosure that they'll disclose after you pay them something. Typically you just report them as spam and move on.
But if I got one that had some credible evidence of them reporting me to a government agency already, I'd immediately get a lawyer to send a cease and desist.
It seems like OP was trying to be a by the book law abiding citizen, but the sheer amount of fraud in this space makes it really hard to tell the difference from a cold email.
Anthropic regularly publishes research papers on the subject and details different methods they use to prevent misalignment/jailbreaks/etc. And it's not even about fear of being sued, but needing to deliver some level of resilience and stability for real enterprise use cases. I think there's a pretty clear profit incentive for safer models.
Anthropic is investing, conservatively, $100+ billion in AI infrastructure and development. A 20-person research team could put out several papers a year. That would cost them what, $5 million a year, or one half of one percent? They don't have to spend much to get that kind of output.
Not to be cynical about it BUT a few safety papers a year with proper support is totally within the capabilities of a single PhD student and it costs about 100-150k to fund them through a university. Not saying that’s what Anthropocene does, I’m just saying chump change for those companies.
Sometimes I think people misunderstand how hard of problem AI safety actually is. It's politics and mathematics wrapped up in a black box of interactions we barely understand.
More so we train them on human behavior and humans have a lot of rather unstable behaviors.
> You are very off (unfortunately) about how little PhD students are being paid
All in costs for a PhD student include university overheads & tuition fees. The total probably doesn't hit $150k but is 2-3x the stipend that the student is receiving.
Someone currently working in academia might have current figures to hand.
Worth mentioning that numbers for the US are unlikely to be representative when discussing it as a whole, though might be relevant to this specific case.
In the UK the all in cost of a PhD student starts somewhere around £45k once you include overheads I believe. If you need expensive lab support then it probably goes up from there.
So about $75k for the bottom end? The quoted numbers sound about right in PPP terms in that case.
> Github doesn't show timestamps in the UI, but they do in the HTML.
Unrelated tip for you: `title` attributes are generally shown as a mouseover tooltip, which is the case here. It's a very common practice to put the precise timestamp on any relative time in a title attribute, not just on Github.
Unfortunately title isn't visible on mobile. Extremely annoying to see a post that says "last month" and want to know if it was 7 weeks ago or 5 weeks ago. Some sites show title text when you tap the text, other sites the date is a canonical link to the comment. Other sites it's not actually a title at all l but alt text or abbr or other property.
> If it was really an autonomous agent it wouldn't have taken five hours to type a message and post a blog. Would have been less than 5 minutes.
Depends on if they hit their Claude Code limit, and its just running on some goofy Claude Code loop, or it has a bunch of things queued up, but yeah I am like 70% there was SOME human involvement, maybe a "guiding hand" that wanted the model to do the interaction.
> Heroku remains an actively supported, production-ready platform, with an emphasis on maintaining quality and operational excellence rather than introducing new features [1]
This sentence is what really seals the death for me. I used to be a big Heroku fan. And used them as late as 2023. But tbh it very quickly fell behind the capabilities and devex of products like Supabase and Vercel.
While I agree that it will probably stick around in zombie mode for another decade, if Salesforce doesn't want to improve the product, it will just slowly bleed users until the cost to maintain it is less than the revenue.
Honored to be included, working at Bitscribe actually got me to sign up to HN which subsequently, between working for YC companies and the HN Hiring posts, has driven a significant majority of my career, so it all kind of comes full circle.
I may predate you! We were shared an office with Heroku back during their/our YC days.
Fun memory - James and bitscribe helped me with my prior startup. I remember them brainstorming a 'collaborative IDE' while they helped us set up our servers.
You may have been at bitscribe at the time with pedro and morton?
Heroku <feedback@heroku.com>
Tue, Nov 6, 2007, 1:03 PM
to jason
Hello -
You've been invited to the Heroku beta by your friend james@heroku.com.
They included this message:
--------------------------------------------------
So we're up and running, and can officially talk about our YC funding now. You are probably outside of our audience, but feel free to kick it around and send me any feedback you have. Going to invite Colin and the Bracy's too.
I got you beat, Oct 25, 2007 - the original post I mentioned that they had a bug (a before_filter for authentication on the invite route) so it was unusable before they fixed it based on my report - a classic rails error! Vintage stuff, great nostalgia.
> You may have been at bitscribe at the time with pedro and morton?
It's been so long I had to go back and find my prior post, but I think so!
We’re building the AI-powered infrastructure layer for small-business lending. Banks and fintechs use Omni to automate document collection, financial modeling, public-record research, and ongoing borrower communication.
Our set up is:
Using mintlify for the docs, just points to the markdown files in the docs folder. And then a line in the claude.md to always check /docs for updates after adding new code.