today I posted a show HN that got insta-flagged by an AI detector. I had written the post then gone back and forth with claude, getting to a refined version I thought was much better. claude said it would hit way better with hn (me?).
after getting flagged I found the original, reposted, and it is actually so much better. not polished, not a masterpiece, I'm nowhere close to a professional writer. But it's like a human story with the marks of human care.
i fear my brain's getting smoothed down by reading all the generation and I need to drop the AI crutch for a while, at least for copy/writing.
I wanted to share a side project we’ve been using and iterating at Hedgy for the past couple weeks.
It’s our take on a self-hosted company brain that is powered entirely by our claude subscriptions. It includes a ClickHouse data lake for ingesting data and light knowledge infrastructure for storing knowledge about the data (e.g. this is how we count a user as active, check column X when determining LTV). This is exposed to our AI’s via MCP. No inference happens in setoku itself, it’s just a data tool you give your agent. The MCP encourages the AI to record gotchas and insights as it finds them and there’s a minimal admin interface for auditing and pruning knowledge [0].
I had been pretty impressed with claude code’s data analysis abilities on my local postgres, so I was excited to ship this capability to my non-technical teammates to use from claude.ai and cowork. They thought of way more things to do with it than I did and the additional log data makes agentic debugging faster and more reliable. I honestly hooked up the log drains just to test the system with more data, but now I couldn’t go back to my log-blind claude code.
After slacking a bunch of screenshots with charts we were making, we wanted a better way to save and share them. Since the data is in the lake, we added a little protocol so that LLM’s could take a static chart or dashboard and publish it to the box as an app hooked up to the live data.
I’m running them on OVHCloud VPS’s. The Hedgy instance and demo are both running on a VPS-3 ingesting spending and account data from Mercury, Vercel logs, Render logs, a few slack channels, and Github activity.
I also deployed a family instance which is running well on a $5/mo OVHCloud VPS-1. It ingests our finances from Monarch Money (love this product!) so that Fable can give me grounded financial advice (this is not financial advice!).
Technically setoku ships as a docker image + a set of claude code skills that cover initial server setup and adding connectors.
Anyway I’m rambling, check out the tool and LMK if you have any questions/thoughts or want help setting it up :)
it's not random at all. that's literally what the resume is supposed to be and what the hiring manager wants to know! This is a real human hiring manager sharing candid feedback on his acceptance criteria. So there's at least one company where this is pretty sound advice.
your point about the dumb filters is plausible but at some point a human is going to read it and try to decide if they want to work with you. If all they have is some AI output, it's going to be an easy no.
in my experience models will answer introspective questions decisively but can't actually introspect. this is pretty similar to humans (e.g. ford's faster horses quote).
with AI stuff you can actually run real tests though.
A version of this I've been enjoying is mostly solo dev but with a biz-guy partner. I mostly just build software but have a partner who will go find out how people are using it, what they want, handle inbound, and be there to chat about ideas even if we're not getting into the technical weeds.
~was truly surprised and delighted by this today. claude's recs are usually pretty good. imo as a product engineer this is a good default and all the cranky experts in here can turn it off.~
edit: didn't read closely enough to realize this can't be disabled / think about the perverse incentives. seems bad actually
This is cool. I was skeptical of MCP's until I made one recently. They're essentially the exact same as 1) giving your agent a CLI tool or REST API and 2) pointing it there in an AGENT.md/CLAUDE.md. Agents are great at using built-for-human CLI tools and IMO they don't need anything purpose-built for agents. The key difference, which ends up being a usability win for non-technical users, is that the MCP bundles 1 and 2 - harnesses inject the MCP tool descriptions on every session after install. Of course, that's also why you need to be careful about context bloat when using/building them
can you host this yourself or do you need to use helix-cloud? the chat thing on the side seems to push me to helix-cloud but it looks like that starts at like $600/mo which is above my experimentation budget.
looking for a db for an agent memory application and i'd probably start with something that's just self-hosted / freeish. postgres is working ok but I want to start ingesting server and chat logs.
very cool. session replays are so crucial for understanding new features qualitatively, especially before you have enough users for useful quantitative metrics.
fwiw my experience building a small tech talent agency / recruiting shop disagrees with this. Cold application pipelines are overwhelmed by gen AI applications and many of the (very qualified) candidates we place report getting totally ghosted on all cold applications - even when we’re able to get them several interviews a week with companies in our network.
Seems like companies still value a curated pipeline. 15-20% of first year salary (numbers we see these days) appears to be worth saving the company time interviewing unscreened candidates. Recruiting can be a real time suck and a bad hire can be catastrophic.
> Agents can now get an email inbox by themselves. (This also means a lot of email nobody wants to read gets processed by AI instead of your inbox being cluttered with spam and slop)
Can you explain this? I would think it means the exact opposite.
i love this because it seems like you've turned vibecoding up to 11 unleashing the non-technicals in your org to ship vibecode slop straight to prod. it's an idea so obviously terrible to most engineers that maybe it's actually really really smart. much bolder than yet another AI-driven dashboarding tool or smart notebook.
to address the elephant in the room... how do you think about technical debt incurred by users who likely do not understand the underlying data models, consider auth, etc?
I built a toy BS with very dumb agents in school in 2012 or so. Multi-agent systems were kind of fringe and retro. It's nuts that modern agents look about the same (basically an OODA loop) but replacing the hand-coded "orient" and "decide" with an LLM is so much easier (for me, a frontier model user) and 100x more capable.
The use-case for multi-agent systems are intuitive from a human perspective (one person can't be an expert in everything) but a little less so with LLM's since, so far, it seems like we use a single frontier model that's pretty good at everything. That said, even a single person has little breakthroughs with different context / sleep / "mind space" so I'd guess there's some useful way to scale that idea usefully with multi-agent systems. I think a lot of these old AI ideas are rich for exploring with LLMs.
>I’ll share more details about where the Ghostty project will be moving to in the coming months. We have a plan but I'm also very much still in discussions with multiple providers (both commercial and FOSS).
what a cliff hanger!
As someone with similar warm feelings for GitHub, it's kind of sad to see the fragmentation but I have similar frustrations with the recent outages. Perhaps it's time to explore the idea of unbundling the social/discovery layer from the code hosting/dev tool so we can live between the myriad git/jj hosts but still do "social coding" together.
I'd guess the same has always been true for READMEs / human dev docs. Of course it doesn't transfer directly but still feels incredible to be in an age where we can measure such (previously) theoretical things with synthetic programmers.
Now building https://www.hedgy.works and https://setoku.com
W: https://campsh.com
T: https://twitter.com/_0_
YC Badge: 0x2363643a1bf43fe9d8fbc0630c706003b4e062bf