Hey HN, I build custom AI agents for businesses, and one of the things I've noticed is that everyone wants to use AI to do real, critical work, but it's hard to trust it because sometimes agents can go a little off the rails. No one wants their agent deleting all their customer data or screwing up their books. So they tend to limit its use to searching, summarizing, and as a writing assistant. The problem is that most ways we store data, from notes & spreadsheets to databases, were designed fo...
I’m an architect at my company and everyone on my team is now using Claude/Codex agents for their work. Managing context in claude.md/agents.md files is fine but so much of the knowledge required doesn’t fit cleanly in one repo or the other. Plus, much of the team (product, design, support) does not work in repos but still needs a place to share context. I searched for existing solutions to this but only found products meant for managing deployed AI agent memory for use in AI applications. I wan...
ok this is kinda funny. aisdr and valley are both ai sdr tools, and aisdr just teased signing something with valley. so the bots built to book meetings are now buying each other. the category is like 2 years old and already consolidating, if this is even real.
I have seen a lot of different ways of implementing long term memory for agents and curious to know which mechanism works better for different usecases and what infrastructure are people using. For me the architecture that has been most robust but more token heavy is agent as memory controller and is built on neon postgres - agent has tools to add memory entries with tools (save, list, update, delete, search) represented as heirarchial directory memories/topic/sub-topic.md backed by serverless...
Training a video model to prefer physically plausible results is not the same as giving it a simulator. The distinction matters whenever a model is described as understanding the physical world. LingBot-Video uses a reward system that includes physical rationality and task completion alongside more familiar video criteria. That can push training away from obviously impossible motion. It does not introduce explicit mass, friction, collision geometry, or a numerical integrator. The model is stil...
Two things happened this week that are worth paying attention to if you work in tech or run a business. On July 16 the European Commission issued binding orders under the Digital Markets Act requiring Google to give rival AI assistants the same system-level Android access it reserves for Gemini. Right now if you install ChatGPT or Claude on an Android phone, you get an app. Gemini gets to hear a wake word, hold the home button, read your screen, and act inside other apps. That gap is now illeg...
saw a stat this morning that's been bugging me all day. 47% of small businesses are using AI somewhere in HR now, screening resumes, onboarding, all that. fine whatever, expected at this point but then i saw the other half of it. more than half of applicants are using AI to write their resumes and cover letters too. linkedin is apparently getting like 11,000 applications a minute right now which is insane to even think about so just sit with that for a sec. candidate uses AI to write the res...
Before the World Cup Final, I tested an AI-based structured analysis model on Spain vs Argentina. The goal was not to predict the winner based on rankings, media opinions, or betting odds. Instead, the model analyzed the match from three angles: Spain’s own match structure Argentina’s possible winning path External factors that could change the normal game flow The prediction was: - Spain had the more stable championship structure - Argentina’s path depended more on ada...
I've been building an app that lets students snap a photo of an assignment, a whiteboard, a printed worksheet, whatever, and turns it into a structured task with subject, due date, and estimated effort. Want to admit that I went in underestimating how hard this would be from an actual parsing pipeline POV. The pipeline is roughly: photo in → Claude's vision API reads the image → a prompt asks it to extract specific structured fields (title, subject, due date, estimated effort) → returned as JS...
I use AI for research almost every day now, and it's amazing for getting a quick overview of a topic. But I've also noticed that once the questions become really specific, you eventually need input from someone who's actually done the work. That's especially true for industries where small details can completely change a decision. While reading about how companies solve that problem, I found Expert Network and thought the idea was pretty interesting. Instead of relying only on reports or public ...
I wrote this article as a bit of an experiment. Specifically, I'm rejecting the purist view that writers should write without AI and avoid it to remain authentic. (This view is particularly common on LinkedIn, where people are concerned about losing their job to AI writers and AI-enhanced writing.) I think that human writers have a completely different challenge that matches the theme of this subreddit: AIs are a new intelligence and audience for all of us. And it happens that as they ingest i...
Assume a future where every human has a sovereign personal AI. Not a government AI. Not a party AI. Not a platform algorithm. A personal AI controlled by the citizen, built around their values, preferences, history, constraints, priorities, and decision patterns. An AI that knows its human well enough to say: “Based on what I know about you, you would probably care about this issue, lean toward this option, object to this trade-off, and want to be notified before any decision is made...
An OpenAI-compatible API generator backed by Cloudflare Workers AI — chat, speech, vision, embeddings — deployed to your account in one click and with a testing dashboard. No login, just need a Cloudflare account and a single-use CF token to deploy. I'm a big fan of testing the latest models deployed on Cloudflare, and the 10,000 "neuron" allotment per day is pretty generous. I built this to supplement a side project I've been working on for the last six months. https://notebooker.ai , which ha...
Application Signal is a platform that collects every YC-backed startup from 2020 to 2026, embeds them into a vector database, and visualizes them on an interactive map. It also provide an AI-powered analysis service. Users can generate a detailed report for their startup idea using credits.
Comments URL: https://news.ycombinator.com/item?id=48993394
Points: 1
# Comments: 1
