Dev Systems
Does your AI stack need a session layer? A maturity framework for teams building AI agents
Most teams building AI agents start with HTTP streaming. It's the right starting point. Every major agent framework defaults to it, it gets tokens on screen fast, and for a single-user prompt-response interaction it works well.The question is when it stops being enough - and how to recognise that before it turns into user experience problems, engineering waste, and technical debt that constrains what your product can do.Over the past year, I've spoken with over 40 leading AI companies building p
Why AI support fails in production: The infrastructure problem behind every incident
HTTP streaming – the default transport underneath every major agent framework – was never designed for sessions that survive a tab close or hand off cleanly between participants. Two failures surface consistently in production CX products because of this. Both generate support tickets about conversation state and prompt quality. Both trace to the transport layer.The scenario that illustrates them: a customer contacts support about an order that's partially shipped and partially stuck. The AI age
AI for American-Produced Cement and Concrete
Meta is continuing its long-term roadmap to help the construction industry leverage AI to produce high-quality and more sustainable concrete mixes, as well as those exclusively produced in the United States. Concurrent with the 2026 American Concrete Institute (ACI) Spring Convention, Meta is releasing a new AI model for designing concrete mixes – Bayesian Optimization for Concrete (BOxCrete), as well as the foundational data used to develop award-winning concrete mixes.Meta’s open source model
How Aigen transformed agricultural robotics for sustainable farming with Amazon SageMaker AI
This post is cowritten with Yuri Brigance, and Usman M. Khan from Aigen. Aigen builds autonomous robots designed to help farmers remove herbicide-resistant weeds and improve crop yield through AI-driven technology. These robots operate without chemicals, using renewable energy, and provide real-time, field-level data to enhance decision-making. Using advanced computer vision AI, Aigen’s robots autonomously identify and remove weeds without harming crops, giving farmers an eco-friendly, cost-effe
Stateful agents, stateless infrastructure: the transport gap AI teams are patching by hand
Every major layer of the AI stack now has a name. Model providers - OpenAI, Anthropic, Google - handle inference. Agent frameworks - Vercel AI SDK, LangGraph, CrewAI - handle orchestration. Durable execution platforms like Temporal make backend workflows crash-proof.Between the agent and the user's device, there is nothing comparable. The stateful connection through which the agent and user actually interact - the session itself - has no dedicated layer, no recognised category, and no purpose-bu
Show HN: Anvil – Desktop App for Spec Driven Development
Very excited to share Anvil. I built Anvil to take back control when working with parallel coding agents. It comes with one click worktree isolation, and first class spec support.Claude Code and similar coding TUIs are very eager to get into writing code, even before their human baby sitter fully understands the implication of what they are about to build.The core insight with Anvil is that it is much easier to write high quality code which matches the author's intent after iterating on an
Show HN: Shoofly – pre-execution security for Claude Code Cowork and OpenClaw
Anthropic says on their safety page: "these filters are not a security boundary." Snyk found 36% of ClawHub skills contain security flaws. Trend Micro documented malware being distributed through ClawHub.These agents have shell access, file access, and connected accounts. We built Shoofly to sit in front of tool calls before they fire.- PreToolUse / PostToolUse hooks intercept every tool call
- Blocks prompt injection, credential theft, unauthorized writes, malware in tool results
Show HN_Mnemosyne: A 10^38 Joule Thermodynamic Barrier for Post-Quantum Edge AI
Mnemosyne: A blueprint for computational pluralism via physical laws.-Security: Establishes a 10^38 Joule physical floor (kappa >= 804 bits) derived from Landauer’s Principle and the Margolus-Levitin limit.Infrastructure: Targets 2GB RAM edge devices using BFT-MESI cache coherence.-Verification: Formally specified in TLA+; preliminary TLC verification completed for minimal configurations (n=4, f=1).-Economics: Proof of Useful Work (PoUW) for socially beneficial AI compute.-Current Openings:Se
Show HN: Book Grounded AI Learning
I constantly ask Gemini/ChatGPT/Claude about books on a certain topic, then ask the llm to give me a summary of the book its chapters - this helps me:
1. keep the llm grounded as a real expert (instead of saying pretend to be an engineering expert, give it a distributed systems book and then ask opinions on it)
2. helps me discover and buy books! Once i go through book concepts and chat with llm about it, i'm more convinced to buy it and dig deeper on low level on itSo i built an
How Google Does It: An inside look at cybersecurity
<img src="https://storage.googleapis.com/gweb-uniblog-publish-prod/images/How-Google-Does-It---Keyword-he.max-600x600.format-webp.webp">Learn how Google approaches some of today's most pressing security topics, challenges and concerns, straight from Google experts.
MLB pitches AI-powered commentary in its play-by-play app
<img src="https://storage.googleapis.com/gweb-uniblog-publish-prod/images/AIMLBPitches-hero.max-600x600.format-webp.webp">MLB Scout Insights feature, powered by Gemini and Google Cloud AI, provides baseball commentary.
Ask HN: Running legacy IE/ActiveX clients without local admin rights?
We are currently maintaining a very old client-server architecture. The server collects real-time data from a large number of sensors and controllers, transmitting it to a legacy database under continuous, massive load (writes every few seconds).The problem is the client side. It’s ancient, strictly requires Internet Explorer, and heavily relies on ActiveX. If a standard domain user launches the browser, the data fails to load and the browser completely hangs. It only functions correctly if run
Ask HN: Building a deterministic AI substrate on legacy hardware
I have been working on this in isolation for the last 7+ years. I have reached the absolute limit of what I can do as a solo architect without "production plumbing." I am at a survival wall and am looking for a bridge and technical partners to move this from a research PoC to a production-grade engine. I am developing a deterministic alternative to the current probabilistic "Token Economy."The Asset: A zero-inference, symbolic AI substrate. It is not an LLM wrapper. It is a l
New Open Source from Non-Traditional Builder
Let me begin by saying that I am not a traditional builder with a traditional background. From the onset of this endeavor until today it has just been me, my laptop, and my ideas I learned how systems work through trial and error, and I built these platforms because after an exhaustive search I discovered a need. I am fully aware that a 54 year old fantasy novelist with no formal training creating one experimental platform, let alone three, in his kitchen, on a commercial grade Dell stretches cr
Show HN: Arxitect – Claude Code plugin for software design principles
Show HN: Arxitect – Claude Code plugin for software design principlesModern coding agents are getting exceptionally good at implementing a given coding task. And with validation-in-the-loop, you can be reasonably confident they will implement a correct solution. However, their implementation often leaves a lot to be desired. It doesn't adhere to the decades of software design best-practices that the community has established and is often myopic to broader software quality attributes like ma
Ask HN: Is DLSS 4 Multi Frame Generation on RTX 4000 a HW or SW Lock?
NVIDIA markets Multi Frame Generation 3x/4x as an RTX 5000 exclusive feature.
But is this actually a hardware limitation, or is it software-enforced?<p>I'm curious if anyone has looked into this. Specifically:
- Whether Ada Lovelace architecture is technically capable of MFG 3x/4x
- Whether the limitation lives in the driver or the hardware itself
- Any prior research on nvlddmkm.sys or DLSS internals
Architecting for agentic AI development on AWS
If you’re architecting cloud systems for AI development on AWS, you’ve likely discovered that traditional architectures create friction for AI agents. Many cloud teams are experimenting with AI coding assistants but quickly discover a gap between what these tools promise and what their architectures allow. When an AI agent generates code, it often takes minutes—or hours—before you can validate whether that change actually works. Slow deployment cycles, tightly coupled services, and opaque code b
GenAI-based development platform - part 3: Announcing Isolarium, three flavors of secure sandboxes for GenAI-based coding agents
I’m pleased to announce that I’ve open-sourced Isolarium, a companion project to Idea to Code workflow that provides secure sandboxes for running GenAI-based coding agents like Claude Code.This article is part of a series about the GenAI-based development platform (a.k.a. harness) that I’ve been developing to make GenAI-based coding agents like Claude Code more productive, more secure and less frustrating.The complete list of articles in the series is as follows: Part 0 - My GenAI development w
LiveObjects now available: shared state without the infrastructure overhead
Shared state is a hard problem. Not hard in the abstract, computer-science sense (the concepts are well understood). Hard in the someone has to actually build this sense, where every team that wants a live leaderboard, a shared config panel, or a poll that updates in real time ends up reinventing the same wheels: conflict resolution, reconnection handling, state recovery.Most teams do not want to spend their time building and maintaining that layer. They want to ship the feature that depends on
Show HN: Running AI agents across environments needs a proper solution
Hi HN folks,I have been building AI agents for quite some time now. The shift has gone from LLM + Tools → LLM Workflows → Agent + Tools + Memory, and now we are finally seeing true agency emerge: agents as systems composed of tools, command-line access, fine-grained system capabilities, and memory.This way of building agents is powerful, and I believe it is here to stay. But the real question is: are the systems powering these agents ready for that future?I do not think so.Using Docker for a sin