I use both Codex & Claude Code. With Codex, their long running agents are still not as good as Claude Code. But I don't complain because they burn very less tokens as compared to Claude Code for the same task. But having said that, if I am provided limitless tokens for each of them, my go to harness will be Claude Code
In the software engineering world, in 2026 we saw a wave of code assistant products. In 2026, we will see a wave of designing software architecture products, not just on greenfield projects but also brownfield projects.
Great approach, since you are following a pre-defined set of approaches, have you thought of building an agent for it? or do you have a prompt which helps you with getting the relevant answer?
In distributed systems, at least we have the variables, functions, pods, log traces, spans etc some pre defined structure, and some level of determinism. I would say Causality is still not fully explored territory when it comes to human brain.
When I think of human brain or may be to some extent LLMs, it's difficult to understand what is invisible. For distributed systems we will build tools, there is ongoing research in LLM Observability, but I wonder what about human brain
I come to HN because it's one of the few places left where you can sill watch people think and present new ideas. The people in the world has done intellectual outsourcing, where instead of grappling with the idea themselves, they ask LLMs to predict the next sentence, which has hollowed the real curiosity.
What makes HN valuable to me is the opposite impulse: people trying to understand things for themselves from the community and in the process, maybe, discovering new ideas which LLMs can't supply. Because what LLMs don't know they don't know, LLMs think they know(that's why they predict next sentence) but we know they don't know and still some people think they know
Sometimes, I wish, there is no thinking tax: as a reminder of why this place exists & also to reward people who are still curious and thinkers
I WANT HN TO REMAIN A PLACE WHERE HUMAN CONVERSATIONS STILL HAPPENS.
BloodHound team: blood is in your hands.
You’ve taken the name of an established security tool and attached it to what, based on your description, looks like a lightly engineered LLM-driven wrapper
First of all, thanks for the contributing to the community. This is a good observability tool to reproduce reasoning paths using structured trace of prompts. During this process, I sometimes do replay the prompts also, especially for some of the use cases where we have observed near deterministic behaviour for some standard prompts. Do you have any plans for adding this feature?
Another question: can your architecture scale to thousands of developers & millions of traces?
This article argues that the real fragility in engineering teams isn’t the bus factor, it's the loss of decision context. Most systems depend on knowledge locked in a few heads, and when that disappears, teams pay for it in outages, rewrites, and slow recovery. The post proposes a simple operating model: capturing intent through decision logs, enforcing a technical "constitution" and using automation to reduce drift, so teams can scale without accumulating invisible risk.
I would love to hear from the community how are you handling tribal knowledge within your organisation
Really interesting project, it nudges you toward learning instead of mindless feeds.
One suggestion: could you add tags to the research papers so readers can more easily filter by their interests? For example, I’m looking to follow recent work from top venues like NeurIPS specifically on training code-oriented LLMs. Tagging would make it much easier to dive into topics like that.
Really interesting idea, thanks for sharing. When working on new projects, a lot of my learning comes from studying what others have already built so I can reuse patterns around architecture, UX, and product decisions. Those learnings tend to be transferable across tech stacks, so they’re not tightly bound to a specific programming language. Could you share the thinking behind using programming language as the primary way to organize or distill these examples?