Everything in this world is relative. In our case, it was more like $10 or more, but yeah, I can see how it could go either way depending on the niche and intent.
I’ve been playing a lot with durable agents, can’t say that I know what the final thing should look like but this could be a part of it. For now I have celld in front of sandbox, if I get this right: agent -> dasp -> celld -> sandbox
atm just trying to collect them all. the problem itself is super non-trivial imho, and my feel is that AI empowerment is getting more people to build these systems just because they can. some of the solutions are really interesting, so it could also be a learning resource or we could figure out after some time what happened and if they have really made meaningful impact on those companies
Claude and I have collected here every public mention of “internal agents” which have been popping up a lot recently. This was quite interesting for me to understand how different or similar these approaches are. I have dumped all that I’ve stumbled on x and here but I have prob missed some. I had run deep researchers few times with less luck, as mostly these things are still not well documented publicly. I feel the ownership of agents would become tremendously important in the coming years.
Idk, spending 10x less time dictating a thought to a coding agent and publishing it immediately is incredibly valuable. Take it at face value and don’t overthink the style.
Yes, this is frustrating, but it doesn’t occur in CC. I run the conversation logs through an agent and opencode source, and it identified an issue in the reasoning implementation of opencode for Zai models. Consequently, I ceased my research and opted to use CC instead.
This wqw an interesting postmortem for me. As a regular user, February definitely felt a bit shakier than usual.
I ran into the unicorn error page a few times and had intermittent issues with PR pages and bunch of weirdness. Nothing long-lived, but enough transient failures that my overall feel was it is degrading in general.
was listening to music while coming with groceries and simultaneously juggling stuff to open the doors and change the track with Siri (the only use for Siri I have)
FWIW I work at Steel (not the OP). While we’ve been iterating on the “right shape” for agent tooling, I’ve been building a benchmark harness to measure how different surfaces affect real web task completion: raw API context, CLI-only, opinionated “skills” (structured outputs + artifact capture), and combinations.
If you’ve run agents on the open web, I’d love suggestions for nasty-but-representative workflows to include in the benchmark.
This rings true, as I’ve noticed that with every new model update, I’m leaving behind full workflows I’ve built. The article is really great, and I do admire the system, even if it is overengineered in places, but it already reads like last quarter’s workflow. Now letting Codex 5.3 xhigh chug for 30 minutes on my super long dictated prompt seems to do the trick. And I’m hearing 5.4 is meaningfully better model. Also for fully autonomous scaffolding of new projects towards the first prototype I have my own version of a very simple Ralph loop that gets feed gpt-pro super spec file.
For no special reason, beside I could, I’ve slop coded this AI agents ephemeral VM orchestrator which I use inside any agent to manipulate and maintain my coding VMs on Proxmox. Probably it could make sense to simplify it further and move from Proxmox to something like this. Link: https://github.com/nibzard/agentlab
This is exciting. But I had to read and check everything twice to figure it out, as some already commented. Strong Feedback loop is an ultimate unlock for AI agents and having twins is exactly the right approach.
For sure! Just ask enough times "why" and you will find the root. The main issue here it is, how many people do that for real, and how this is becoming even more critical now.
Reproducibility is a fascinating topic for me, and today with AI coding agents we could have automated reproducibility at least in some fields. The concept they touch on in the paper, of post publication verification could replace or add onto existing research valorization.
Strong point. I’m considering to tag patterns better and add stuff like “model/toolchain-specific,” and something like “last validated (month/year)” field. Things change fast and for example “Context anxiety” is likely less relevant and should be reframed that way (or retired).
Author here (nibzard). I started this back in May as a personal learning log. I agree with the skepticism about jargon and novelty. However, if something reads like overly complex common sense, that’s a bug, and I’d like to fix it. If you can point out 1–2 specific pages that feel sloppy or unactionable, I’ll rewrite them (or remove them). I’m also happy to add flags or improve the structure. Also, contributing new patterns would be grand. Of course, some or even all patterns are explicitly “emerging.”
At some point, we need to begin. My initial thought was that this is a growing and evolving resource, primarily for my own use. We are slowly but steadily learning what makes sense annd patterns emerge. Also, if others find it interesting and contribute, that would be even better.
This type of things are time sink holes, as surprisingly it takes a lot of time to figure everything. I was hoping to dedicate a decent amount of time to review and structuring but sadly life got in the way. If you have a suggestion how to structure it better I am all ears.
Most of the patterns should link to external resources since they were derived from them. If there’s no link, it was probably obvious or I’ve derived it from my own project.
The flow was, me finding interesting pattern -> Claude ingesting the reference and putting it in a template -> Me figuring out if it makes sense -> push
Author here. Yes, CC is the maintainer. When I stumble on a decent idea, I would just feed it to CC to create a pattern out of it. This was my quick and dirty approach to a public learning log with an idea that I would get back to it at some point and clean it up. Which I did on a few occasions.