But who to say we actually need 10x when people are more comfortable with 3x. I really don't know. I do know the more I try to use traditional tools and approaches I just end up abandoning them and rewriting them. But it's hard to know if that means anything, because that's always been my approach.
I just don't think you need to redo your process for AI. You just need to AI enhance it. The success of AWS was not architect for the cloud, it was lift and shift. Meet people where they are.
I think a lot of the power of AI is about enabling individuals to leverage it to do their job. So first, you need to make devs more productive developing with AI. Then enterprise will naturally freak out and want visibility and control. The enterprise sale is always the same thing.
I realize I'm building something for the individual developer. Not teams. Teams is a whole different concern. AI can be used by managers to do managery things. I want to use AI to be used by a devs to do cool things.
Breaking: Browser Use + Jev = Ultrafast ⚡
Findings flights took 7s and cost only $0.0039 🤯
> new action space every step
> DOM state space
> small LLM fallback to type
(this video is at 1x speed btw)
Built a tiny open source browser agent. try it below ↓
the jev release is giving react major version drop vibes. just a massive crashing wave, everyone rethinking fundamental stuff and thinking through the implications of a novel new primitive. wonderful.
We're adding support for AGENTS.md to Claude Code.
Starting today in version 2.1.277, if there is no CLAUDE.md in a folder, Claude will check for and use AGENTS.md.
You can toggle this behavior in /config.
We're adding support for AGENTS.md to Claude Code.
Starting today in version 2.1.277, if there is no CLAUDE.md in a folder, Claude will check for and use AGENTS.md.
You can toggle this behavior in /config.
This basically sums up the difference between codex-cli and claude code.
Codex memories are a git repo, with an extensions system, and seeded with 38 lines of markdown.
Claude code is an empty MEMORY.md file the model records TOC lines that point to other markdown files the model writes.
In the next ty release, we have full support for recursive implicit type aliases (at arbitrary depths).
For the type theorists among us, this is implemented using mu types.
@cjc@karrisaarinen@40yroldstartup@linear I think this is it. The work already exists. They are agent sessions. I need to manage those. I already have a system of record, the sessions. I don't want to flip the paradigm and make the system of record the board and have the agent subscribe to the board.
Ugh, I'm studying project management. What is wrong with my life. I literally trying to understand JIRA. You know I've never once actually used it. If you've ever worked with me that would come as no shock at all.
I'm actually not looking to use either. This is the basic issue. I have a lot of agent session doing things. I need to work better, so I need to herd cats a bit and also automate more of the behavior of agents. So I need to get them to follow a process instead of me kicking them along at each step. So it starts to basically look like a kanban board process of agents moving across lanes. From that perspective I started looking at integrating with existing systems, and trying to learn the best approaches.
But the issue though is I'm a developer and I'm just trying to make my single developer life better. If I care about other people that would be a team issue and I need JIRA or Linear.
As a single developer, the last thing I want to do is write and manage tickets. Even if you think it will make my life easier, it's not what I want to do. I want something like a process to exist, but I don't need to really ever see it or really want to see it. So this is more bottoms up. All products like jira/linear tend to be top down. It's all about giving visibility to what is happening to the people below. I don't need so much visibility, I need to herd cats.
As soon as I'm looking at an issue board I'm already to abstracted from the actual work.
@bnjorogedev I started with vz. It was super simple and worked. I later added libkrun for linux. That's the only reason. Now that you bring it up I might just use libkrun on both platforms so I have less to support the golang vz binding I'm using I think I had to fork, so not great.
I figured out this cool thing. It's called tags. You can just put tags on things. And then can you randomly track state and progress. It's like giving a gold sticker to a child. Good job item, you are now a super star!
@karrisaarinen@40yroldstartup@linear I'll put it this way: I'm a developer that wants to develop better with agents. I'm not a manager that wants to manage agents. I don't want to track work. I want to be able to do more work.
@karrisaarinen@40yroldstartup@linear But that's using agents to solve linear. It first assumes quite a bit. It's many layers I don't need unfortunately.