In a similar vein, here's my favorite prompt: "Please review the tests you've written. Will the tests actually test what they're meant to? If the code breaks, will the test fail?" It's amazing how often LLMs will write tests that don't test anything.
Neat throwback to Google Summer of Code, when tech felt so much simpler (at least to me). Anyone know anything about CodePath, Social Finance, or the nonprofits listed?
I have been working on this exact problem, and I suppose now is as good a time as any to talk about it.
To make any agent "good", there are two components: the model and the harness. Very few companies can train models, but anyone can build a harness. How much does the harness matter? Can I build a harness that's good enough that I can use open source models with opus level performance? That's the question I've been trying to answer by building better harnesses. None of the existing frameworks have the functionality I need to build a good harness. The features I need are language-level... and so I started building a language called Agency[0].
It's been six months and its going well. Some of the things Agency can do are wild:
- It can pause and serialize execution at any point, making HITL easy
- It has some neat safety capabilities such as handlers[1] and PFA[2]
- You can bundle up any agent as an HTTP or MCP server[3]
- I'm now working on a built-in optimizer to optimize agents (think DSPy).
Obviously, it's a huge undertaking, but having worked with the Agency for six months, I can't imagine going back to another framework. It makes things so easy. I'm working on its built-in agent now [4]. My goal it to get it to be as good as Claude Code, but using open source models. It's still early days, lots of rough edges, but if this sort of thing interests you, I'd love to have a few more people test it out.
It's important to note that thousands of people are giving their time every day for climate change work, doing thankless jobs for very little pay. The fact that the US chose a corrupt man over these people, the fact that millions of people sat out the vote, is a real slap in the face of the people who try to fight this, day after day after day. And many young people seem to have just given up. It's your future, others are fighting for it, join them.
I'm all for people writing their own coding agent harnesses... is there anything different about this one? Its not clear why I'd choose this over pi, opencode, or other existing options
As other comments have said, it would be great to add what this does that existing solutions can't. I see the project has been active since Feb, and has < 150 commits. I'd assume this is still pretty immature. So why use this? I think more explanation is needed.
Why don't all these distro maintainers add their own back doors, and mine crypto off our machines without our knowledge? Surely, there is some legal fine print they can add that would let them do that. There is very little incentive for them to maintain these systems, given how thankless and underpaid the work is.
Do you not think that that would have a negative impact on the world? or do you not care, or is this "the world is going to hell so I may as well get mine"?
I actually don't think Getting Killed the album is well mixed, what turned me on to Geese was their From The Basement performance https://www.youtube.com/watch?v=qIol9hig2G4 the music and the mixing are incredible. I've followed From The Basement for a while, ever since their collaboration with Radiohead. So maybe this was a psyop, but the music is genuinely really good.
I've been thinking about this a lot recently, and I don't know if it is possible to stop. I've been thinking the most impactful thing would be to create open-source tools to make it easier to build agents on top of open-source models. We have a few open-source models now, maybe not as good as Gemini, but if the agent were sufficiently good, could that compensate?
I think that would democratize some of the power. Then again, I haven't been super impressed with humanity lately and wonder if that sort of democratization of power would actually be a good thing. Over the last few years, I've come to realize that a lot of people want to watch the world burn, way more than I had imagined. It is much easier to destroy than to build. If we make it easier for people to build agents, is that a net positive overall?
Armin is pretty well-known in the tech space. He has contributed a ton to open source and generally seems like a fairly principled person. I think this may be the first case where you turn out to be wrong :)
I have been very curious to try Ohm. I'm currently using a hand-rolled parser combinator library, but Ohm looks slick. The online editor is nice too https://ohmjs.org/editor/
> The human is left doing whatever the machine can't, often a narrower slice of the original role
I haven't seen anyone talk about AI and its impact on flow yet. It's pretty easy for me to achieve a flow state while coding without AI, but with AI, I'm not so sure. I spend my time managing multiple Claude instances as they work on different tasks, and there's no time to go really deep into anything.
Flow was such a productivity boost for me. Even though Claude definitely helps me finish tasks quicker, I've started wondering how much quicker it actually is, vs getting into flow.