I'm building ETL layer for agent traces. Agent traces is most valuable data that people are generating that can be utilized to improve agents and models, track costs,learn user behaviour, etc.
I've been extremely frustrated with any large new work that i do with agents. Then plan multi step, multi hour work with extremely large code changes running for 30+ hours. In the end what you get is sometime completely useless code because it made an assumption that wasn't true at all. In the end, i end up wasting hours.
I've been working with self improvement harness a little bit and one thing i've come to conclusion is harness task fit. The learning can be significantly improved if we understand the behaviour of task and how it should be learned.
I'm pretty sure a general solution will definitely exist which will do fine, but we are yet to see one.
I agree with the premise that LLMs reward experstise, but people without expertise can very eaily learn to prompt correctly and get to a result that is very good. I remember somebody proved a mathematical conjecture by just asking 'keep going' in plain english without a mathematics background.
Strong agree with the point that "agents" shouldn't be called "agents", as agency lies with human. That being said, I don't like the word clanker either.
Congrats on the launch. Currently all my hobby projects go like procure auth from clerk, database from neon, vercel/cloudflare for deployment, etc etc. This could actually solve that problem. Is there a local simulator that can be used for testing and then replicated easily in prod?
Haven't used clawvisor, but what I understand is that its trying to solve the same problem but very diffefrently. Authsome doesn't bundle any tooling and works as an MITM proxy with credential injection. This means you can work with your own tools, cli's or MCPs.
Been tinkering with something of my own at https://github.com/manojbajaj95/authsome. Core goal was to do credential management, from an ease point of view and not security.
I attended a talk from Yann LeCun, and he always had a strong opinion about auto-regressive models. Its nice to see someone not just chasing hype and doing more research.
Early in my carrier, i saw code written by people both junior and senior. Every single time i saw a great code, it made me feel like its so simple. heck, i could have written it. It was completely opposite for junior-mid folks.
Yes. This is important, rather than treating it as a gospel, each company/founder has to find out a way that work for them. It doesn't have to Steve Jobs model nor the John Sculley model.
In footnotes, he mentioned that C level execs are good at managing up rather than down. As I've seen, this happens with a lot of founders who are busy manging VCs and other stakeholders that they completely rely on VPs.