Max Krueger

If you're hyper-ambitious but without direction, spend all your time doing this right now: 1. Keep up with what's happening in AI – the latest tools, releases, news. 2. Tinker with vibe coding and AI agents. Continuously try to automate more of your work. 3. Write on LinkedIn and X, or make TikToks. Build your surface area for future optionality. 4. Find your community of equally ambitious people to tinker with and share advice. 5. Push your boundaries of what you believe is possible with technology and media, every single day.

So formal verification?

@johnhanacek @steipete @BetterStackHQ I run this group. Come join san-diego.aitinkerers.org

@decaf3398 @steipete @BetterStackHQ San Diego needs more meetups for openclaw and AI generally, where are my friends? Haha

I asked @BetterStackHQ for a cli to get logs, instead they built an MCP with 50 tools that takes ~7000 token to clutter up your context and it's logging fetching is pretty sub-par. Use my cli instead. github.com/steipete/bslog

@steipete @BetterStackHQ @steipete If you're ever in San Diego we'd love to host you at one of our awesome AI events!

@steipete @BetterStackHQ We should probably just start writing CLIs for any MCPs that are lacking them haha

😂

@YoniSmolyar I'm working on getting the SD chapter of AI Tinkerers off the ground. Come hang!

san diego founders: we don't live in SF. there aren't thousands of us there are only a few dozen of us I think i think great opportunity to form a community with events and coworking spaces nobody is gonna do it for us! we should all be friends or at least acquaintances comment if interested and tag any SD builders out there?

imagine falling for mcp

@Austen Some of the safety stuff is annoying now but will probably pay dividends in the long run. There will be many uses cases where safer models will be required.

It’s starting tbh

In SF the next couple days if anyone wants to grab coffee

@levie I think your first point is most important. It turns out to be really hard to build agents where you are not the domain expert. Coding almost has to be the domain it takes off in because developers have the skills and domain expertise to make agents effective.

Coding is the best possible domain for AI agents to initially take off in. 1. The very builders of the systems know the workflows so they can improve the tech faster. And they have a huge incentive to make their tools better to create a perpetual flywheel. 2. The limited loyalty developers have for coding tools turns out to be an advantage for the rate of progress, because the ecosystem is constantly chasing performance gains and will switch to whatever works better. 3. The entire value cycle is entirely digital and nearly self-contained, so the cycle times on value creation are near instantaneous. So you can prove immediate upside if it works. 4. As @karpathy points out, code processes digital information, which means it eventually acts as a lever on all other sectors that are digital too. So the TAM is enormous. My hunch is that many of the best practices that are emerging in coding agents will eventually transfer to other categories of AI agents across all of knowledge work. The tools will have to get even simpler, but the paradigms will persist.

@kepano I’ve built many agents but am building an obsidian based agent right now. It’s frankly a great way to interact with the agent.

AI agents built into cloud apps will always be sub-par compared to state of the art agents that you can use with Obsidian. This is because your Obsidian data is in your control, in plain text formats that are ideal for LLMs to process. You can choose to run any of the best-in-class agents and AI tools: Claude, Gemini, OpenAI Codex, etc. You also have the choice to run local models and privacy-preserving options. Or you can use none of these and enjoy not seeing sparkly AI buttons everywhere. Remember that cloud-based note-taking apps are not end-to-end encrypted. This is inherently less secure, open to breaches, and allows companies to make your private thoughts part of their training set. Obsidian is not working on adding agents because we don't have to. We'll let the AI labs focus on that. We're 4 developers and our focus is on making great editing tools for plain text Markdown files that are always in your control.

I guess I'm a fan of lots of threads haha @AmpCode

I suspect we will rewrite the entire codebase within 6 months anyway.

But sometimes it is just better to rewrite the entire codebase from scratch.

When using AI in large codebases it is almost always prudent to go through a few rounds of refactoring to improve context.