Leon Builds Agents

We're entering a new era

The cool part about the “find things ai can’t do and do that” economy advice is it’s typically given to people who don’t know the first thing about ai (or the economy)

@trq212 Is this based on when our first chat in a session begins or is applied automatically to all messages within that time frame?

To manage growing demand for Claude we're adjusting our 5 hour session limits for free/Pro/Max subs during peak hours. Your weekly limits remain unchanged. During weekdays between 5am–11am PT / 1pm–7pm GMT, you'll move through your 5-hour session limits faster than before.

@DanielMiessler This type of thing is why I’ve advocated for a “model nutrition facts” concept where the model providers need to provide guarantees and information around, not only the model, but also the quantization or other efficiencies of what they’re serving and when to customers

CORRECTION: I have multiple signals indicating that I am likely be wrong about this. The basic vibe is that it's easy within a company's infrastructure to just redirect traffic to different locations, and one location could simply be like a lesser model or version or whatever. Not saying Anthropic or OpenAI or whoever does this (that's a separate claim), just saying it appears to be possible. So while it wouldn't be CHANGING the model itself, it would be changing the model we're experiencing, which is kind of the point. Thanks to everyone who corrected me on this, including @Jhaddix.

claude rate limits yesterday and today have been insane. maybe 80% reduction in actual usage and completely unacceptable (not even in CC, just normal web usage) anthropic not addressing this when there are hundreds of post like this is a major red flag

in the age of ai, not thinking big enough is a technical limiter on success

i think a big part of the cognitive dissonance right now is that the person who doesn't really know code but is vibe coding projects for their own work/customers every day is more of a developer than many new grads who are "hand coding" and refuse to use ai

@lukeburgis Are you clearing context/starting new chats frequently? Could be context rot

My openclaw agent definitely seems to be getting dumber. Was there some update made this past week that would've significantly affected things? Driving my crazy tbh

@curious_vii similar vibe github.com/hamelsmu/resea…

Free α: Using Claude (Code) as an orchestrator that invokes both Gemini and Codex with all the settings cranked up all the way as needed is definitely the move. Probably the highest leverage thing you can do to get the most out of this technology at the moment. In addition to connecting via MCP to all of the services that you use, symlink an agents.md file in your root directory with a claude.md file that just has some basic context about those different CLI commands with encouragement to run them headless as sub-agents when a thought partnership is needed. Checks would be useful if there's any uncertainty, et cetera.

@curious_vii yes it's a healthshare but different than some others i've seen. yea i've been on marketplace rollercoaster and it's terrible. we have concierge pediatrician which isn't covered but between that + crowdhealth we get a lot of coverage. my tldr notes from my convo with them

Is that a healthshare plan? I was actually on Medicaid when working on my last startup and then had to transition to an ACA marketplace plan off cycle, which was an unpleasant Kafka-esque experience, but I’m now on an HMO with sky high deductibles (still not cheap). Haven’t seen a PCP in about 10 years, and I don’t really have any plans to.

The administrative friction in the US healthcare system is incredibly demoralizing. Generally unpleasant. Dehumanizing to all parties involved. Humiliating, even—a feeling of being ever so slowly crushed by a soulless, creaky bureaucratic machine, and there’s nothing you can do.

@curious_vii We ditched insurance and started with Crowdhealth this year. Haven’t had to use it yet but just the concept alone is worth it for us Coverage is better than insurance, $1-2k less a month and you can see transparently what they cover each month

@leonjcoe I should have

@curious_vii have you looked into own your labs? ownyourlabs.com/how-it-works

“Can I pay cash for basic bloodwork?” “No, you must pay for insurance first. Then, you must find a PCP. And get them to write a prescription. It has to be faxed. We need to verify all of these things—your identify and coverage—five different times. The front desk will berate you. Only then may you have the privilege.”

most people: don't want to learn new things don't want to understand how things work want things done for them don't want to take responsibility for themselves are uncomfortable with risk/uncertainty AI is taking people who can harness all these and giving them superpowers

@FracSlap I run the Houston AI Club and would love to participate/collaborate

we were pretty early adopters of Claude Code implementing it as the standard at our company 5 months ago. we're on track to hit 1 million lines of code with it this month. we're thinking about doing a claude code workshop in houston and midland for people in the Collide community. leave a comment if you'd be interested in that so we can gauge interest.

@StewartalsopIII @elonmusk *drunk don draper meme* Ai lawyers

@elonmusk I think we can safely say that property rights are a thing of the past in the US (unless you have the cash to hire lawyers)

Anthropic is guilty of stealing training data at massive scale and has had to pay multi-billion dollar settlements for their theft. This is just a fact.

@curious_vii @StewartalsopIII I think in 2 years it will be solved and commonplace

@curious_vii @StewartalsopIII Also the crypto x payments x ai intersection is small so not a ton of teams well positioned to do all 3 well

It’s actually going to be nuts when these things (agents, whatever) can freely move money around. Right now, they can code, produce images and text, etc. All fine. But, things will get interesting when they can actually transact. And without needing a HITL (or personhood!)

@StewartalsopIII @curious_vii Actively being worked on by at least a dozen companies big and small. This has been in the works for over 2 years and many picked up on it in the last 18 mos

@curious_vii how close are we to that tho?

the modern ai consulting company should look like a mix of playing video games and be run like a mid century ad agency but who will build it?

also just being willing to make 10-20 versions of something until it's right (still saves incredible number of hours vs doing it by hand and quality goes way up)

(suprisingly) still alpha: -deep research (seems forgot about/not integrated in workflows) -prompting multiple models at the same time on same task -having models (self or others) eval the work on rubric -have models ask you questions/build prompts -paste screenshots/error logs