KevinSimback

Why run models locally?
Typically two main reasons - privacy and cost
Let’s explore the cost side with an example and see how it maths
Let’s say you want to run an Autoresearch loop overnight like @karpathy
If you have access to an H100, you can run 100 experiments overnight using Opus-4.6 and the API cost will likely be in the $10-25 range
But most of us are not lucky enough to have access to an H100
We can still run 100 Autoresearch experiments on a MacBook for the same $10-25, but it won’t be apples to apples
The H100 will complete 50-100x more training steps over the same timeframe
So if

Some personal reflections on AI
I started off studying CS at uni, but I was never a good coder
I took a corp finance class my junior year and aced it, I was far better at finance than coding so I switched majors
But I always had an inferiority complex towards people who could really code
I knew they could run circles around me when it came to building anything and I was at their mercy for technical dependencies
So I stuck to my Excel models and PowerPoints as my edge
Then working as a consultant at Goldman I met some real finance quants
Most of them could code AND they knew finance
Once again

Think like a Strategist, work like a Dev -> that’s the new model

Claude Max plans are like all-you-can-eat buffets
I’m already full, but sure, I’ll take another plate of tokens please

“Hey Kevin, you’ve been doing all this openclaw and agent stuff, are you more productive?”
100% yes, but at same time my intensity has ramped up with it
So if you think it’s a way to sit back and let agents work while you sip margaritas, that’s not been the case
I suppose it could be for some, but my natural inclination is to just do way more things in same amount of working hours
It’s a double-edged sword

AI agents is the “internet moment” of this generation
For startups it’s a chance to reimagine how products are built and work gets done
For enterprises it’s an operating model redesign that reshapes competitive positioning
For investors it’s an entirely new addressable market

Link to the analysis:

Here are the 3 areas I’m most interested in these days, would love to hear or connect with anyone in these domains
1. Agentic adoption
Believe we’re at an inflection point with AI agents and there’s so much to explore, early innings on this
Super high conviction that agents will proliferate at massive scale, super low conviction (currently) on how value will be distributed and captured as a result
2. AI infrastructure
Something I talk less about but behind the scenes working with one of the most exciting projects in this space, more to come on this
If you believe in #1 above, then you have to

“Agentic Economics” is an emerging field of pontificators on the timeline who proclaim to know how agents will make money in the future
My pov - we don’t yet know, keep experimenting and we will begin to find out
My overarching question - if agents are cheap to build and run, why wouldn’t they keep undercutting each other until the price they charge for various services is just the cost of inference + maybe a small margin?
This to me is the defining question for Agentic Economics and anyone who doesn’t have a good answer to this question has not thought deeply enough on the subject

Here are the 4 best use cases I’m seeing with OpenClaw / agent platforms:
1. Personal assistant / Chief of Staff
Really what it was made for, your personal Jarvis, can be used equally well for personal, professional or both
If you do nothing else with agents, this alone is worth it
2. Coding assistant / coding team
Agents work really well to build things and when you get multiple specialist agents working in a team it’s pretty magical
But you can basically do the same with Claude Code / Codex and tools/MCPs
So the lift here is really just making it more accessible via a persistent agent, for

One of my OpenClaw projects spent $100+ on tokens this morning trying to troubleshoot an issue
Turns out it solved it from the start but pushed it to a new repo for some reason while the deployment was pointing to the original repo, this kept looking broken
🤦‍♂️ AMA

Who is really good with AI video creation?
Need someone to help me with a 20-30 sec video, will pay $100 + credit you whenever I post it
Reply or DM with some examples you’ve created