Running two AI models is cheaper than running one good one. A frontier model as architect, a cheaper one as engineer. New post on how I set this up and where it's heading. samihonkonen.com/posts/two-mo...
Two models are cheaper than one
How splitting the work between a frontier model and a cheaper one cuts costs without a quality hit.
samihonkonen.com
Smaller models forget system prompt instructions. What if we attach instructions to specific tools and deliver them the moment the model tries to use them? I call this guidance injection. samihonkonen.com/posts/guidan...
Guidance injection: reliable instructions for local LLMs
System prompts don’t work reliably for smaller models. Guidance injection delivers instructions at the exact moment they’re needed instead.
samihonkonen.com
I've been building a capability library for my AI agents. A global index, docs that agents read on demand, maintained with AI. Hopefully, over time, it will compound. samihonkonen.com/posts/an-age...
An agent capability library
A growing collection of tool docs that agents read on demand
samihonkonen.com
I had a Mac Studio mostly sitting idle, so I started leaving it on and running a local LLM on it. Now I've got a small collection of purpose-built AI agents on disk, each shaped to a specific job, with the data staying on my own machines. New post:
Purpose-built local AI agents
How I turned an idle Mac Studio into a private AI backend, and built purpose-built agents on top of it.
samihonkonen.com
My first self-hosted server lived on zort.org in the early days of the internet. Letting that domain go is one of my more lasting regrets. exe.dev brings back the feeling, but sadly not the domain. New blog: samihonkonen.com/posts/a-love...
A love letter to exe.dev
Why having a computer on the internet feels good again.
samihonkonen.com
Installing 500-series signal processors still has the same feel as installing RAM chips to your PC 25 years ago. What's not to love ❤️
Today I saw a woman reading a research paper in a cafe. I mean a real, printed paper. No LLM, nothing. It wasn't a summary, it was an actual, long research paper. And she was just reading it raw. Like a psychopath.
I just analyzed my whole angel investment portfolio with deathbyclawd.com 😅 My weighted average (based on the amount of capital at risk) was 46.6. Among the best were nohewa.fi. @tangled.org also did well. I won't mention the one's that didn't do so well 🙈
Are You Just a .md File?
Find out if your SaaS can be replaced by a Claude Skill. The SaaSpocalypse Survival Scanner.
deathbyclawd.com
today, we're announcing our €3,8M ($4.5M) seed financing round, led by byFounders with participation from Bain Capital Crypto, Antler, Thomas Dohmke (former CEO of GitHub), Avery Pennarun (CEO of Tailscale) among other incredible angels. read more on what's next: blog.tangled.org/seed
announcing our €3,8M seed round
and more on what's next
blog.tangled.org
