how to decide on full AI vs human-in-the-loop
A heuristic that’s helpe me decide if a given problem should be given completely to AI or use a human-in-the-loop model is “transaction symetry”:
Alex's notes
Recent content on Alex's notes
Latest posts
it's harder to sell exotic cars to people who can afford it
I was watching an interview with a exotic car salesman when he said something that struck me.
AI agent security
AI agent security is application security, network security PLUS controls for prompt injection (the lethal trifecta) and LLM fallibility (trusted/untrusted input).
trusted input
AI agents are able to do good things and bad things.
OpenAI Codex App
OpenAI released a macOS Codex desktop app (I’m pretty sure Codex was CLI only until now).
how to run claude code with local LLMs
Install claude-code-router Create a config file in /.claude-code-router/config.json Run ccr start Run Claude Code ccr code Config for LM Studio { 'Providers': [ { 'name': 'openai', 'api_base_url': 'http://<lm-studio-host-here>/v1/chat/completions', 'api_key': 'not_needed', 'models': ['glm-4.
add a new dokku app
Runbook for creating a new dokku app on aws. You can replace all of the the dokku command using ssh like ssh dokku@mydomain.
my favorite local LLMs
My current list of best large language models for local AI:
AI skills (for humans)
At time of writing, AI needs humans to do anything useful and there is a big difference between the best and worst employees at using AI.
bundler-less JavaScript
Lately I’ve been de-node-ifying my personal projects. I’m tired of the maintenance of dependency churn and security incidents.
RustPython with wasm
RustPython is a python interpreter written in rust. What makes it interesting is that it does no use cpython which makes it much more straightforward to target WebAssembly and do things like execute untrusted code on a host machine.
AI code sandbox
An increasingly important part of the AI stack is running untrusted code in a sandbox.
ChatGPT is a search engine now
OpenAI announced Apps SDK which allows developers to integrate apps into ChatGPT.
the lethal trifecta
When giving LLMs the capability to access private data, view untrusted content, and externally communicate, bad actors can trick AI agents into leaking private data via prompt injection.
export meeting videos from fathom
Annoyingly, Fathom does not provide a way of downloading all meeting videos from an account.
personal infrastructure tour
I build personal infrastructure around the things I do constantly, refined for my workflow (quirks included), with built-in privacy, and for fun.
when to show error details
Only show error message details when the user can recover from it.
local AI
I’ve been spending more time with local large language models. By some estimates, open-weight models (Qwen, Deepseek, GLM, Gemma, etc.
why I don't write (as much) anymore
It’s been awhile since I’ve published anything. I don’t feel as motivated about my note taking practice.
AI web browser
Just like AI coding agents are forking VSCode so that agents have more control over the development environment, AI companies are forking Chromium so that agents have more control over the browser environment.
Comet AI browser from Perplexity
Some initial impressions on the AI-first web browser from Perplexity, Comet:
how to lead a meeting
Meetings are not useful unless they lead to an outcome or clearly defined tasks needed to reach an outcome.
malleable software
Today’s applications provide closed systems that can not be modified or extended to suit the needs of the user.
DB connections in healthchecks is a bad idea
Here’s how I’ve seen DB connection/checks in a load balancer healthcheck become bad idea:
unsubscribe links are a sneaky phishing vector
Maybe you want to be nice. Maybe you don’t want to hurt another well-meaning business’ email delivery because you know from experience how painful debugging deliverability issues can be.
ask AI to ask you questions
A surprisingly useful technique for getting the most out of AI tools is to prompt it to ask you questions that would help it to complete the task you want.
AI agent design patterns
It’s still early days for building user applications with AI but we are starting to see things progress as the industry progresses from LLM workflow to AI agents.
CSM to customer benchmark
Based on a data from Gainsight the average ratio of customer success managers to customers is 144 for low-touch businesses with less than $10k in annual contract value (ACV).
Axum and dynamic dispatch
I kept getting an opaque rust compile errors when writing an axum API handler.
What does it mean to be AI native?
With the capabilities of large language models getting more useful for real work, the pressure is on to incorporate them everywhere.
install a PWA in macOS
Apple keeps making it more and more difficult to “install” a PWA.
typed languages are best for AI agents
Typed languages should be the best fit for useful AI agents.
goose coding AI agent
I’ve been testing out goose an AI agent for writing code that runs on your machine instead of as an IDE co-pilot.
prospects will thank you for disqualifying them
When I started doing founder-led sales, I thought it was my job to find a way to sell to anyone who got in touch and scheduled a meeting.
the bitter lesson
Ironically, one of the most efficient strategies for building with AI is to wait for better models.
AI browser automation
An incomplete list of AI-powered web browser automation and AI agent projects.
why I like ring binders - Plotter notebook
The Plotter notebook from Designphil is a minimal ring binder for planning and writing.
n8n for automation
I started trying out n8n for setting up automated workflows.
ink reviews
This is my personal list of ink reviews based on my usage and preferences.
how to be productive
Here’s a collection of things that have helped me be more productive.
judgment disqualifies automation
When a process requires human judgement for an unknown number of possible decisions, automation is not possible.
do higher temperatures make LLMs more creative?
Higher temperatures tell LLMs when generating a completion to not always use the highest probability next token.
CSAT benchmarks
Customer satisfaction (CSAT) surveys measure how much people like a product or service.
llm workflow patterns
Workflows are a sequence of actions with LLMs to process input into a desired output.
lab notebook for founders
A lab notebook is where research scientists keep track of their experiments so that they can be reproduced and verified.
Prompt Engineering for LLMs - literature notes
Notes from reading Prompt Engineering for LLMs by John Berryman and Albert Ziegler.
latent space reasoning
Rather than converting to text at every step in a chain of thought process with large language models to solve a complex problem, new research suggests that reasoning can happen in a latent space using the internal representation of the model.
mushy systems
As large language models proliferate into every service and ultimately replaces business logic, we will be left with the horrible burden of maintaining mush.
rust build caching with docker
Compiling rust dependencies every time a docker image is built can take a very long time.
AI replaces business logic
Satya from Microsoft talks about how orchestrating between business applications is the next step for artificial intelligence which will replace business logic with AI.