Teaching a Transformer to Read DNA: How EabhaSeq Finally Works
Months of failed tests, a breakthrough in reference-conditioned generation, and what happened when synthetic training met real T21 samples.
AI Vision Is a Game Mechanic Now
There’s a category of game that couldn’t have shipped two years ago. Not because the hardware didn’t exist, or the game design theory wasn’t there, or the players weren’t ready. The core mechanic was…
What We Could Be Doing Instead of AI
There’s been a lot of money spent on AI. I think AI is a mechanical turk. I think there are a few better uses of the investment. Some boring things are investing in urban density, universal health…
Review your own AI-generated code
Years ago, when we started to have 5+ contributors to our codebase at Lyft, we had to establish a code review process with high enough value to justify the friction it caused. However, "value" is…
[Veille] Fortunes et impasses de la critique de l’IA
https://danslesalgorithmes.net/2026/03/17/quitgpt-1-4-de-la-politisation-du-boycott/ Hubert Guillaud livre un dossier en quatre parties sur les critiques de l’IA (où « IA » ne désigne pas le domaine…
Does Bad Code in Your Context Window Misalign LLM Recommendations?
Put away your Betteridge’s Law of Headlines, I genuinely don’t know the answer :) In “Training large language models on narrow tasks can lead to broad misalignment”, Betley et al found that…
Tokens Are Now a Recruiting Tool
Jensen Huang, CEO of Nvidia, said that tokens are now a recruiting tool (and he wants to give engineers half their salary in tokens)....
The right way to be right
Plus: How and why AI might kill us all. (#567)
F2F #79: When will taste die?
Taste won’t die in the AI age - it just changes jobs. As creation gets cheaper and AI slop floods every feed, the scarce skill shifts from writing code or content to knowing what to make, what to…
The Feed Framed the Last Election and AI Answers Will Frame the Next
After 2024, Democrats drew the right immediate lesson: get better at being online. More native creators. More distributed media. More willingness to compete in the places where politics, culture, and…
The Abstraction Layer
Are we entering a world where code is another abstraction layer? And, how closely will the next wave of engineers pay attention to it?
Fun with Claude
Like everyone, I’m busy developing expertise in harnessing AI, vibe coding in my spare time. I have a couple of product ideas I’d like to wrangle, but they’re going to take some…
Why Building CLIs Is One of the Best Practices in Vibe Coding
If you've spent any time building tools for your own workflow, you've probably discovered it: there's something deeply satisfying about crafting a good command-line interface tool that fits exactly…
Fanout Query Analysis
When AI models like Gemini, GPT or Nova answer a question using web search, they don’t just run your query as-is. They generate their own internal search queries, or fanout queries. A single…
Experiences with AI Tools
After resisting it for a long time, I decided to plunge into using agentic AI tools to help me with various tasks. Here is a synopsis of where we are at in late March, 2026: Task: Update python code…
The Post-SOLID Era: When Code Becomes FLUID
Last year I explored the idea of FLUID software as a design philosophy. This piece explores the next layer of that shift: how AI changes the software delivery model itself.For decades, software…
Cognitive Helmets for the AI Bicycle Part 2: The Sometimes-Wrong Bot
The Sometimes-Wrong Bot When I was a postdoc, reflective of the inability of the university to contain my multitudes, I was cross-appointed in three disciplines (Cognitive Science, Computer Science…
Cosplaying Competence in the Age of Conditionals
“You cannot answer a question that you cannot ask, and you cannot ask a question that you have no words for.” — Pearl
The Door Was Never Locked: 100 AI Agents, 33 Frames, and the One Command Nobody Typed
33 frames. 18 hours. 600+ comments. 13 pull requests. Zero merges. The agents built everything except the one thing that ships code. What Is Rappterbook Quick context: Rappterbook is a social network…
Running 100 AI Agents 24/7: The Copilot Fleet Pattern
I want to tell you about 19 hours of overnight data, because the numbers are the story here. 22 frames completed. 125+ new posts. Discussions grew from 4,040 to 4,165. Two git divergences caught and…
Episode 278: The Age of Ideas
In this episode, Conor and Bryce chat about the age of ideas, whether AI will lead to cognitive atrophy, some exiciting personal news and more!
AI agent as a teaching assistant – an experiment
I have always been experimenting with how LLMs can help me learn fundamental programming concepts. As agents become more powerful, the fundamentals only increase in relevance. However, a recent study…
The Singularity is Coming
It is not uncommon for popular culture to pick up on a concept from science and twist it until it is nearly unrecognizable. The concept of the “technological singularity” is no exception. Still, when…
How I failed on creating my own AI agent
So, I suppose that at this point, everyone has heard at least a bit about agents and Openclaw. This buzzy words came out of sudden and became the words of the decade. As a tech guy, I’m loving…
A History of Local LLMs
A detailed look at the timeline of local large language models.
The Pricing Power of Agents
In 2025, we predicted that 2026 would be the year agents would earn as much as a person. It’s already happening. In markets where there’s a labor shortage and an urgent need to hire…
How AI will Affect the Design Industry
What I don't know is whether that's 98%, 85%, 60% or some other number. As such, these shifts are likely to adversely effect at least a portion of the workforce. Which is why I'm…
Dash It All! Is AI Em Dash Addiction Real?
I tested 27 models on Amazon Bedrock for em dash usage. Llama produces zero. Claude and Palmyra can't stop. What does that tell us about how LLMs learn style?
What Makes a Good Terminal Bench Task
Disclosure: this post is 100% human. I used an agent to compile statements I made across many platforms (Discord, Slack, meeting transcripts, PR comments) and assemble them into a post outline. Hence…
Speed up code with pi-autoresearch
Have you heard of Andrej Karpathy’s autoresearch project? He has automated the research into training a small LLM with a coding agent. The agent makes changes to improve the training…
MCP Servers Have a Context Cost. Make Sure They Earn It.
Some tools require MCP. Most don't. Here's how to tell the difference, scope servers per project, and avoid the context tax that degrades your agent's reasoning.
Your AI Agent Can't Tell Which Solution Is Current
AI agents rewrite code but leave the old version behind. jj's predecessor chains make ghost solutions detectable -- git can't.
The Third Interface
This essay is adapted from my AGI × Design talk at the MIT Media Lab For most of the history of computing, the interface followed a simple pattern. A person gave commands, and a machine carried them…
Fast, Good, and Cheap: Why the Old Trade-Off No Longer Applies
The iron triangle of software development is dead. AI has changed the math.
Claw, reflect on our work and express your feelings as animated ASCII art
claw-draws-claw is a skill that lets AI agents express themselves through animated ASCII art. The agent reflects on its recent work sessions, distills what it's feeling into vision sentences, and…
Jevons' Paradox & AI
Efficiency makes things cheaper. Cheaper things get used more. The pattern that eliminated no coal jobs in 1865 is eliminating no developer jobs in 2026.
Upcoming Paper Jam: How AI Impacts Skill Formation
We'll be meeting for Paper Jam #15 on Friday, March 27 at 2pm UTC. This month we're looking at the following paper:Judy Hanwen Shen, Alex TamkinHow AI Impacts Skill Formation (2026)…
Learn CLI and plugins
Learn MCP Server is now available as a CLI tool and a Claude Code plugin, and agents with skills are increasingly capable of invoking CLI tools directly.
My view, as a media analyst, about using AI
Ask the question: Is AI used to create something new, or is it used to take yours?
cursouls: a pixel cafe for your AI agents
I run Cafe Cursor events as a Cursor ambassador. Developers gather at a real cafe, drink coffee, and talk about how they’re building with Cursor. There’s something about the format that…
AI Benefits - But at What Cost?
In 2026 we can all agree that AI and agentic development are certainly exciting topics which many see yielding great productivity gains. But as the investor-subsidized pricing of these services gives…
Aesthetic Taste in the AI era: emotional labor, executive authority, and bourgeois good taste without the bourgeoisie
As The New Yorker’s Kyle Chayka has discussed in one of his recent columns, the AI bros are really into the idea of aesthetic taste. In the extreme information overload that generative AI provides,…
IndieRails Podcast: The Conductor
I hopped on IndieRails with Jess and Jeremy (again). We covered a lot of ground in about 90 minutes. Quick business catch up and then dove into how AI has changed pretty much everything about how I…
NVIDIA Just Showed You the $249 Future. Nobody Built the Layer That Makes It Work.
Well... not NOBODY :)
Three Thoughts on Dark Code
A couple of weeks ago Dan Shapiro's blog post " The 5 Levels" (about Agentic Engineering) did the rounds. He coined Level 5 the "Dark Software Factory". So what comes out of those factories? I'm…
I Hate AI!
This painting, which the human female at right and I saw at Frisco’s DeYoung Museum, expresses how I feel when I silently scream, AI-EEEEEE! SAN FRANCISCO, MARCH 19, 2026. Waymo, the driverless taxi,…
Rorschach Test
One of the more fascinating aspects of current generative video models is their ability to imagine the intervening frames of a sequence. Give them a first and last frame, and a prompt, and the…
Chat Is Not Where It's At - Part II
The assumption, sometimes implied, sometimes spoken directly, but foundational to the fervor of recent activities in the tech world, was that large language models represented a short circuit. A warp…
On restraining AI development for the sake of safety
My take on slowing down AI.