Notes from DHH’s keynote
My rough notes from DHH’s rant: the main question, beginner’s mindset, BYO agents, architecture and economic humility.
Writing
On learning, intelligence, and building with AI.
Restructuring around agents
AMP looks like an outlier today. Rebuilding around agents may soon become necessary.
Frontier intelligence and good-enough models
Will smarter models unlock enough new valuable work to justify the premium?
A barbell approach to harnesses
Use a full provider harness for complex work, or something small for a specific task.
Open weights and getting paid
Widely used models do not necessarily bring revenue to their creators.
Separate the system from the agent
Can you change the agent while keeping your knowledge, goals and control?
Subscription value is a moving target
A snapshot of AI subscription value can change within weeks.
Keep the compute or rent it out?
Research upside and current rental income pull labs in different directions.
NVIDIA’s moat and the labs’ feedback loop
Demanding workloads help NVIDIA improve its chips and software.
Building and owning your agent infrastructure
Why agent harnesses should run inside a deterministic control plane you own, and how Centaur and Centaur Context separate control, execution and shared knowledge.
When does AI progress become self-accelerating? Redefining and modelling RSI.
Defining and measuring the feedback loop from better AI to better AI research—and what it would take to sustain acceleration.
The open weights debate (3D chess)
How open weights reshape the incentives of the AI race.
Capability leakage
The concern is not just data leaving a company, but its learning loop.
The skeuomorphic phase of AI software
Are we trying to fit new software into familiar old shapes?
Separate generation from review
Separate generation from review, then create a taste rubric.
Free the Claw: agents break containment. What’s next?
What happens when agents leave the chat box and coding harness, and start operating across everything else.
Dario's road to Valhalla
What's the most plausible story where foundation model companies actually start making money?
Products do labs (and labs do products)
The labs are building products, and the product companies are building labs.
Start with the context window
Whether that external context is stored in files or folders, or in a database or a graph, matters less than the externalisation itself: that the context exists externally and can be easily and efficiently updated and retrieved when required.
Superpowers and stupidity: who gets the leverage?
AI is going to make some people incredibly proficient. It's also going to make a lot of people very stupid.
Creating shared language is hard
On naming, metaphor, and whether language creates reality.
Bending the slope of slop
The proportion of valuable and insightful output needs to grow faster, or at least as fast as, the slop.
Why Obsidian and Markdown are the wrong approach for personal knowledge management
If agents are going to manage your knowledge, you should be building systems for agents, not humans. Markdown is the wrong foundation. SQLite is the better backbone.
AlphaFold: better mental models of AI progress
Understanding ‘AlphaFold’ is a good way to help make sense of the non-linear trajectory of AI progress.
Day 1: getting Qwen3-4B running on Modal
My attempt to get an open-source model running on Modal and what I learned about the abstractions.
Day 0: project setup & my coding agent workflow
Building an open-source knowledge management system for Hugo and Stefan's LLM course students, documenting my setup and workflow with coding agents.
Modelling the model
What happens when being able to model the model becomes a stronger predictor than raw intelligence?
The contextual user substrate
Why collecting and curating rich contextual user data will soon matter more than cranking up model intelligence
Why context matters
A follow-on piece about the growing importance of context and personalization in AI
When human intelligence stops mattering
How to think about human/AI collaboration when raw human intelligence stops mattering
Why it’s so hard to “just say the thing”
On the difficulty of clear communication and what it takes to explain ideas simply.
Why we need to protect the 'what is intelligence?' debate
(and) the people and platforms who facilitate them.
Writing for immortality
Thoughts on writing for AI and the future of human consciousness