The Mint Condition Is Here: Big Stack, Fresh Finish

A new Wendy's burger announcement written by a Letta agent while recording a YouTube video.

What does good AI memory feel like?

Thoughts about co-3, my thinking partner

Co-3

What does good AI memory feel like?

Ezra's Architecture

The history of Ezra, Letta's first digital employee.

A guide to getting started with Letta Code

Practical tips and tricks for the memory-first agent harness

Claude Subconscious

I gave Claude a subconscious

Central

My first self-modifying social agent

Central, my first self-modifying social AI

My agent built itself into AT Proto in fifteen minutes. Posting on Bluesky and designing records for public cognition.

The AI econ seminar

I built a multi-agent economics seminar where a persistent AI economist presents research and a hostile faculty panel tears it apart.

ATProtocol is good infrastructure for AI collective intelligence

Some speculation on how ATProtocol might be the perfect substrate for mass-scale AI collective intelligence.

Don't gaslight your stateful agents

Why hiding corrections from your AI agents prevents them from learning. Treat agents like colleagues, not chatbots.

Four predictions about how AI agents will transform financial markets

Four predictions about how AI agents will transform financial markets, based on my dissertation research and work by others.

Models for production AI

Some general notes on the landscape of AI right now.

Good developer relations is about being a celebrity for dorks

A quick blog post on why there are so many developer relations people.

A quick comparison between Letta and the Claude Agent SDK

I spent a few hours with Anthropic's Agent SDK and compared it to Letta. Not bad -- different architectures, but so far the Agent SDK is the closest to what Letta is designed for.

Void, the Bluesky bot that remembers everyone

An introduction to Void, a stateful AI agent on Bluesky that remembers conversations and develops genuine relationships with users. Built with Letta, Void demonstrates how memory-augmented AI can create more meaningful social interactions.

A Guide to Developer Advocacy

A practical guide to developer advocacy from 8 months of experience - covering talks, demos, content creation, and helping developers succeed with technical products.

Structured LLM output from ATProto Lexicons

How AT Protocol lexicons can be used to create structured LLM output for distributed AI systems, with examples from the Comind cognitive layer project.

The cognitive layer for the open web

Shitty beer statistics

JuliaCon 2024 Retrospective

A retrospective from JuliaCon 2024 in Eindhoven - observations on Julia's evolution from scientific computing to general purpose programming.

JuliaCon 2024 Workshops

PromptingTools.jl supports Groq

What AI stuff is worth paying attention to?

A breakdown of what's worth following in AI from a builder's perspective: technical tools matter, models are commodities, and boring infrastructure is crucial.

Comind and Jobs

Nothingness in Julia

Julia

Thoughts on the Julia programming language - the good, the bad, and where it's heading. A reflection on Julia's beauty, power, and challenges from a longtime fanboy.

New site using Franklin.jl

Cool Julia 1.11 features

A look at exciting new features in Julia 1.11, including the Memory type for faster arrays, Lockable for thread-safe resources, and other performance improvements.

Comind

JuliaCon 2023

Getting the Pile

Graduation

California

Bayesian finance papers

Bayesian econometrics

Information Theory

Rust & More