BAIR Berkeley
★ 9/10 signal
Intelligence is Free, Now What? <br> Data Systems for, of, and by Agents
agenticmulti-agentmemory
What happened
A perspective from UC Berkeley's EPIC Data Lab explores how near-zero inference costs will transform data systems. It proposes three core paradigms: 1) Data Systems *For* Agents (handling agent swarms, speculative execution, and structured memory); 2) Data Systems *Of* Agents (managing state, consensus, and coordination across thousands of long-running agents); and 3) Data Systems *By* Agents (using LLMs to synthesize custom data systems from scratch).
Why it matters
It provides a foundational architectural blueprint for the next generation of databases built specifically to support autonomous agent swarms.
The take
This is a seminal framing of how database architecture must evolve for an agent-first world. Instead of optimizing for human queries, databases must optimize for high-concurrency, low-latency agent swarms that read/write structured memory and require transactional state management. The concept of 'agentic speculation' (pre-fetching data based on predicted agent steps) is particularly compelling.
Do this
Read the full BAIR blog post and start thinking about how your application's state and memory systems will scale when queried by hundreds of parallel agents instead of a single human.
Don't read this site daily. Get it in your inbox.
The daily brief and Sunday deep dive — distilled, scored, and opinionated. For builders only.