The Signal

Everything that matters in AI, with our take.

Updated through the day. Every headline links straight to the source. The two lines underneath are ours.

agents149 research125 infra70 claude64 labs52 capital51 safety40 culture36 security34 regulation31 anthropic30 openai24 evaluation18 enterprise13 reasoning12 benchmarks11

Latent SpaceArticle2h ago

Building Legal Agents When You Can't See the Data - Jacob Lauritzen, CTO of Legora

This is the same piece as Item 2 under a slightly different framing. Legora's core problem is a feature, not a bug: they need eval strategies that work without observing agent behavior. That's a hard technical constraint that applies to any agent handling sensitive data. The techniques they developed probably translate to healthcare, fintech, and government AI work. Worth reading if you're shipping agents in regulated contexts.

agents legora legal-ai

Latent SpaceArticle2h ago

Building AI for Lawyers - Jacob Lauritzen, CTO of Legora

The interesting tension here is between what makes agents valuable in law (domain-specific reasoning over messy data) and what makes them hard to build (you can't see half the work because of client confidentiality). This is a real constraint in regulated verticals that most agent tutorials ignore. If you're building agents for finance, healthcare, or law, this conversation covers the evals and blind-spot problems you'll hit.

agents legora legal-ai