Videos

Simular's neurosymbolic computer use agents pair a VLM agent with a deterministic coding language (Simulan) to...

Browserbase's auto-browse loop applies Karpathy's auto-research idea to browser agents, using a doer agent and a...

Static evals and benchmarks create a 'benchmark illusion' — they don't reflect real customer usage. Amazon AGI Lab's...

PlanetScale argues that the 'move fast and break things' ethos is outdated for AI-era infrastructure; instead, you...

Evals are reliability instrumentation for non-deterministic AI systems, analogous to fuzzy unit tests. The talk...

A browser agent with 99% per-step accuracy achieves only 36% overall success on a 100-step task because cost...

Postgres 19 is removing several features before release — property graphs, online checksums, FOR PORTION OF,...

Voice agents are moving from demos to production at scale, but reliability remains the top blocker, with even a 1%...

Small, specific adjustments to coding agent workflows yield disproportionately large reliability improvements. The...

AI agents have become distributed systems that can cause side effects; treating them as such with deterministic...

The core tension in LLM gateways is that availability, latency, guardrails, and cost cannot all be maximized during...

Only about 5% of demo calls with AI startups result in a signed enterprise contract, and the main barriers are not...

AI system architecture evolves faster than evaluations, creating a widening gap between what models can do and what...

Maven Clinic's transition to an AI-native company required rethinking development processes: they abandoned...

The debate between loop engineering and graph engineering for AI agents is settled by the task's reliability...

Most production agent failures are not model failures but harness failures—the system that owns state, orders...

Current AI agents can reliably perform individual steps and use tools, but they fail at end-to-end knowledge work...

Reliability in multi-step AI agents is a control problem, not a prompting problem. Microsoft's ACE voice tutor uses...

An agent's identity is fundamentally its log—the append-only event history capturing every input, output, and tool...