Core AI owns the horizontal runtime, intelligence, and improvement capabilities that Product and Infrastructure both depend on, along with the Sycamore Forge experiences that make them usable. Engineers here own end-to-end slices, from cloud service and API design through the React interface. The team covers four related areas, and you do not have to pick one to apply. Tell us what you have built and we will work out the fit together.
The agent runtime. Multi-turn sessions, model routing, tool execution, memory, durable workflows, and the APIs that expose them. The hard part is correctness across long horizons: surviving retries, provider interruptions, partial failure, and context growth without losing the thread.
Harnesses and environments. An agent that writes and runs code needs somewhere to do it. That environment has to start fast, isolate genuinely untrusted execution, reach only the services it legitimately needs, and carry credentials it can use but never read. The same area owns the verification layer that decides whether what an agent produced actually works rather than only appears to.
Evaluation. Agent quality is genuinely hard to measure. A change that looks better on a handful of examples often is not, and a judge model can be confidently wrong in the same direction as the system it grades. This is where the offline suites, replay corpora, and statistical discipline that gate a release come from.
Self-improving systems. Every agent run produces evidence, and almost all of it is currently thrown away. This area turns it into improvement: structured trajectories at fleet scale, failure clusters surfaced from production rather than guessed at, and proposed changes that are versioned, measured, staged, and reversible.