Agent Infrastructure
person concept tool org talk claim — click a node to jump to its page; hover an arrow for the relation
The systems, harnesses, and tooling built to deploy, run, and improve AI agents in production contexts — distinct from the underlying models themselves, and typically customized to the specific workflows and data of an organization.
YC's Internal Build
Y Combinator began constructing its own agent infrastructure approximately a year before the talk, motivated by the need for YC-specific agents rather than relying on off-the-shelf solutions. As described in Inside YC's AI Playbook: "we started building our own harness inside of YC for kind of YC specific agents about a year ago" (2:46). This suggests that organizations with specialized use cases find general-purpose agent platforms insufficient and invest in bespoke infrastructure.
Self-Improvement as a Core Component
A notable element of YC's agent infrastructure is the Self-Improving Dream Cycle, which is treated as a constituent part of the broader infrastructure stack. This cycle involves a general agent that operates on a nightly schedule, reading through all agent conversations to identify improvements (18:36). This positions agent infrastructure not as static tooling but as a system capable of iterating on itself using its own operational data — blurring the line between infrastructure and the agents it supports.