Descriptions:
Madison Faulkner, partner at NEA and former Meta AI researcher, and Hugo Santos, CEO of infrastructure startup Namespace, argue at the AI Engineer conference that traditional CI/CD pipelines are fundamentally broken for the era of agentic software. Where human developers submit one or two pull requests per week, AI agents generate thousands of short-lived branches simultaneously—overwhelming GitHub Actions with cache misses, merge conflicts, and serialization bottlenecks that the system was never designed to handle at that volume.
The talk introduces “continuous compute” as the successor paradigm. Rather than treating the PR as the atomic unit of work—a construct optimized for human review latency—the speakers propose an architecture built around intent and spec-first workflows, with a cache layer becoming the central orchestration plane. Key components include ingress shaping and rate limiting, intelligent infrastructure routing, and agentic identity systems that support retries at scale. Mitchell Hashimoto, co-founder of HashiCorp, is cited as an independent validator who has written publicly about the need to evolve GitHub for cloud-native and inference-era workloads.
A particularly sharp analogy frames high-volume merging as a database serialization problem: when humans commit slowly, lock time is large but infrequent; when machines commit at machine speed, minimizing time-to-merge becomes the dominant engineering constraint. Namespace’s hardware-software co-design approach is positioned to address this gap, and the speakers suggest the traditional DevOps layer will see its most significant innovation in the coming year.
📺 Source: AI Engineer · Published May 13, 2026
🏷️ Format: Deep Dive
