Skip to content

Agent Orchestration 2026

Shared thesis across two Coding Agents Conference talks (zach-lloyd, niels-bantilan): 2026 is the year orchestration goes from novelty to core infrastructure. Lloyd stakes the consumer/dev-tooling side; Bantilan stakes the platform/durability side.

Lloyd's framing (Warp)

  • Engineers are now "orchestrators of fleets they don't even run locally anymore."
  • Local laptops hit physical limits with 4–5 concurrent agents against a million-line codebase → laptop-limit.
  • 2026 goal: make launching an agent into the cloud as frictionless as starting a local conversation, without losing visibility or programmability.

Bantilan's framing (Union / Flyte)

Why this matters

Lloyd is the interface layer, Bantilan is the runtime layer. Together they describe the emerging orchestration stack: UI/UX on top, durable-task substrate below, sub-agents (subagent-architecture) in between. The unresolved seam is authentication — see harrison-sam-general-purpose-agents-2026 for the auth argument that plugs this stack into real enterprise work.