Skip to content

CONCEPT Cited by 1 source

Fast feedback loops

Definition

Fast feedback loops = the architectural property that each proposed code change can be validated (pass/fail, latency, correctness) in seconds rather than minutes or hours. Named by the 2026-03-26 AWS Architecture Blog post as the primary architectural constraint of concepts/agentic-development: "Agentic development depends on feedback speed. The faster an agent can observe the impact of a change, the more effectively it can refine its output."

Why it matters more for agents than for humans

Humans batch work across minutes-long feedback; agents iterate every few seconds. A 30-second cloud deployment is a yawn for a human reviewer but "turns every iteration into a high-friction exercise" for an agent running a read-edit-test-observe loop. Multiply per-iteration cost by the agent's iteration count and the feedback tier dominates total cost.

Feedback tier ladder (from AWS 2026-03-26)

Each unvalidated change should use the cheapest tier that can falsify it:

Tier Latency Tool
Local emulation seconds SAM local, systems/dynamodb-local, Glue Docker
Offline dev (data/ML) seconds systems/aws-glue ETL libs in Docker, reduced sample data
Hybrid cloud minutes minimal CFN / CDK stacks invoked via SDK
Preview env minutes short-lived full-app IaC stacks, torn down after validation
Production deploy minutes–hours full CI/CD pipeline

Failure modes

  • Cloud round-trip for every change — agent blocked on deploy + pipeline + log-tail; hours of wall-clock per productive minute.
  • Tight coupling blocks local test — the agent can't isolate a change because the handler reaches directly into a cloud SDK, so no test runs without credentials.
  • Deployment-only failures — bugs only surface in the real cloud environment, not in any local or emulation tier; agent has to deploy to learn.
  • Opaque code bases — agent can't reason about where a change belongs; iteration wastes compute on the wrong file.

Seen in

Last updated · 200 distilled / 1,178 read