Name: The Edge That Thinks
Author: Kurt Overmier

What this is

This is the canonical home for the 0.2.1 public-review edition of The Edge That Thinks: a practical field guide for teams building reliable generative AI systems on Cloudflare.

The premise is direct: a model call is not a product architecture. Production systems need explicit state ownership, durable units of work, governed tool access, retriable side effects, and observable routing decisions.

Independently authored by Kurt Overmier and published by Stackbilt. Not affiliated with or endorsed by Cloudflare.

Free download Download the free PDF E-reader Download EPUB Offline reader Download HTML ZIP

Free public-review edition

Version 0.2.1 · August 2026. A substantive early edition shared for practitioner review. The complete book is free; corrections and review findings will inform version 1.0.

This is not the final 1.0 edition. Practitioner review, final assistive-technology review, and remaining publication gates are still open.

Read the free sample online View updates

The curated web sample remains available alongside the complete downloadable edition:

Editable diagrams

Diagrams are committed as Mermaid source plus SVG output under /public/field-guide/figures. They are designed to remain legible in grayscale and avoid color-only meaning.

Reference architecture showing client requests flowing through an edge gateway, admission, policy, job records, queues, coordinators, inference, artifacts, and telemetry. — **Reference architecture.** The minimum separation of concerns for a generative system that may invoke inference or cause a consequential side effect. Mermaid source

Lifecycle diagram showing a request becoming an accepted job, then moving through queued, processing, completed, failed, canceled, and reconciled states. — **Request and job lifecycle.** An HTTP request should create or locate durable work. Cancellation is a state transition: queued work can be canceled directly; processing work can be canceled only when the coordinator can safely stop or ignore the side effect. Mermaid source

State ownership map assigning identity, job truth, live coordination, transport, artifacts, retrieval, cache, and telemetry to separate owners. — **State ownership.** Each consequential fact needs one owner. Storage primitives are implementation details after the ownership decision is made. Mermaid source

Routing ladder from deterministic code through replayed procedures, small models, specialist models, expensive reasoning, and human review. — **Routing ladder.** Start with the cheapest proven capability. Escalate only when evidence, capability, risk, or policy requires it. Mermaid source

Checklist preview

The checklist is a launch conversation, not a maturity score. Start with these gates before adding model calls to production work.

Contract and state

Is work a request, job, session, or workflow with explicit cancellation and terminal states?
Is there one authoritative record and an idempotency key for every effect?
Can duplicate delivery, timeout, partial completion, and retry converge safely?

Identity, models, and tools

Is tenant scope resolved server-side for reads, retrieval, and tools?
Are prompts, source data, derivatives, and telemetry separately classified?
Can deletion reach artifacts, vectors, cache, and queued work?

Reliability and operations

Can long work return an accepted job rather than an implausible finished response?
Is large data kept out of queues and model context?
Are downstream failures clear states rather than fabricated output?

The Edge That Thinks