Vol. 3 of 5 · White Paper · 2026

When the Builder
Can't Sign Off

Copilot. Cursor. Claude Code. Autonomous coding agents. The code is shipping faster than it can be understood — and the Supportability Engineering framework was built on assumptions that agentic development breaks. This paper addresses what happens when no human fully authored what went to production.

AuthorJohn A. Bowman
Covers4 Broken Assumptions
FormatFree · PDF

Download Vol. 3 Free

Supportability Engineering for agentic development workflows — context blocks, structural gates, and accountability without full authorship.

Your information is never sold or shared with third parties.

You're all set.

Your download is ready.

Vol. 3 — When the Builder Can't Sign Off

John will be in touch — dooohhead@gmail.com

The Problem

Four Assumptions. All Broken.

The Supportability Engineering framework — like every engineering quality framework — was built on the assumption that humans make design decisions. Agentic development breaks that assumption in four specific ways. This paper names them precisely and fixes each one.

The answer is not to stop using agentic development tools. It is to adapt the framework so that the standards that used to live in engineers' heads now live in the agent's context — and are enforced automatically.

Assumption 1: Someone knows why the code was written this way
The institutional knowledge that used to live in the engineer's head now lives nowhere. When support escalates at 2am and asks "why does it work this way?" — the honest answer may be: "The agent wrote it."
Assumption 2: The architecture was designed
Agentic development produces architecture through accretion. Blind spots appear not because a designer missed them — but because nobody designed anything. You cannot review an architecture that was never drawn.
Assumption 3: The reviewer understands what they're reviewing
Humans reviewing AI-generated code are less likely to catch subtle supportability issues. The code looks clean. The review confidence trap means problems slip through that a structured gate would catch in seconds.
Assumption 4: The feedback loop has a human memory
If the next sprint is also agentic, SFL findings need to flow into agent context — not just a quarterly slide deck that no agent will ever read.

The Solution

The Framework Adapts

Each of the six framework phases gets a specific adaptation for agentic development. The structure stays intact. What changes is what each phase produces — and how those outputs reach the agent.

Phase 1 · Requirements
The Supportability Context Block
The SRD produces a machine-readable context block injected into every agent session as a hard constraint — not advisory guidance. Failure modes, logging requirements, sensitive data exclusions, dependency rules.
Phase 2 · Design
Continuous Architecture Observation
Post-session architecture extraction instead of a pre-build gate. Gap alerts block the next session until reviewed. The architecture is always current — not a diagram that was drawn once and forgotten.
Phase 3 · Build
Structural Gates, Not Manual Checklists
Automated checks enforce the standards the agent was given. Correlation ID propagation scan. Sensitive data log analysis. Dependency registry check. Human reviewers focus on meaning, not mechanics.
Phase 4 · Test
Generation-Aware Test Cases
Tests specifically designed to catch what agents miss: edge cases handled literally rather than with business logic, pattern-matched error messages, dependencies added without fallback behavior.
Phase 5 · Release
The Accountability Declaration
Engineering lead attests to what can be verified — structural gate results, architecture gap reports, test coverage — not to full authorship they don't have. Honest accountability replaces the appearance of it.
Phase 6 · Operate
Context Block Updates From Production
Every incident that reveals a gap produces a versioned update to the context block. Operational experience flows back into future agent sessions — not just human meeting notes.

"The agent is only as supportable as the context it was given. The job of Supportability Engineering in the agentic era is to make that context precise, complete, and continuously improved by what happens in production."

— Supportability Engineering, Vol. 3

This is Volume 3 of 5.

Vol. 1 is the foundational Shift Left framework for traditional software. Vol. 2 extends it for agentic AI systems as the product. Vol. 4 is the governance framework for the AI systems now operating your support stack. Vol. 5 maps the complete framework to SOC 2, ISO 27001, GDPR, SOX, and FedRAMP. Get all five together and see the complete picture.

Get the Full Series

About the Author

John A. Bowman

Supportability Engineering Practitioner

dooohhead@gmail.com
Agentic DevelopmentAI Governance Support EngineeringShift Left SRECode Quality Operational Readiness

John A. Bowman is a Supportability Engineering practitioner whose work sits at the intersection of support operations, software design, AI governance, and organizational reliability.

This white paper is Vol. 3 of a five-paper series. Vol. 1 is the foundational Supportability Engineering framework. Vol. 2 extends it for teams building agentic AI products. Together they form a complete framework for every layer of modern software development — including the layer where AI builds the software itself. Vol. 4 applies the framework to the AI systems now operating your support stack. Vol. 5 maps the complete framework to SOC 2, ISO 27001, GDPR, SOX, and FedRAMP — compliance evidence as a byproduct of building correctly.

John is available for consulting engagements, staff roles in support engineering, AI governance, or operational readiness, and advisory work with teams adopting agentic development at scale. Reach out directly.