DEV Community

Cover image for I Tracked My AI Development Waste: Why Governance Alone Wasn't Enough
Jon Schuck
Jon Schuck

Posted on • Originally published at jschuck9.substack.com

I Tracked My AI Development Waste: Why Governance Alone Wasn't Enough

A two-week experiment revealed that managing shared context and execution state mattered more than the governance framework itself.

I measured two weeks of AI-assisted development and discovered that the biggest source of engineering waste wasn't the model—it was unmanaged context drift across sessions.

Over two separate weeks, I tracked the opportunity cost of building the same local project using two iterations of an AI engineering workflow.

Both weeks followed a governance framework.

Both weeks used design documents, implementation plans, and structured execution.

The critical difference wasn't whether governance existed—it was whether the workflow continuously managed shared context and execution state across AI sessions.

That single change fundamentally altered the economics of development.

The Metrics at a Glance

Workflow Strategy Issues Closed Total Time Reactive Waste Est. Cost per Issue Closed
Week 1: Governance without Shared Context 3 14.1h 56% $1,175
Week 2: Governance with Shared Context 7 5.2h 0% $186

The Missing Piece Was Shared Context

What surprised me most wasn't the model's capability.

It was how quickly context drift accumulated into reactive engineering work.

As implementation progressed, assumptions slowly diverged across sessions. Even with governance artifacts in place, the AI's working context became increasingly disconnected from the project's intended state. The result was repeated debugging, unnecessary rework, and hours spent correcting issues that should never have existed.

The biggest improvement came from changing the workflow—not the model.

Instead of treating governance as static documentation, I treated shared context and execution state as first-class engineering artifacts that had to be continuously verified, synchronized, and propagated across every session.

That relatively small investment fundamentally changed the outcome:

  • More than as many issues completed

  • Roughly 63% less total development time

  • Zero reactive cleanup

  • An 84% reduction in estimated engineering cost per completed issue

The takeaway wasn't that the model suddenly became smarter.

It was that governance without managed context is only partial governance.

I documented the complete experiment—including time logs, cost calculations, failure analysis, and the specific issues that consumed hours of unnecessary effort—so others can judge the results for themselves.

Read the full breakdown:

What 7 Issues in 5 Hours Taught Me About AI Workflow Economics


The views expressed here are my own and do not necessarily reflect those of my employer or any organization with which I am affiliated.

Top comments (0)