When AI ships code faster than anyone can review it, velocity metrics go vertical — and drift accumulates silently. The Production Drift Ratio is the first metric designed to make that cost visible.

Technological advances in the Industrial Revolution wiped out artisinal craft. We're at a moment when AI could repeat history, but we don't have to let it happen. AI can be a tool, not a replacement. We can reweave code and design and keep humans in control.

AI now writes code faster and more prolifically than any team can review it. The result is a new category of problem that most teams feel, but no one has named clearly. That problem is drift — and understanding it is the first step to controlling it.

Every team that ships software eventually faces the same invisible problem: the design and the code start out in agreement and then quietly stop being the same thing. The roundtrip is what fixes it — if it actually closes.

What happens when everything *looks* like it's working, but what the AI produced isn't what you asked for. Instead, it's just a more impressive-looking version of something adjacent to what you asked for. This is called the AI Distraction Stack. And it's a problem.

Eighteen images and eleven exchanges about rope physics, and a masterclass in what happens when an LLM anchors on its first output and won’t let go.

Imagine that your team runs AI-generated code through every check you’ve built into the pipeline. Security scanner: clean. Test coverage: above threshold. You ship. Three months later, you’re in a war room trying to figure out why adding a seemingly simple feature has become a two-week engineering project—and why the proposed solution keeps breaking three other things. The code isn’t buggy. The architecture is wrong. And it’s been wrong since the first sprint. This is the AI

There's a moment every team experiences about six weeks after they start shipping AI-generated code to production: The demo went perfectly. The sprint velocity numbers were off the charts. Then something breaks. Not dramatically. It's quieter than that: A user reports something strange. Then another. The team digs in. What they find isn't a bug—it's an architecture. Welcome to the production problem with vibe coding. And it's not what most people think it is.

What a packaging error at Anthropic reveals about the moment we’re all in — and what we choose to do next.

AI hedging is the result of the model generating output token by token, moment by moment, with no mechanism to check whether what it's saying now is consistent with what it said ten minutes (or ten seconds) ago.

Figma, Claude, and Codex just announced bidirectional design-code workflows. Here's what they actually shipped — and the pain it leaves completely unsolved.

Reweaving refers to using the power of AI to continuously detect and surface where code has drifted from design and engineering standards, but leaning on human expertise to apply the fixes. Unlike pure AI code generation, which optimizes for speed alone, reweaving optimizes for sustained alignment between a codebase and its evolving standards — treating quality maintenance as an ongoing, systematic process rather than a one-time cleanup.

The best tools are those that reliably extend human capacity rather than act independently. We should be crafting AI that augments our abilities, but stays under our control and direction.

A little over a year ago, I was curious about all the hype around “vibe coding.” I’ve spent my decades-long career working in software and exploring new possibilities with this amazing tech. Okay, vibe coding, challenge accepted. I spent months pushing it as hard as I could on genuinely complex projects. Some of it was impressive. Other aspects? Not so much.

TL;DR Most teams don't have a code quality problem — they have a drift problem. AI accelerates drift because it has no access to a system's history, reasoning, or intent. The answer isn't better inspection after drift happens — it's governance that prevents it structurally. "Human in the loop" means chasing drift; "human in control" means expressing intent through constraints that keep AI within bounds. When we create a document and print it, we expect that what comes out of

n the rush to build faster with AI, there's a hard-learned lesson about why we build carefully in the first place that's being overlooked. It's not about process or methodology — it’s a fundamental question that speed doesn’t answer: Are we building the right thing at all?