The AI Slop Spiral — On Resonance

All writing<br>I open the document. My computer screen glares brightly as the sun sets outside my home office windows. A dying lamp hums faintly in the background, interrupted only by the click of the mouse wheel as I scroll.

Twelve pages.

At least this one is only twelve pages.

It is the seventh document I have received today. A steady stream of PRDs, technical specifications, incident reports, and QA testing plans. Each one looks impressive. They are thorough, professionally formatted, and full of tables, edge cases, diagrams, and implementation details.

At a surface level, they all seem smart and comprehensive, and it gives the confidence that the organizational engine continues to churn. Work continues moving through the process.

I pull up the latest PRD. It is long, however, and I have a growing pile to get through, not to mention all the other tasks on my plate. I put it into AI and ask for a summary, risks, and priorities. Once I get the response, I then take the time to roughly conceptualize the project in my head and ask a few questions and validate several assumptions. AI assures me that the requirements are clear and the important specifics have been accounted for, so I sign-off with a few notes that someone downstream should validate certain implementation details.

The PRD is asynchronously handed off to engineering and design. Design uses AI to build out assets and mockups. Engineering uses AI to turn it into a technical plan. QA runs it through AI to generate acceptance criteria.

Work is underway.

Faster than ever, a pull request comes in, the code is done. Copilot has already reviewed it and left a series of notes about error handling, naming, and possible edge cases. After some back and forth with a few coding agents, the code gets the approval of AI and another real human engineer and it moves to testing.

QA uses AI to help validate the implementation against the acceptance criteria that AI generated from the AI-generated PRD and designs that AI distilled from someone higher-up’s AI stream of conscious notes. The PM signs off. Design signs off. QA signs off. All are real humans.

The feature goes to production in record time.

A day later, we roll it back.

Several events are missing. Important data is not being populated. There is a duplicate-record bug that can allow someone to create multiple accounts, and even if it had gone out, we learn that the feature was not actually what was envisioned by the key stakeholders at conception.

The next two weeks are spent triaging, level-setting, and trying to figure out what happened.

Meetings are scheduled. Requirements are clarified. Ownership is debated. Tests are added. The team goes back through the original documents and assets, and discovers that some things were never specified, other things were technically specified but interpreted differently, and a few critical assumptions existed only in someone’s head or not at all.

Eventually, we ship the feature again. But now it has taken much longer than it should have. Multiple stakeholders are frustrated. The team has lost some trust in the process and in each other.

Still, from the outset, every individual step appeared to move faster.

The AI Slop Spiral

AI has changed the way we work.

This is not coming. It already happened.

It is an enormously powerful tool. It allows people to write, research, analyze, design, and build things faster than ever. A single person can produce an amount of work that would have seemed absurd only a few years ago.

But AI has also made it much easier not to think. Worse, it has made it possible not to think while producing something that looks very much like thinking .

A twelve-page PRD gives the appearance of rigor. The corresponding technical specification is filled with jargon and detail that exudes understanding. A set of polished mockups makes it look like the user experience has been carefully considered. Forty acceptance criteria look like comprehensive and well-thought coverage. A pull request with dozens of comments and several follow-on commits looks like a careful review.

We live in a world now where none of those things mean that anybody understands what is being built. More to the point, none of this means that at any point did the team have a shared understanding of the work being done and its impact area.

An AI product requirement document begets → AI mockups and assets begets → AI technical specification begets → AI acceptance criteria begets → AI-generated code begets → AI code review begets → AI testing. It is a game of telephone played between teams and within teams, with AI participating at every step. Context is compressed, transformed, expanded, and then transformed again, and shared understanding is lost along the way.

I call this The AI Slop Spiral.

Artificial Confidence

The real danger is not that AI produces bad work, to be fair, humans have always produced bad work. The danger is that...