Every: why some AI workflows stick and others don't

What the article is about

Published in Every’s Working Overtime newsletter in July 2026, this piece by Katie Parrott addresses a pattern that many knowledge workers recognize: the AI workflow built with care last month that no one is using anymore. Rather than treating abandoned systems as a personal failing, Parrott draws on behavioral psychology research to explain why most automated workflows fail, and introduces a practical four-category model for deciding what to do with each one.

Context: company, task, scale

Parrott is a staff writer at Every, a publication that documents AI use in editorial work in close detail. The piece reflects sustained daily experimentation with AI workflows across writing, research, and communication tasks, including tools like Codex, custom assistant configurations, and automation built around the publication’s content process.

Key method: keep, redesign, revisit, retire

The core of the article is a decision framework with four options for evaluating an AI workflow: keep it if it is delivering consistent, immediate value; redesign it if the underlying need is real but the implementation is wrong; revisit it later if conditions are not right yet; or retire it if the original problem no longer exists or never warranted automation.

The framework gets its analytical weight from two psychological mechanisms Parrott identifies as the primary causes of failure. The first is “symbolic self-completion” — the tendency to build workflows for an idealized version of yourself that does not reflect how you actually work. One of her examples is a social media automation tool that generated post recommendations she accumulated without using, because her actual behavior was spontaneous rather than systematic. The workflow solved a problem she did not have. The second mechanism is immediate reward: research by Woolley and Fishbach shows that workflows delivering quick feedback sustain engagement far better than those promising distant benefits. Parrott’s Compound Writing plugin succeeded where others failed partly because it provided visible value — a paragraph to develop, an argument unblocked — within the same working session.

Maintenance cost is treated as a first-class consideration alongside workflow quality. An AI assistant she calls “Margot” initially delivered value but eventually required more time to maintain than it saved. Retiring it was rational, not a failure.

Who it is useful for

Writers, editors, and any knowledge worker who has built AI workflows and struggled to determine which ones are worth preserving. The piece is particularly useful for people who have accumulated several AI tools and automation setups and need a principled way to evaluate them rather than abandoning all of them or persisting with all of them equally. It does not assume any specific platform or toolset.