It's been one month since OpenAI folded Codex into ChatGPT, and the Every team has already scrapped and rewritten their year-old guide Codex for Knowledge Work from scratch — a coding tool has become a general-purpose workspace, and every old setting is now obsolete. The same wave of changes also brought an executive AI-adoption checklist, a copy-editing benchmark calibrated against real edits (KateBench), and an experiment to clone a colleague's judgment. The tool got a new skin, and the growing pains are just beginning.

Think of it as an office relocation: everyone had a fixed desk, familiar supplies in the drawers, and an unspoken understanding of "who handles what and who to ask." Then one morning OpenAI bulldozed every desk and drawer, rearranged them all, and handed out new badges — without telling anyone where the pen holder and sticky notes went. The four pieces in this Every roundup are the new floor map, a cross-department directory, a reminder to stop trusting vendor benchmark scores, and a more radical play — turning that one colleague who's always being borrowed by other teams into a clone card that anyone can pull. That's where the analogy ends. The real difference: in an office move you can slowly learn your way around, but ChatGPT's "the merge" changes the API, the billing, and the workflow entry point — and the cost of getting lost is a production incident, not a missing pen.
Event

Less than a month after the merge, the May guide is already obsolete

The guide published in May became obsolete just one month later.

Why did a guide published in May go stale within a month? In May, Every's Katie Parrott published Codex for Knowledge Work, treating Codex as a coding tool. In mid-July, OpenAI folded Codex into ChatGPT, and what had been a coding tool became a general-purpose workspace — internally at Every it's called "the merge." A month later, Parrott tore the May guide apart entirely and replaced it with ChatGPT for Knowledge Work.

The new version overhauls all the workflows: the workspace is split into three zones — Chat, Work, and Codex — with chatting, office work, and coding each given their own home. It also adds three new feature categories: goals, projects, and scheduled tasks. All settings from the original version are void, and the same entry point now has to handle a completely different set of tasks.

16
Workflows included in the ChatGPT for Knowledge Work guide
Source: Every
400
Executives who joined Every's enterprise AI adoption Q&A
Source: Every
33
The most-asked AI adoption questions from executives
Source: Every
30,000
Real editing samples by Kate Lee used to train the KateBench copy-editing model
Source: Every

Also rewritten in the same batch was the March OpenClaw guide. The new version tackles the same question: when should you run a personal agent (a personal dedicated AI assistant) and when should you switch to a team-shared version like Every Agent. The Every team's take is that personal and team agents each have their own use cases, and the old version failed to draw that line.

Why such a radical overhaul? A coding tool became a general-purpose workspace, and every workflow habit built up over the past year no longer holds. This is a product-level repositioning — keyboard shortcuts, custom prompts, and workspace configurations all have to be redone.

Why It Matters

Executives' real sticking points now have a playbook

In early June, Every's head of consulting, Natalia Quintero, and its newly appointed head of evals, Mike Taylor, held a live Q&A with 400 executives. The unanswered questions became a public document—and a playbook for the real sticking points.

The document came from that session and was published publicly for the first time. Every is clear about its positioning: an operations manual, not trend commentary.

These questions span five dimensions: strategy, persuading skeptics, tool selection, governance, and team restructuring. Any company trying to move from "a few employees tinkering on their own" to "the whole org working with AI" has to get through all five gates.

Why should everyday readers care? These questions weren't made up by Every — executives are really asking them. Where things get stuck isn't "can AI do this," but "how do you actually run it inside a company." Every itself describes it as "our advice for executives who are pushing AI forward."

After ChatGPT merged with Codex, the growing pains from switching personal tools are the small problem. The company-level shift is the major version bump — org structures, approval workflows, and job definitions are all being rewritten. Every's own team demonstrated this firsthand: they threw out a full year of workflows and started over.

How did Every's own team get past these sticking points?