What a Cowork rollout actually involves

The work behind the rollout

An AI rollout needs decisions about access, training, permissions, and data. This issue looks at Petti's framework for rolling out Claude Cowork across a company, and at GC Academy, where we're inviting readers to test courses.

What's happening in AI

What a Cowork rollout actually involves.

Claude Cowork launched as a research preview in January and became generally available on April 9, bringing the approach behind Claude Code to work outside software development. You can hand it a multi-step task involving files and connected tools, then review what it produces. That changes the assignment from asking for an answer to delegating a job, with more responsibility for deciding what the AI can access and do.

Companies exploring that way of working have a rollout to plan. In its April announcement, Anthropic described early adopters who saw Cowork take hold in one team and wanted to roll it out more broadly, and it introduced controls for access, spending, usage visibility, and connector actions. That is the practical work behind making these AI products useful across an organization, whatever product a company chooses.

Alex "Petti" Pettigrew has been doing that work in the field since these AI products' early days. His November 2 video brings the patterns he has noticed into a five-part rollout framework. He uses a fictional company to walk through access, training, connectors, a marketplace for shared skills, and the data foundation beneath them.

Access comes first. Choose the product, decide whether to start with a pilot, and plan how people will get the usage they need. At 2:40, Petti moves to training. He describes people working alongside employees on their actual workflows until the process clicks. Assign time for that hands-on work when planning training.

The connector discussion at 4:13 separates two decisions: what information AI can see, and what actions it can take. An AI product might prepare a change while a person retains the final approval. Connecting a system means deciding those boundaries, not simply turning on an integration.

At 6:48, the framework turns to a marketplace for shared skills, reusable instructions for recurring work. We've covered skills management on our site: Enterprise AI skills have a management gap.

The fifth part is the data foundation. In his fictional example, records for the same organization live in different systems without reliable links between them. Connecting all those systems still leaves the AI trying to work out which records belong together. At 16:38, Petti argues for treating the data model as something the company actively maintains. The relationships between records need attention even if the AI product changes.

These activities can overlap. Petti says the framework need not run as a strict sequence once access is in place. The fictional example is not evidence of adoption rates or measured results.

By the way: GC Academy

Five self-paced courses are available at GC Academy: AI Fundamentals for Everyday Work, AI Changemaker, Citizen AI Developer, How to Become an AI Consultant, and Your Experience, Reinvented: Position Yourself for the Future of Work.

We're looking for people to test GC Academy courses for free. Reply if you're interested, and tell us which course you'd like to try.