Becoming an AI Power User

A private win helps one person until it is made legible to others. A private workflow improves your work. A documented, governable workflow can improve a team's shared capability. Writing the playbook is what turns the private win into something a colleague can run, and it sets up the two questions that follow: what stays hidden when AI use is kept secret, and how a few organizations made it visible.

The 'secret cyborg' pattern is a real productivity drain

Ethan Mollick coined the term 'secret cyborgs' for workers who use AI to boost productivity but do not tell their employers. The phenomenon is fairly well documented. Over 50% of US workers self-report some AI use at work. Many of them hide it because they fear punishment, displacement, or reputational risk.^[1]

Hidden AI use costs the team the chance to learn from it. When a process stays private, it can stay intuitive and half-articulated, and no one else can copy what works or catch where it fails. McKinsey's 'Superagency' report shows the organizational cost: executives dramatically underestimate how many employees already use gen AI for a large share of daily work. When that usage stays invisible, the standards and failure modes people are discovering never reach anyone else.

Three organizations that turned individual AI use into team capability

Several organizations have created explicit structures for making AI use visible and replicable. The patterns below circulate widely in practitioner writing; the Ramp and Roblox specifics are reported secondhand and worth confirming against a primary source before you cite them.

• Shopify incorporated AI use into performance reviews. Employees are expected to demonstrate reflexive AI use as part of their work. This removes the stigma and creates positive incentives for sharing playbooks.
• Ramp published internal 'AI power user' rankings based on measurable engagement: employees with 5+ AI actions per week across tools like Claude Code, Cursor, and ChatGPT. Making AI use legible and recognized turns hidden behavior into a visible skill.
• Roblox implemented screen-share-the-prompt meetings where team members share their AI workflows live. This creates a natural knowledge-sharing mechanism and normalizes AI use as a professional skill.

Some AI skill is tacit and often transfers faster by watching than by reading

The Roblox screen-share example points to something the documentation alone can miss: a lot of AI skill is tacit. You can read someone's saved prompt and still miss the workflow around it. It often helps to see when they pause, what they paste in, how they judge the first answer, when they switch models, what they save, and what they ignore. That kind of knowledge tends to transfer well through proximity: screen shares, pair sessions, live demonstrations, and shared experiments. Some of it can also be captured back into the playbook, by having the AI watch a session transcript and write down the moves the expert made without naming them.

Etienne and Beverly Wenger-Trayner's communities of practice research describes learning as something that develops among people who share a domain and improve through regular interaction. An AI community of practice can be as small as two colleagues who compare workflows weekly, or as structured as a team that runs regular screen-share meetings where each person demonstrates one workflow. The format matters less than the regularity. Watching someone else's AI workflow, even for five minutes, teaches things that documentation cannot capture.^[2]

Writing a workflow down lets you climb to a higher-leverage version of it

When you explain an AI workflow to a colleague, intuition that was vague has to become explicit. You name the task, the setup, the prompt, the model choice, the quality check, and the failure modes. That same explicitness is what lets you jump to a higher level of abstraction: once you can articulate what each step does and what consequence it produces, you can package the steps into a template, a reusable prompt, a workflow, or an agent instruction that runs with less of your attention. Sharing is one good way to force that articulation, and learning-by-teaching research (Nestojko et al., 2014) suggests it tends to help: people who expected to teach material recalled it better and organized it more effectively than people who only expected to be tested on it.

Sharing also puts more minds and more model runs on the same workflow. Every colleague who runs your playbook runs it on their own tasks: more experiments, more curiosity, more edge cases surfaced, more candidate standards to fold in. You can route that back into the system rather than carrying it by hand. Point the AI at a batch of recent runs and ask it to find where outputs drifted from the quality standard, then have it propose a revised prompt or rule. A team that shares playbooks tends to evolve them faster, because the workflow is improving in parallel across many people and runs instead of inside one person's head.

A shareable playbook is the bridge from individual to organizational gains

The smallest useful step between secret cyborg and visible power user is documenting one workflow and sharing it with one colleague. A playbook does not need to be comprehensive. It needs to answer three questions: what task does this workflow handle? What does the AI setup look like? What quality checks do you apply?

A good playbook is specific enough that someone can follow it without guessing. It includes the actual prompts, complete and ready to paste, the model selection, the reference materials, and the verification steps. A test of a good playbook is whether a colleague can replicate your results without you in the room.

Visibility without governance creates a different kind of risk

Making AI use visible solves the knowledge-sharing problem without solving the trust, confidentiality, or governance problem. When employees use AI tools on work data without shared rules, the risks extend beyond missed productivity: sensitive information pasted into tools with unclear data handling, outputs used without review standards, and inconsistent disclosure about what AI helped produce.

Before sharing an AI workflow with a team, document four rules:

1. Approved tools. Which AI tools are allowed for this workflow? Some organizations have data-processing agreements with specific vendors. Others prohibit certain categories of data from leaving the org. Know the boundary.
2. Data boundary. What information may never be pasted into AI tools? Client names, financial figures, health data, employee records, legal correspondence, and anything covered by NDA or regulation should have an explicit rule.
3. Human review standard. What must a human check before the AI output is used? For some workflows this is a quick scan. For others it is a line-by-line verification. Name the standard for each shared workflow.
4. Disclosure norm. When should the team say that AI helped produce the work? Some organizations require disclosure on client-facing deliverables. Others are more informal. A shared norm prevents awkward situations where one person discloses and another does not.

These four rules keep the chapter's bias toward adoption while making your shared playbooks credible to organizations that take data governance seriously. A workflow shared without governance rules may be impressive. A workflow shared with governance rules may be adopted.

How to have the AI conversation with your team

If your organization does not yet have an explicit AI policy, raising the topic can feel risky. Three approaches that tend to work:

1. Start with results, not tools. Share the outcome of an AI-assisted workflow without leading with 'I used AI.' When people see the quality and speed of the work, they are more receptive to hearing how you did it.
2. Propose a no-punishment pilot. Suggest a 30-day experiment where the team tracks AI use openly and shares what works. Frame it as a productivity experiment, not a policy change.
3. Identify one workflow that benefits the whole team. Choose a workflow where AI-assisted output is clearly better: meeting summaries, research briefs, client prep. When the team sees the benefit firsthand, the conversation about broader adoption happens naturally.

Share one workflow this week