Add Sponsored tag must : The AI Strategy Blueprint
Add Sponsored tag must : The AI Strategy Blueprint
Add Sponsored tag must : The AI Strategy Blueprint
What Shadow AI Tells You About Your Own Tools
The AI Strategy Blueprint reads unauthorized AI use as a sign that employees have work they need help doing.
This review was prepared for a paid placement. The book link leads to Iternal Technologies, the author’s company.
An employee uses an unapproved assistant to turn meeting notes into a project update. The security problem is easy to see if the notes contain information the tool should never receive. An operations question sits beside it. Why did the employee pick that method, and what would the organization offer someone in the same situation tomorrow?
John Byron Hanby IV returns to this problem in the competitive-risk and governance sections of The AI Strategy Blueprint. Shadow AI is the use of AI tools or workflows outside the organization’s approved arrangements. The book argues that a prohibition on its own leaves much of the demand in place. That is an important diagnosis. An approved tool also has to solve enough of the underlying problem to earn habitual use.
The workaround has a job description
Ask what the employee was trying to finish, what material they needed, and where the result would go. The answer may be mundane. A weekly report takes too long to draft. A worker can access meeting notes but cannot find the decisions in them. A team relies on a complicated internal tool that is unavailable to someone covering a shift. Whatever the explanation, the details matter because they tell a manager what a sanctioned alternative must do.
The book favors giving employees practical, approved ways to use AI. Depending on the work, that could mean a centrally managed service or a local system with defined sources. The fit matters more than the hosting label. A local assistant may have no access to the current notes. A managed service may work well on public material but have different handling rules for confidential records. A person trying to follow the rules needs a clear answer about both the tool and the type of information allowed.
A weekly update exposes three different problems
Suppose an operations team keeps using an outside assistant to prepare a one-page handoff. A manager can reconstruct the task without asking anyone to reproduce the unauthorized upload. Use a synthetic set of notes, have the employee explain what they normally need to extract, and watch what happens with the available internal tools. The first question is whether employees can find an approved way to do the work. A rule posted on an internal site is of limited help when people need to make the choice inside another application.
The second question concerns the inputs. A sanctioned assistant may be easy to use but have an unclear policy for notes that include sensitive information. Someone has to explain which notes can be used and what belongs in the handoff. During a trial, the team could start with ordinary operational information and identify the few fields that require a different process. That gives operators a concrete boundary at the moment they handle the information.
The third problem is whether the work is actually improved. Compare the updated handoff against what the receiving team needs. If the output regularly omits who owns a decision, adding a field or a review step may be more valuable than choosing a different model. Record which part of the task improved and which step still creates friction. An approved tool earns credibility when people can see it solve a familiar problem within rules they understand.
An approved choice has to compete at the point of work
Imagine that the organization offers a replacement for the weekly update task. It should be easy to find, tell staff what it can receive, and make the review step obvious. A simple example using ordinary meeting notes can demonstrate how to detect a missing decision before sending the result. If employees keep struggling to finish, the support team should have a way to record what is missing and act on it.
Hanby’s account could distinguish more clearly between curiosity, an accidental policy mistake, and a serious disclosure. The circumstances call for different responses, and a manager will learn less if employees expect every question about a failed workflow to be treated as an offense. The data and security implications still matter. The book is most useful when its urgency turns into a well-designed alternative with clear rules and a way to ask for help.
There is a measurable outcome for leaders. Pick a recurring job that drives people toward outside tools and offer a credible way to handle it. Watch whether employees use it and whether the underlying work gets easier. If the same workaround remains necessary, the investigation has taught the institution where its supported workflow still falls short. That makes the book’s argument about shadow AI useful to the people who own the work as well as those who write the policy.

