8 urgent updates your IT playbook needs to survive in the AI ​​age

Follow ZDNET: Add us as a favorite source On Google.

ZDNET Highlights

• AI may make your technology playbook obsolete.
• Be open to revisions that help people check their ideas.
• Focus on key areas like use cases, data sources, and training.

Do you or your team use a technology playbook? If yes, what’s in it? There’s a good chance that your playbook is quickly becoming outdated.

Also: 10 ways AI could cause unprecedented harm in 2026

This is the challenge recently posed by Thomas Earle, a prolific technical writer and teacher. Interview With Matt Strippelhoff, Partner and CEO of Red Hawk Technologies. Erle calls for new playbook revisions and tried-and-true practices to help AI proponents and developers test their ideas, run safe pilots, and prove the return on investment of their projects.

Playbooks, whether formal or informal, detailed or simple checklists, ensure everyone works from the same page strategically for consistent operations and deployments with strong security policies. However, in today’s increasingly digital world, you may need to revisit those guidelines if you or your team is working with AI.

Also: Worried about the job market? 5 ways to stand out in the age of AI

A Playbook for the 2026 Enterprise contains many new requirements, but it also builds on previous IT guidelines. Strippelhoff and Earle review some of these ideas.

8 guidelines for the AI ​​era

1. Start with a meaningful problem: Identify where AI will really make a difference, versus AI for AI’s sake. “Some companies are looking for ways to implement AI, but they haven’t identified the problem they want to solve,” Strippelhoff said. “So, they have a solution looking for a problem. Traditional strategic planning is important to make sure you’re identifying a meaningful problem.”
2. Start with desired outcomes and prepare a business case: This approach was common for earlier technologies, but with AI initiatives it takes on added urgency. “The most important thing is to understand the organization’s readiness for the idea,” Strippelhoff said. “Someone needs to take the time to prepare and define that vision. Then you need to engage subject matter experts in those systems, data sources, and other things, and determine whether you’re really ready, whether it’s time to make the investment. Often, a lot of organizations are not as prepared as they may think they are.”
3. Include an extra layer of caution: AI is not just about building and running software. Deploying AI also means diving into an organization’s deepest wells of knowledge. Training data comes from those wells, Strippelhoff said: “It also includes a means of validating the responses generated or being produced by the AI.”
4. Make room for exceptions: This is an area where even the most well-planned AI systems can stop working. Insufficient data quality, for example, can lead to significant inconsistencies in AI output, Strippelhoff warned: “Exceptions in the quality of your data can create a lot of challenges for training AI models.”
5. Time for AI model training include: People need assurance that the training data is fresh and accurate. For example, Strippelhoff said, in the healthcare sector, a wide range of billing codes makes automated revenue-cycle management challenging, “because there are thousands of codes to choose from.” As a result, the process must be closely monitored by humans until assurance is obtained that the code is properly classified through an ongoing feedback loop.
6. Make sure your data is ready: “Some companies may believe that with their digital assets, standard operating procedures and governance in place, they are ready to move forward with an AI initiative, only to find that their data is in such a poor state that they have to ‘do’ the project. I have seen projects stalled or permanently shut down because of this.
7. Always keep humans in the loop: AI may seem synonymous with complete automation, but it is not. An essential part of the AI ​​output validation process is to maintain human oversight at key points. This will likely be a “subject matter expert who validates the output,” Strippelhoff said, “which takes time to train.”
8. Check platform limitations: “If your solution is dependent on extracting and transferring data into the system via an API endpoint, there may be limits to the number of calls, the availability, and the type of information you can retrieve, as well as the frequency at which you can retrieve it.”