Recently, the CIO Professional Network, a private community for CIOs, CISOs, and CTOs, convened another Roundtable discussion. Hosted by Jerry Heinz and William Novak, the focus was on the integration of AI skills and workflow tools within business settings. Through live demonstrations and interactive dialogue, the session aimed to empower leaders in creating AI systems that are not only repeatable but also governed to enhance daily operations.
The dialogue illustrated how AI can function as an operational layer that can be designed and audited much like other enterprise technology systems.
Participants discussed various topics, including skill design, governance of agents, automation of workflows, version control, and the adoption of enterprise AI across different business functions. Examples such as daily news digests and calendar synchronization illustrated how well-structured AI workflows can add significant value, particularly when accompanied by robust safeguards.
Why This Matters: AI platforms open up opportunities to streamline processes, which in turn presents leaders with a new set of responsibilities. The potential is in automating mundane tasks and enhancing team responsiveness. However, it is crucial to ensure that these systems are managed with the same rigor as any other enterprise application. This Roundtable underscored that the true advantage lies in developing AI tools that are structured, repeatable, and aligned with business needs.
- Skills Transform General AI into Specialized Workflows: Attendees noted that regardless of how a platform labels them—skills, custom GPTs, Gems, or Copilot agents—the underlying concept remains the same. These are specifically tailored workflows designed for particular tasks. By providing a fixed role with clear instructions and expected outputs, skills yield more consistent and useful results over time. The group elaborated on how this enhances AI’s utility as a reusable process rather than merely a chat tool.
- Robust Skills Require a Defined Framework: A clear structure for skill design emerged as a critical theme during the discussion. Participants learned that an effectively constructed skill offers the agent precise directions, the background necessary to grasp the task, and defined boundaries concerning access or actions allowed. The design allows for specific actions when required—such as searching for information from sanctioned sources or carrying out established workflows. Attendees recognized that these instructions could be organized into simple markdown files, facilitating a process approachable even for non-programmers to refine over time.
- AI Enables Teams to Achieve More Without Expanding Staff: One participant shared that his journey began when downsizing compelled him to take on additional operational responsibilities while still improving his team’s productivity. This led to automations for onboarding and log analysis, which eventually evolved into more sophisticated AI-assisted workflows. Other leaders echoed similar experiences, emphasizing how well-designed agents can enhance staff capacity to perform repetitive tasks that would otherwise require additional hires.
- Early Governance and Access Controls are Crucial: Many questions from the audience centered on compliance and monitoring actual accomplishments. Jerry explained that his systems rely on narrowly defined permissions, established tool access, and platform hooks that track agent activity. This means agents can only access approved contexts and systems, with permissions administered similarly to any other user or application identity.
- Version Control and Observability Foster Safe Adoption: When discussing the maintenance and enhancement of these systems over time, an attendee highlighted the importance of Git-based versioning. Using Visual Studio with automatic saves to Git creates a preserved version history, allowing for reversion of unwanted edits. Participants also described employing monitoring and local observability tools to scrutinize agent behavior. Together, these controls contribute to managing AI workflows safely as they become increasingly complex.
- Tailored Platforms Meet Diverse Business Needs: Participants illustrated how similar concepts can be adapted to various professional services environments. Using Claude for generating instruction sets and transitioning those into Copilot, members described tools for tasks like accounting news management, contract dashboards, HR policy assistance, and insurance policy reviews. These insights suggest that organizations may require multiple AI platforms based on specific tasks and existing systems.
- Enhanced Outcomes Begin with Comprehensive Requirement Gathering: Some attendees shared strategies for refining agent design. A key recommendation was to prompt the model to ask the user questions prior to building a skill, which helps clarify requirements that could be overlooked. Participants noted that voice prompting tends to elicit more nuanced responses than typing. Additionally, limiting the number of questions posed by the model can provide sharper focus and facilitate a smoother process.
