A 'digital HR' function is a set of teams, processes, and tools dedicated to managing AI agents as if they were a new kind of workforce. As AI agents start to permeate the enterprise and interact with both humans and other systems thousands of times a day, they need structured oversight. This function focuses on: - Governance and security: Defining what each AI agent is allowed to access and do, and assigning a clear digital identity before it’s deployed. - Monitoring and observability: Tracking how agents behave in real time, across all their interactions, so issues can be spotted and corrected quickly. - Evaluation and feedback: Continuously assessing performance against business goals and adjusting prompts, data access, or workflows. - Cross-functional collaboration: Bringing together IT, traditional HR, and end users to decide where AI agents fit, how they change roles and KPIs, and how to integrate them into day-to-day work. In practice, this can be a centralized “digital HR” team or a coordinated set of functions across the organization. The goal is to ensure people and AI agents perform at their best together, rather than letting AI adoption grow in disconnected silos.

AI agents will gradually take on both routine and more cognitively demanding tasks, which means organizations need to revisit how work is structured and measured. Key shifts to plan for: - Role redesign: Early on, agents will automate routine work. For example, developers may rely on AI-generated code to ship features faster, and marketing teams might use an AI agent instead of a junior associate for competitive research. Over time, as deep reasoning improves, agents will also support more complex tasks, such as scanning medical records and X-rays to provide a preliminary diagnosis so physicians can spend more time on direct patient care. - New KPIs: Performance metrics will need to move beyond volume-based measures. In a call center, for instance, once an AI agent handles common inquiries, human agents may be evaluated less on number of calls and more on service quality, problem resolution, and uniquely human skills like empathy and complex issue handling. - Hiring and workforce planning: HR, IT, and business stakeholders need to collaborate earlier in the process. Before investing in a new AI customer-support agent, for example, IT should work with call-center staff to identify pain points and pilot specific tasks. HR then uses those insights to adjust hiring plans, job descriptions, and KPIs. - Reporting and incentives: As AI becomes embedded in workflows, reporting lines and incentives may need to change so teams are rewarded for effectively using AI agents, not just for individual output. Overall, the organizations that benefit most will be those that treat AI agents as part of the workforce design conversation, not just as another IT tool.

You can think about AI agents much like hiring decisions, with three main options: build, buy-and-train, or outsource basic capabilities. 1. Build for mission-critical, differentiating use cases Reserve your largest AI investments for agents that directly drive competitive advantage or core outcomes. This includes: - Use cases that are tightly tied to your unique domain, data, or processes. - Scenarios where off-the-shelf tools can’t meet your performance, security, or integration needs. In these cases, you’re effectively building an AI leadership team: in-house specialists who design, deploy, and scale agents that are specific to your business. 2. Buy and train when you need customization but not a full build Sometimes off-the-shelf AI isn’t good enough out of the box, but the ROI doesn’t justify a fully custom build. Here, you: - Start with a commercial or open source agentic solution. - Use your proprietary data and internal expertise to train, tune, and configure it. This is similar to hiring associates and training them with guidance from more tenured employees. 3. Outsource basic capabilities through existing software vendors Most enterprise platforms (CRM, HCRM, and others) are adding embedded AI features. These can: - Improve performance within that specific software ecosystem. - Handle standard, non-differentiating tasks. You should invest in these where they make sense, but recognize that they’re usually limited to that vendor’s environment. When possible, connect data from these silos into a unified platform so you can build higher-value applications on top of shared assets. Across all three options, you still need strong governance, monitoring, and observability. Just like human hires get a digital identity and access profile before day one, AI agents need clear permissions and real-time oversight before they’re put into production.