Why we need an “HR department” for AI agents, and how this accelerates AI adoption

Why do 74% of companies still struggle to achieve tangible value from generative AI in 2025? While AI promises to be “transformational”, a survey found that knowledge workers believe AI has improved efficiency 35% less on an organizational level compared to the personal level.

For many organizations, AI transformation has become a key company-wide initiative for 2025. But the problem is there’s no formal playbook yet for how to actually deploy AI across an organization at scale and get real value, leaving the vast majority of enterprises stuck in the proof-of-concept phase.

So what should leaders do to close that gap? 

Consider creating an “HR department for AI agents” — a concept to think about how to embed AI across the organization. Just like how a traditional HR department is responsible for setting new hires up for success, the HR department for AI agents is accountable for getting AI to successfully integrate across the entire organization and achieve full buy-in. 

This doesn’t necessarily mean there needs to be a discrete department, but the responsibilities of a regular HR department can be mirrored for managing AI agents and integrated into existing roles, or into new roles created specifically for this purpose (i.e. an AI deployment task force).

Who Runs this Department?

For AI to succeed, there needs to be clear ownership — whether that’s a single person, or ideally, a dedicated team leading the initiative across the organization. NVIDIA’s founder, Jensen Huang, has suggested that IT should take on this role for AI. This makes a lot of sense, given IT’s expertise in cloud infrastructure, data pipelines, cybersecurity, and vendor management. From our conversations with Credal customers, we’ve found that IT often plays a central role in AI adoption.

However, the responsibility doesn’t always fall on IT. In practice, we see a range of ownership designations, including:

1. IT departments
2. AI/ML platform teams 
3. AI task forces, made up of business / operations leads
4. Individual department heads choosing their own tools for their teams
5. Engineering leaders, looking after AI adoption as part of their broader responsibilities

We go into detail about this in our other blog post on getting to 90% AI adoption. No matter who’s running it, establishing an “HR department for AI agents” brings much-needed structure and governance needed to actually capture value from AI.

The New HR Playbook for AI

If AI agents are to function as real employees to the organization (which is the goal), then they need structured management. That means applying core HR principles to ensure AI is effectively integrated for long-term success.

1. Recruiting (Procurement)

AI procurement can be treated like executive recruitment — very resource-intensive and requires close alignment with the organization’s long-term strategy to work well. Purchasing an AI tool requires a very careful decision making process because it means committing to integrating it into a company’s workflows, data pipelines, and employee training.

Who owns it?

• IT departments
• AI/ML platform teams 
• Individual business departments 

Why does it matter?

• AI procurement is generally more effective long-term when there’s centralized oversight. While individual business units can adopt AI solutions more quickly and tailored to departmental workflows that fit their immediate needs, the tradeoff is that it can result in fragmented systems, data silos, and integration challenges over time
• Having a good procurement process ensures that AI can be a scalable investment that supports long-term enterprise-wide transformation. This is similar to how HR ensures that new hires fit into the company’s long-term workforce strategy, rather than just filling short-term gaps.

We have another blogpost called the “AI tool scorecard” that shares what specifically to look for when shopping for an AI platform.

2. Onboarding (Deployment & Integration)

When an employee onboards, they get a company email and software logins to access the organization’s data; similarly, an AI platform will need access to relevant databases, APIs, or internal software platforms so it can access the information it needs (customer data, transactional logs, sensor data, etc.). 

Who owns it? 

• IT / DevOps / Infrastructure teams

Why does it matter?

• Proper system integration guarantees that the AI platform can access up-to-date data, which is critical for incorporating it in daily workflows
• The organization needs to implement role-based access controls and secure credentialing for proper data governance

3. Training, Organization-Wide (Employee Enablement)

Similar to regular organization-wide training, where a new hire is introduced to the company’s culture and procedures, employee enablement is designed to bring everyone in the organization up to a baseline level of AI literacy — covering how LLMs work, what AI tools are available at work, the company’s governance policies, and how to incorporate AI as part of everyday work.

Who owns it?

• IT directors 
• HR / Learning & Development
• *Less commonly, vendors provide this training. However, at Credal, we offer deployment strategists to support organizations in this process

Why does it matter?

• An enablement plan can create a culture of AI utilization as people configure shareable AI agents and teach each other 
• Many people associate AI solely with chatbots, but there’s actually a lot of untapped power in AI tools that connect to your internal data; proper training can help employees learn how to tap into these tools and get daily usage
• Positioning AI as a “digital employee” can make employees more effective at getting value out of AI. 53% of people in the workplace see AI as a tool, while only 11% see AI as a teammate (but this number is growing). Those who see AI as a teammate are more likely to get productivity gains because they’re thinking about scalable ways to invest in the AI tool to get it to be more productive
• Governance policies, and making sure employees understand them, eliminates hesitation around using AI. When people are not sure what they are allowed to do, they tend to do nothing at all (counterintuitive, but empirically, we’ve found this to be true)
  

4. Department Training (Specific Use Cases)

When a new employee onboards, they receive broad training on the organization’s culture and processes. Similarly, most LLMs are initially “pre-trained” on vast, general datasets—akin to that broad onboarding process. But just as every department has unique requirements (customer support, operations, engineering, etc), these models need customized, in-depth training on specific domain expertise.

Who owns it? 

• Subject matter experts (SMEs) – they provide domain context and training data, and evaluate the AI’s outputs for accuracy in production (ex: The finance lead can upload new regulations or internal policies to the AI assistant, effectively “teaching” the AI to flag potential compliance risks in future transactions)

Why does it matter?

• Training AI with subject matter expertise makes it more accurate and reliable, especially when data stays synced and up to date, and this increases the AI’s value over time
• Designating “AI champions” (in this case, the subject matter experts) in each department helps employees in that department adopt the tool more quickly and confidently 

5. Performance Monitoring (Tracking KPIs) 

The goal of performance monitoring is to make sure an employee is working as expected - outputting at a high quality, or to provide constructive feedback for areas of improvement. Similarly, AI platforms must also be evaluated regularly to make sure performance is as expected. Depending on the application, KPIs might be accuracy (for Q&A), internal NPS, or utilization (% of the org using the platform, frequency of usage per active user, etc.)

Who owns it?

• IT directors – to track overall usage and ROI across the entire organization 
• Business stakeholders / department heads – to monitor specific departmental outcomes linked to AI usage (cost savings, product growth, customer satisfaction, etc.).

Why does it matter?

• Without tracking KPIs, it’ll be difficult to assess what’s going well or poorly, and why. If AI usage is low across the organization, is it due to poor training, lack of awareness, unclear governance policies, or a tool that’s too complex or unreliable? 
• If one department is rapidly adopting AI, performance tracking can determine what’s driving it. Is it a strong internal champion or they’ve found particularly high-value use cases? This can allow for replication across other departments.
• Tracking KPIs is crucial for proving ROI of AI initiatives to executive leadership, and to advocate for additional resources—whether that’s upgrading infrastructure or expanding deployment teams

Why This HR Approach Accelerates AI Adoption

Everyone in an organization understands (or at least has experienced) recruiting, onboarding, training, performance reviews, compliance, and development. Mapping AI processes onto these established concepts makes AI feel more like a part of standard operations.

Just as HR processes have well-defined owners (e.g., recruiters, hiring managers, trainers, compliance officers), the AI adoption curve can mirror that level of clarity. Each step has accountable parties, objectives, and KPIs.

Only 1% of C-suite leaders say their organization has mature internal gen AI, according to a McKinsey report — a huge gap between AI’s potential and reality. The HR analogy ensures that AI initiatives aren’t just short-term experiments; they follow a proven organizational template for acquiring and retaining critical resources (in this case, AI capabilities). Employees aren’t simply “hired” and left to their own devices. They’re onboarded, trained, monitored, developed, and governed. Similarly, AI systems are not “one-and-done” solutions but continuous investments. The HR framing enforces that mindset. This view helps sustain funding, executive attention, and a roadmap for unlocking actual productivity gains. AI shouldn’t be treated as a static purchase but rather continually managed as a “digital employee”.