When someone asks what AI skills employees need, the typical answer is a list of tools. Learn prompt engineering. Get comfortable with ChatGPT. Understand how large language models work. The list usually runs about ten items long, and almost none of it explains what will actually change how your workforce performs.
Sharma Vedula, Head of Solutions at Data Society, has a different view. After years of working with enterprise and government clients on AI implementation and workforce development, he has seen what separates organizations that successfully build AI capability from those that spend money on training programs and see nothing change.
His answer is less about which tools people learn and more about how they learn them, and who gets brought into the process first.
“They need to invest in people before they invest in technology. That should be the biggest takeaway because the more you loop in your teams upfront, not after the fact, the more you will get a better buy in and you’ll get more patience through from them because they would understand.”
That framing shifts the AI skills conversation from a curriculum problem to a change management problem. And it changes how enterprise leaders should think about building capability across their organizations.
THE SKILLS THAT ACTUALLY MOVE THE NEEDLE
There are two categories of AI skills that enterprise employees genuinely need, and they are not the same as what appears on most upskilling curricula.
The first is practical AI literacy: the ability to work with AI tools effectively in the context of a specific job. This is not generic AI knowledge. It is the understanding of how a particular tool behaves, where it fails, and how to use it well for tasks that employee actually does. A financial analyst using AI to model scenarios has different literacy needs than a customer service rep using AI to draft responses. Treating both the same way produces training that is too broad to be useful for either.
The second is critical judgment: knowing when to trust AI output, when to verify it, and when to override it. This is the skill that organizations most consistently underinvest in. Teaching people to use a tool is straightforward. Teaching them the instinct that something is wrong when a tool returns a confident-sounding answer that should not be trusted is harder, and more valuable.
Both skills require hands-on practice in context, not passive instruction. Organizations that design for this get different results than those that default to self-paced e-learning.
Related reading: AI Upskilling vs. AI Implementation: Why Enterprises Need Both and in the Right Order
THE PROBLEM WITH “LEARN THIS OR YOU ARE OUT”
A pattern is emerging in enterprises where employees are handed access to AI tools with instructions to figure it out, or worse, told that failure to adopt AI will cost them their jobs. Vedula is direct about why this approach backfires.
“I wouldn’t go with that message that you know you won’t have a job if you don’t use it but I would encourage them by showing them how to use it because there are still a lot of unknowns and there are some best practices out there that you may not have to teach them every single way to do it but at least sharing some best practices would help them not only understand how to use those AI tools but use them efficiently.”
The efficiency point matters more than it might seem. Most enterprise AI tools are priced on consumption. Employees who do not know how to use them well do not just produce worse output. They also drive up costs. An employee who prompts a model inefficiently burns through tokens and time and still delivers subpar results.
A fear-based message also destroys the psychological safety employees need to experiment, ask questions, and actually learn. AI tools require a certain amount of trial and error to understand well. Employees who are afraid of making mistakes will avoid the tools or use them superficially. Neither outcome produces the capability the organization is trying to build.
WHAT AN EFFECTIVE AI SKILLS PROGRAM LOOKS LIKE
The alternative to “figure it out” is not a series of mandatory modules. It is a structured, iterative approach that brings employees into the process and keeps them there.
Vedula describes what this looks like in practice:
“Have a cadence where you meet with them try to understand what their issues are if they have any and maybe that’s how you sort of teach them you don’t have to actually teach them the beginning but you know have a constant sort of loop back on how things are progressing and how things can be done.”
This is less like a training program and more like a coaching model. The organization establishes a regular cadence for checking in, gathering feedback, and addressing the specific friction points employees hit in their real work. The training evolves based on what people are actually struggling with, not a predefined curriculum.
This approach also produces better adoption. When employees feel supported rather than pressured, they are more likely to engage genuinely with the tools and bring their colleagues along.
Related reading: How to Build an AI-Ready Workforce in 2026: The Enterprise Leader’s Complete Guide
THE PROOF: WHAT STRONG AI UPSKILLING ACTUALLY PRODUCES
When AI skills programs are built well, the results are measurable. Data Society’s work with the Department of Health and Human Services offers a clear example.
The COLLAB program started as a focused boot camp for 25 employees across different parts of the agency, including NIH and the Office of the Secretary. The goal was not to hand off technology and hope people figured it out. It was to build internal capability so that when tools were applied at scale, people already understood the data, the workflows, and the decisions they were supporting.
Capstone projects from that first cohort generated more than $500,000 in annual cost savings and freed up four full-time staff positions. When the program expanded, it received 450 applicants for 30 available spots. That number is not incidental. When training is genuinely useful and career-relevant, employees want it. Demand becomes a signal of quality.
That kind of result does not come from assigning courses and tracking completion rates. It comes from investing in building real capability in a structured, supported environment. For enterprise leaders evaluating their own AI skills programs, the core question is whether the program is designed to check a box or to produce the kind of capability that shows up in business outcomes. The two require fundamentally different approaches.
See the full landscape of enterprise AI readiness in The 2025 AI Readiness Report
Frequently Asked Questions
The most valuable AI skills for enterprise employees are practical tool literacy specific to their role, critical judgment about when to trust or verify AI output, and the ability to apply AI to real workflows. Generic AI literacy matters far less than role-specific applied capability.
Start with best practices specific to the tools your organization uses. Build in regular check-ins where employees can surface friction. Focus on the capabilities that produce the most value for each role rather than trying to cover everything at once.
Mandating training rarely produces genuine capability. A more effective approach is creating structured opportunities, sharing best practices clearly, and building a cadence that supports employees through the learning curve. Employees who understand the value of AI for their specific work will engage voluntarily.
Completion rates are not a useful metric. Look for changes in how employees use AI tools on the job, reductions in friction or errors, and whether employees are surfacing ideas for AI applications in their own work. Program demand is also a strong signal: programs that are genuinely useful generate word-of-mouth.
AI literacy is awareness: understanding what AI is, how it works broadly, and what its limitations are. Applied AI skills are the ability to use specific tools effectively in the context of a specific job. For enterprise performance, applied skills tied to real workflows produce more measurable business value than general literacy alone.
READY TO BUILD AI SKILLS THAT ACTUALLY STICK?
Most enterprise AI training programs are designed to demonstrate activity, not produce capability. If you want to build AI skills that translate into real business outcomes, the approach is fundamentally different from assigning a course library.
Data Society works with enterprise and government organizations to design and deliver AI upskilling programs built around the way people actually work. Talk to our team about what a structured AI capability development program looks like for your organization.
