The skills your team needs to work effectively with AI are not the same as the skills they needed two years ago, and they will not be the same in two years from now as they are today.
That sounds like a problem. It is actually an opportunity, if you move on it now rather than waiting until the gap is obvious.
At Enterprise DNA, we have trained more than 220,000 professionals in data and AI skills. The pattern I see consistently is that the teams that have invested in systematic upskilling adapt to new AI tools faster, produce better output from those tools, and catch errors that undertrained teams miss entirely. The technology is the same. The results are significantly different.
The 2026 AI coding landscape makes this urgent in a specific way. Uber confirmed that 70 percent of its committed code is now AI-generated, with 84 percent of engineers using AI tools monthly. Claude Code doubled its weekly active user base in a matter of weeks. Anthropic’s Claude Partner Network generated 10,000 certifications in its first weeks. These numbers tell you that AI tools are being adopted at a pace that is outrunning most organisations’ training programmes.
This guide is about how to close that gap systematically.
Start by auditing where your team actually is
The most common mistake in upskilling programmes is building them around assumed skill gaps rather than actual ones.
Before designing training, run a simple audit. Ask your team three questions:
What AI tools are you currently using in your day-to-day work? Not which tools are available. Which ones are you actually using. There is often a significant gap between what has been provisioned and what is being used regularly.
When an AI tool gives you output — a code suggestion, a document draft, a data analysis — how do you evaluate whether it is correct? What does your review process look like?
What tasks do you currently avoid giving to AI tools because you do not trust the output? Why do you not trust it?
The answers to these questions will tell you more about where your training investment should go than any capability framework built from scratch.
Common gaps that show up in this audit:
Teams that use AI tools but rely entirely on their outputs without a review framework. They get faster outputs and also deploy more errors.
Teams that have strong review processes but have not adopted AI tools at all. They are being left behind on productivity while having the fundamentals to use these tools well.
Teams that use AI for some tasks but have a set of high-value tasks they have written off as AI-inappropriate — often incorrectly. The tools have advanced, but the team’s mental model of what they can do has not kept up.
The three skill layers that actually matter
When I talk about AI skills, I mean three distinct things that require different training approaches.
Layer 1: Tool proficiency. Knowing how to use the specific tools your organisation deploys — how to write effective prompts, how to use the context features, how to configure the tool for your specific workflow. This is the most teachable and the most frequently overinvested in. Teams often spend training budget here because it is concrete, but it is the layer with the shortest half-life. Tool interfaces change faster than anything else.
Layer 2: Output evaluation. Knowing how to identify when an AI output is wrong, incomplete, or suboptimal. For coding tools, this means understanding the code well enough to spot the subtle errors that look correct on first read. For generative content tools, this means knowing enough about the domain to recognise when the AI is confidently wrong. For data analysis tools, this means understanding the data well enough to spot when the AI has misunderstood the question or mishandled an edge case.
This is the layer that most training programmes underinvest in. It requires domain knowledge, not just AI tool knowledge. And it is the layer that determines whether AI adoption produces better outcomes or just faster ones.
Layer 3: Workflow redesign. Knowing how to restructure work so AI is used where it creates the most value and humans are used where human judgment is genuinely irreplaceable. This is a management capability as much as an individual one. It requires understanding what the tools can reliably do, what they cannot, and how to design processes around that distinction.
A training programme that only addresses Layer 1 makes your team faster at using AI tools and does not necessarily make them better at producing quality output. A programme that addresses all three layers builds genuine capability.
Build a structured learning path, not a collection of resources
The most common failure mode in internal AI upskilling is assembling a collection of resources — courses, videos, tutorials, articles — and calling it a training programme.
Resources are not programmes. A programme has a defined learning path, clear outcomes, assessment at each stage, and accountability for completion.
For a team that is starting from a low baseline, a twelve-week structured path looks like this:
Weeks one and two: Foundations. Data literacy basics, AI tool landscape orientation, and hands-on time with the specific tools your organisation uses. The goal is to get everyone to a baseline of competency with the tools.
Weeks three and six: Applied practice on real work. Each person applies AI tools to actual tasks from their current job, with structured reflection on what worked and what did not. This is where the output evaluation layer starts to build — through practice with real output, not manufactured exercises.
Weeks seven and ten: Peer learning and error analysis. Teams share examples of AI outputs they found difficult to evaluate, cases where the AI was wrong in subtle ways, and workflows they have redesigned around AI. This builds the collective intelligence about where the tools are reliable and where they need more careful human review.
Weeks eleven and twelve: Workflow redesign workshop. Each team maps their current workflows and identifies where AI creates the most value, where it creates risk, and what the redesigned workflow looks like. This is the output that actually changes how work gets done.
That structure takes a team from tool uncertainty to genuine capability. It is more investment than sending people a list of courses, but the results are not comparable.
Address the data literacy gap explicitly
The AI tools your team uses are only as useful as your team’s ability to work with the data those tools produce.
This is an area where I see consistent gaps. Teams that are eager to use AI for analysis do not have strong enough data literacy to evaluate the analysis they get back. They accept numbers that are technically correct but misleading because the framing was wrong. They miss issues that a data-literate reviewer would catch immediately.
Data literacy is not about being able to write SQL or use Python. It is about knowing how to ask good questions of data, how to spot when a dataset has not been cleaned, how to identify when a correlation is being presented as causation, and how to understand the limitations of a given analysis.
At Enterprise DNA, data literacy has been the foundation of our training for years, and it is the thing that consistently differentiates teams that use AI well from teams that use AI fast. The fast teams produce impressive-looking outputs quickly. The skilled teams produce accurate outputs that hold up when scrutinised.
If your team’s data literacy is not strong, investing in it before or alongside AI tool deployment produces better returns than AI tool investment alone.
Create accountability for AI capability development
Upskilling programmes fail when completion is voluntary and accountability is absent.
Make AI capability development part of how you measure team performance. Not in a punitive way, but in the same way that you measure any professional development that matters for the role. If being able to work effectively with AI tools is a meaningful part of your team’s job function — and in most knowledge-work roles, it is — then building that capability should be treated as a job requirement, not an optional enhancement.
Quarterly capability reviews are a simple mechanism. Each person can demonstrate competency on their tool set, show examples of their output evaluation practice, and present one workflow improvement they have made using AI. This creates a cadence without being burdensome, and it keeps the conversation active rather than treating upskilling as a one-time event.
The teams that build AI capability fastest are the ones where it is treated as a professional expectation, not a bonus. The tools are changing fast enough that a single training sprint will not maintain currency. Building the expectation of ongoing learning is the culture shift that makes sustained capability development possible.
Enterprise DNA’s learning platform has trained 220,000+ data and AI professionals. If you want to build a systematic upskilling programme for your team, let us show you what that looks like in practice. Or book a session with me to design the right programme for your specific team and context through Omni Advisory.