The Gap Isn't Access

IBM's 2026 CEO Study found that business leaders cite AI skill gaps as the single largest barrier to AI value creation, ranking it above technology limitations, budget constraints, and data quality issues. CEOs who have solved the access problem — bought the licenses, stood up the infrastructure, issued the mandate — report that skills are still what separates the companies capturing value from the ones watching it happen to their competitors. The instinct is to read this as a training problem. It isn't. And the distinction matters, because the wrong diagnosis produces a very expensive wrong prescription.

The skill gap IBM is documenting isn't the gap between employees who know how to use ChatGPT and employees who don't. That gap is largely closed. Most employees with a laptop and a browser have used an AI tool in the last ninety days. The gap IBM is actually naming, if you look at what underlies the CEO frustration rather than the shorthand that summarizes it, is the gap between general AI familiarity and the ability to deploy AI at workflow level. These aren't adjacent skills. They don't even overlap much.

Prompting is a skill, not the skill

Knowing how to prompt a language model well is a skill. It's not the skill. The skill that produces the outcomes BCG's 5% are generating is the ability to look at a specific business workflow, identify where the manual density is highest, design an AI-augmented version of that workflow, integrate the necessary data feeds, write the specification tightly enough that the tool does what the workflow requires, define a quality standard for outputs, and own what happens when the outputs drift. That's a systems design capability layered on top of operational knowledge layered on top of domain expertise. It's rare. It's not produced by training programs that teach employees to use AI assistants more effectively.

IBM's CEOs know they don't have enough of this capacity. What they're calling a skill gap is really an implementation design gap. The people who can close it aren't sitting in corporate L&D programs waiting to be upskilled. They're in agencies that have done this work before, built the methodology through trial and production failure, and can bring the capability to a company that doesn't have time to develop it internally.

Why training and hiring don't close it

This reframes the question companies are actually facing. The typical response to a reported skill gap is a training intervention: buy licenses for an AI course platform, run a lunch-and-learn series, assign mandatory completion metrics, report to the board on percentage of employees trained. None of this is wrong, exactly, but it doesn't produce the implementation design capacity the gap actually requires. You can train everyone in a company on AI prompting and still have no one who can redesign the client intake workflow around an AI classification layer and own the output quality on an ongoing basis.

The second typical response is hiring: post a job description for an "AI lead" or "Chief AI Officer," run a search, wait four to six months, hope the hire can build what's needed. This is more promising but slower than the problem requires, and it doesn't solve the agency quality question that comes immediately after: the AI lead needs to evaluate vendors, and they face the same vetting problem the company had before they arrived.

Access was never the problem

The access framing in IBM's study is worth examining precisely because it reveals how the problem is being construed. Executives understand access problems. They've solved access problems before: deploy the infrastructure, license the software, remove the barriers to usage. The 85% AI tool access rate IBM documents is an access problem solved. Twenty-five percent regular use is the evidence that the underlying problem was never access.

The organizations that cross from 25% usage to embedded workflow transformation don't do it by improving access further. They do it by redesigning specific workflows so that the AI has a home inside how the work gets done, not alongside it. A tool with 85% access but no home in the workflow stays optional. A tool with 40% access that sits inside a redesigned process runs at whatever rate the process runs.

Workflow redesign at that level of specificity requires someone who has done it before, can show what the output looks like, and can own the quality of what gets built. That's not the tool provider. It's not the IT team. It's an implementation agency with a documented discovery methodology and a delivery track record. IBM's CEO Study named the gap accurately. The prescription it implies, though it doesn't state it, is a different kind of engagement than a training vendor offers.

The barrier is the path

That's the gap IBM named without quite finishing the thought. The barrier isn't the technology and it isn't the budget. It's the path to the specific capability that can design and deploy AI at the workflow level for a company that doesn't have the internal depth to do it themselves — and the ability to tell, before signing, whether a given agency actually has that capability or just describes it well.

---

If you want that question answered for your specific situation, the Forge Playbook does it. Answer a few questions about your business and we'll put together a tailored outline of which workflows are worth automating and what a realistic budget looks like for each. Free, no obligation, takes about three minutes.

Get your free Forge Playbook →