AI readiness profiles: the four employee types your training must fit
Candova's 2026 State of AI Jobs and Skills Report, conducted with Study.com, shows the workforce is not one population. Four AI readiness profiles emerge from the data, each with a different bottleneck and a different fix.
In short
AI readiness profiles sort a workforce into four distinct groups, and matching training to each profile is what separates hours saved from a rework queue.
- The four profiles: the AI beginner, the occasional user, the frequent user, and the power user.
- Only 18% of employees feel fully prepared to use AI well, so most of the workforce sits in the two lower-readiness profiles.
- Every profile names the same two gaps, practice on real work and a clear quality standard, from very different starting points.
- Treat everyone as one group and a single program misses in both directions: too basic for some, too advanced for others.
AI readiness profiles are the four groups a workforce splits into once you look at how people actually work with AI, and they are the input most AI training leaves out. They come from the Candova 2026 State of AI Jobs and Skills Report, conducted with Study.com, which surveyed 1,000 employees across industries. The report found a workforce with the tools but not the training: nine in ten employees use AI at least sometimes, only 18% feel fully prepared to use it well, and nearly one in three save no net time from it.
The full dataset behind those numbers, covering adoption, time saved, confidence, barriers, and methodology, lives in Study.com's 2026 State of AI Jobs and Skills Report. This page is about what to do with it. The most useful thing the data does for anyone planning training is stop treating the workforce as one population. Sorted by how people work with AI, four readiness profiles emerge, each with its own bottleneck.
Treat all four as one group and you get one program that misses in both directions: too basic for some employees, too advanced for others. Build around the profiles instead, and the same AI license that returns real hours for one person stops generating a rework queue for another.
The gap the profiles have to close
Readiness here does not mean expert skill. It means feeling equipped for the AI demands already in the job. By that bar most of the workforce is not there yet, so a single training track fails predictably. Adoption is not the constraint: only 5% of employees work somewhere AI is discouraged, and 21% are now required to use it for core responsibilities.
| Self-reported AI preparedness | Share of employees |
|---|---|
| Fully prepared for nearly any task | 18% |
| Prepared for most tasks | 33% |
| Prepared for basic tasks only | 35% |
| Not prepared at all | 14% |
Source: Study.com and Candova, 2026 State of AI Jobs and Skills Report (n=1,000). Full findings: study.com/resources/state-of-ai-jobs-and-skills.html.
What the data behind the profiles shows
- Adoption is effectively universal: nine in ten employees use AI at least sometimes, and 21% are required to for core work.
- Readiness lags badly: only 18% of employees feel fully prepared to use AI for nearly any task.
- Training exists but does not transfer: among the trained, only 18% say it prepared them to work independently.
- The most fixable constraint is time: 67% say two hours a week or less would meaningfully improve their skills.
The four AI readiness profiles
Every AI readiness profile below reports one to two hours a week available for training, and every one names practice on real work among its top gaps. What changes across the profiles is the starting point and the bottleneck, and that is what a training plan has to answer. Read them as a sequence: most employees sit in the first two, and the goal is to move each group one step up rather than run all four through the same course.
The AI beginner
Completes simple tasks only and needs significant editing. Covers the 35% with no training plus those prepared for basic tasks only. Top gaps: guidance on checking accuracy (57%), practice tasks reflecting real work (55%), role-relevant examples (46%). Priority is what AI can and cannot do in their specific job, before any advanced prompting.
The occasional user
Handles most tasks but edits frequently. Understands the potential, lacks the benchmarks to know when output is good enough to ship. Most-requested support is clear quality standards or checklists (25%). Top gaps: practice tasks (58%), accuracy guidance (57%), clear rules for safe use (48%).
The frequent user
Completes complex tasks with minimal editing and is ready for structured challenge. More exposure alone will not move them. Most-requested support is practice tasks with clearly defined expected outputs (19%). Top gaps: practice tasks (58%), accuracy guidance (48%), safe-use rules (46%).
The power user
Consistently produces high-quality output with little editing and represents the current performance ceiling. Wants role-specific advanced examples (22%) and better tools (22%). Top gaps: role-specific examples (51%), practice tasks (48%), time to apply learning (38%). Needs to keep current as models change; the fundamentals are already there.
The pattern across the profiles matters more than any single number. Two employees on the same AI license land in different profiles because their capability differs while the tooling stays identical, and preparedness is what predicts the return. Among employees who feel fully prepared, 70% save three or more hours a week. Among those prepared for basic tasks only, 81% save two hours or less. Seat count makes a poor proxy for return, so a plan that ignores the profiles quietly funds a rework queue for the lower two.
The four AI skills every readiness profile needs
Across all four profiles the same competency set turns up as the practical minimum. Output evaluation, spotting when AI output is wrong, misleading, or unsuitable, is the most-reported skill need and the one only 44% feel confident in. Prompt construction is writing prompts that hold up across varied tasks instead of landing one good result by luck. Task decomposition, breaking real work into AI-manageable steps, sits at 34% confidence. Safe and compliant use is last at 30% and carries the most risk, and no skill in the six-skill set the survey tested reached 50% confidence.
These four are also the competencies least likely to develop through unstructured experimentation, the model that dominates today, since nearly half of trained employees taught themselves. That explains the uneven results, and it explains why all four profiles converge on the same two asks, practice and standards, from very different starting points.
How to build AI training around the profiles
Five design principles that follow directly from what each profile reported needing and the time employees actually have.
- 1
Keep modules short and applied
30 to 60 minutes, each tied to a specific job task. Abstract AI concepts can wait. This is what fits inside the two hours a week that 67% of employees say would be enough, and lack of time is the top barrier at 41%.
- 2
Make every example role-specific
Not general AI literacy but applied examples drawn from the employee's actual responsibilities. Role-relevant examples are a top-three gap for all four profiles, and the most-requested need for beginners (24%) and power users (22%) alike.
- 3
Build accuracy practice into every module
Include evaluation and fact-checking components each time. Guidance on checking accuracy is the most consistently reported gap across profiles, at 57% for both beginners and occasional users, and only 44% of employees feel confident evaluating output.
- 4
Publish an explicit quality benchmark
Employees need a written standard for what good output looks like before they can produce it reliably. Only 32% currently have one, while 35% work from a rough idea with no benchmarks and 19% do not know at all.
- 5
Sequence from evaluation and safe use upward
Start with output evaluation and safe, compliant use, then build toward task decomposition and complex application. Safe use is both the lowest-confidence skill at 30% and the highest-risk one. It belongs at the front of the program. Leaving it for an advanced module is too late.
Where the profiles sit against other 2026 research
The profile picture lines up with what leaders report independently. Kyndryl's 2026 People Readiness Report surveyed 1,100 senior business and technology leaders across eight countries in June 2026. It found AI embedded in core processes at 57% of enterprises, up from 35% a year earlier, while only 23% of leaders believe their workforce is fully prepared, a six-point drop from 2025. Only about a third have fully implemented training programs to prepare staff to work alongside AI.
Two surveys, two populations, the same shape: deployment accelerating, preparedness flat or falling. When employees describe being handed tools without training and leaders say their workforce is not ready, the disagreement is not about the facts. For teams working out where they currently sit, our AI readiness assessment walks through the same diagnosis at company level.
The strongest case against a training-first fix
There is a serious counterargument to reading this as a training problem, and it deserves stating plainly. MIT Sloan's research on redesigning work for the age of AI locates the failure elsewhere: initiatives stall because workflows, decision rights, incentives, and operating models stay unchanged while the tools arrive. On that reading, answering flat AI returns with a training program treats a work-design problem with a curriculum.
The profile data does not contradict that. It sharpens it. What the survey measures is not a motivation deficit, since 54% already want to improve and only 9% are learning because they were told to. It measures missing structure, which is the same category of problem MIT Sloan points at. Protected time is a management decision. A clear standard for good AI output is a management decision. Only 27% of employees find their company's AI rules fully clear, and that is not something an employee trains their way out of. The honest synthesis: training built around the profiles is necessary and not sufficient. It will not survive a workflow nobody redesigned or two hours a week nobody protected. Our analysis of why AI training fails covers the same failure mode from the program side, and what type of AI training actually pays off covers the design choices that separate the two outcomes.
Where to start with the profiles
- Sort a sample of your team into the four profiles by output quality and how much editing their AI work still needs.
- Write down what good AI output looks like for each role. 68% of employees work without a clear standard.
- Protect two hours a week for applied practice, which is what 67% of employees say they need to improve.
- Do not diagnose readiness with a confidence survey. Confidence runs 16 points ahead of output quality, so it inflates every profile.
How to cite this report
Anyone citing these findings should credit both organizations that produced them. Suggested citation: Study.com and Candova, 2026 State of AI Jobs and Skills Report. The full dataset and methodology sit on Study.com's report page; the readiness-profile analysis and the training design guidance are here on Candova.
Methodology
The findings draw on two complementary studies, both fielded via Pollfish among employees across industries. Combined results are presented throughout, and each figure is sourced where appropriate.
| Study | Respondents | Fielded |
|---|---|---|
| State of AI Jobs and Skills (primary) | 1,000 | March 16, 2026 |
| AI Career Confidence Survey (secondary) | 1,000 | December 5, 2025 |
Both studies fielded via Pollfish. Conducted by Study.com and Candova; full methodology at study.com/resources/state-of-ai-jobs-and-skills.html.
Common questions
What are the four AI readiness profiles?
The AI beginner (completes simple tasks only, needs heavy editing), the occasional user (handles most tasks but edits frequently), the frequent user (complex tasks with minimal editing), and the power user (consistently high-quality output with little editing). They come from the Candova 2026 State of AI Jobs and Skills Report conducted with Study.com. Most employees sit in the first two, because only 18% feel fully prepared to use AI well.
How do I tell which AI readiness profile an employee is in?
Judge it from their output. Look at the quality of their AI work and how much editing it still needs before it can ship. Avoid sorting people by a confidence survey: the data shows self-reported confidence runs 16 points ahead of the share who actually produce high-quality output with little editing, so it inflates every profile.
Should AI training be the same for everyone?
No. Treating the workforce as one population produces a single program that is too basic for some employees and too advanced for others. Each readiness profile has a different bottleneck: beginners need to know what AI can and cannot do in their job, while power users need advanced role-specific examples and time to apply them. The training move differs by profile even though the underlying skills do not.
What do all four AI readiness profiles have in common?
Two things. Every profile names practice on real work tasks among its top gaps, and every profile reports only one to two hours a week available for training. That shared constraint is why short, applied, role-specific modules with a clear quality standard work across all four, while long generic courses work for none.
How many employees feel prepared to use AI at work?
Only 18% feel fully prepared to use AI for nearly any task, while 35% feel prepared for basic tasks only and 14% not at all, per the Candova 2026 State of AI Jobs and Skills Report with Study.com (n=1,000). Adoption is far ahead of readiness: nine in ten employees use AI at least sometimes. The full statistics are in Study.com's report.
What AI skills does every readiness profile need?
Output evaluation (spotting wrong or misleading output, 44% confident), prompt construction, task decomposition (breaking work into AI-manageable steps, 34% confident), and safe and compliant use (30% confident, and the highest organizational risk). No skill in the six-skill set reached 50% confidence, and these four are the least likely to develop through unstructured self-teaching.
Train each profile on the work they actually do
Role-specific AI training that meets employees at their readiness profile, with the practice and standards the data says are missing.
Power users save 10+ hours a week. Learn how.
The practical AI habits behind it, one a week.
Written by
Candova Editorial Team
Practical AI guides and tool coverage from the Candova AI team
The Candova Editorial Team byline covers the practical reference material on this blog: step-by-step how-tos, tool comparisons, and setup guides. These pieces answer a concrete question, like how to wire two tools together or which assistant fits a task, rather than carrying one person's point of view, so they run under the team's name instead of an individual byline.