What AI Skill to Learn First? [2026 Data]

Data from 2,527 assessments reveals what AI skill to learn first based on dimension gaps, role, and ROI. Find your starting point.

By Ozan Dagdeviren··14 min read
search intentskillsdimensionslearning pathai-skillslearning-pathpromptingcritical-thinkingassessment-datacareer-development

Figuring out what AI skill to learn first is the single most common question we hear from professionals taking their first serious step into AI fluency. It's also the question most people answer with gut instinct rather than evidence. We have evidence. Across 2,527 completed AISA assessments, we see clear patterns in where people are strongest, where they're weakest, and — critically — which skill you should prioritise depending on your role and goals.

The short answer: for most people, prompting and communication is the highest-ROI starting point. But "most people" might not be you. Let's walk through the data.

The 5 AI Fluency Dimensions, Ranked by Average Score

AISA measures AI fluency across five dimensions, each scored 0–100. Here's how 2,527 people performed, ranked from strongest to weakest:

RankDimensionAvg. ScoreWeight in Composite
1Workflow & Application47.225%
2Prompting & Communication44.523%
3Critical Thinking42.422%
4Safety & Responsibility41.510%
5Technical Understanding39.220%

The overall average composite is 46.2 out of 100 — firmly in the Developing tier (28–59). That means the median person using AI tools today is operating well below the Proficient threshold of 60.

What the Gap Between Top and Bottom Tells Us

The spread from the highest dimension (Workflow, 47.2) to the lowest (Technical Understanding, 39.2) is 8 points. That's meaningful but not enormous. It tells us that most people aren't dramatically lopsided — they're moderately weak across the board, with slight relative strengths in the practical, hands-on dimensions.

This makes sense. People learn by doing. You pick up workflow patterns by using ChatGPT or Claude at work. You pick up prompting habits through trial and error. But the conceptual and evaluative skills — understanding how models actually work, spotting when output is wrong, knowing what data you're exposing — those require deliberate study.

Why Technical Understanding Sits at the Bottom

Technical understanding (39.2) covers knowledge of how AI models work: token prediction, context windows, model selection, and the difference between training and inference. Most professionals haven't needed this knowledge to get value from AI tools — until they hit a wall. When a prompt fails and you don't understand why, when you pick the wrong model for a task, when you can't explain to your team why the output is unreliable — that's the technical understanding gap biting.

Stanford's 2024 AI Index Report found that while AI adoption in organisations doubled year-over-year, formal AI literacy training lagged significantly behind deployment. The pattern we see in AISA data is consistent: people are using AI faster than they're understanding it.

Why Starting With Your Weakest Dimension Isn't Always Right

The intuitive move is to look at the table above, find your lowest score, and start there. That logic works for some people. It doesn't work for everyone, and here's why.

The Dependency Problem

AI skills aren't independent modules. They form a dependency chain. Prompting is upstream of almost everything else. If you can't communicate effectively with an AI model, you can't generate output worth evaluating (critical thinking), you can't build it into a workflow (workflow & application), and you can't assess its risks (safety & responsibility).

Starting with technical understanding when you can't write a clear prompt is like studying engine mechanics before you've learned to drive. Useful eventually, but the wrong sequence.

The Motivation Problem

The weakest dimension is often the one you find least interesting. If you force yourself to start with safety frameworks when what excites you is building AI-assisted workflows, you'll quit in a week. Sustained skill development requires a feedback loop: learn something, apply it, see results, learn more. That loop is hardest to establish in your least-engaged area.

The Diminishing-Returns Problem

Moving from 30 to 45 in any dimension is easier than moving from 45 to 60. But the impact of that improvement depends on which dimension it is. A 15-point improvement in prompting unlocks better output across every task you do with AI. A 15-point improvement in technical understanding is valuable but narrower in its immediate payoff.

The question isn't just "where am I weakest?" — it's "where will improvement compound fastest?"

The Case for Starting With Prompting (44.5)

Prompting and communication is the gateway skill for AI fluency. It's the dimension that most directly determines the quality of every interaction you have with an AI model, and improving it produces visible results within hours, not weeks.

Prompting Is the Highest-Leverage Input

Every AI workflow starts with a prompt. The difference between a vague request and a well-structured one — with clear context, constraints, and output format — can be the difference between useless output and genuinely useful work product. AISA's prompting dimension covers specificity, context engineering, iterative refinement, and the ability to adapt your communication style to different models and tasks.

At 44.5, the average prompting score sits in the Developing range. Most people are writing prompts that work sometimes but lack the consistency and precision that separate a Competent user from a Developing one. The gap between "I got a decent answer" and "I reliably get excellent answers" is almost entirely a prompting gap.

The Feedback Loop Is Immediate

Prompting is the fastest skill to practice because the feedback is instant. Write a better prompt, get a better response. Try a different structure, see how the output changes. This tight loop makes prompting the easiest dimension to improve through deliberate practice — and the one where motivation is easiest to sustain.

Our free Prompting Skills guide covers the core techniques: role prompting, few-shot examples, chain-of-thought, output formatting, and iterative refinement. Most people can move from Developing to Competent in prompting within a few focused weeks.

Role Data Supports Prompting as the Starting Point

Look at how prompting scores vary by role:

RolePrompting ScoreComposite Scoren
Product53.455.0148
Founders52.155.4228
Engineering51.154.3516
Design47.947.255
Data47.849.953
Students36.236.2228

The roles with the highest prompting scores (Product, Founders) also have the highest composites. Correlation isn't causation, but the pattern is consistent: people who communicate well with AI tend to perform well across all dimensions. Prompting is the tide that lifts every boat.

Students, at 36.2, have the largest gap to close — and the most to gain. If you're a student or early-career professional, prompting is unambiguously where to start. We've written more about this cohort in our analysis of 228 student assessments.

AISA

Curious about your AI Fluency?

AISA helps you measure, prove and improve your AI skills — free report in a 20–40 minute chat.

The Case for Starting With Critical Thinking (42.4)

Critical thinking is the skill that prevents AI fluency from becoming AI dependence. At an average of 42.4, it's the second-weakest dimension — and arguably the one with the most dangerous consequences when it's missing.

What Critical Thinking Covers

AISA's critical thinking dimension measures your ability to evaluate AI output: detecting hallucinations, checking claims against sources, identifying logical gaps, recognising when a model is confidently wrong, and knowing when to override or reject AI suggestions.

This matters because AI models are getting better at sounding right. Claude Fable 5.2, released this month, scores 65 on the Artificial Analysis Intelligence Index — the highest of any public model. But higher capability doesn't mean fewer errors. It means errors are harder to spot because the surrounding output is more polished and persuasive.

Critical Thinking Compounds Everything

Here's the compounding argument: every other AI skill becomes more valuable when paired with strong critical thinking. Better prompting gets you better output — but only if you can tell when the output is actually good. Workflow automation saves time — but only if you're not automating the propagation of errors. Technical understanding helps you choose the right model — but only if you can evaluate whether the model's output meets your quality bar.

McKinsey's 2024 report on AI adoption found that 65% of organisations were regularly using generative AI, nearly double the figure from ten months prior. As adoption accelerates, the risk of over-reliance on AI grows proportionally. Critical thinking is the counterweight.

When to Start Here Instead of Prompting

If you're already a decent prompter — say, scoring above 50 in prompting — but you find yourself accepting AI output without much scrutiny, critical thinking is your higher-ROI starting point. This profile is common among Enthusiasts (22% of AISA personas, average score 51.9): people who are excited about AI and use it frequently but haven't developed the evaluative habits to use it well.

Our Critical Thinking with AI guide covers hallucination detection, source triangulation, and the habit of treating AI output as a first draft rather than a final answer.

What AI Skill to Learn First: A Role-Based Decision Framework

The data suggests different starting points depending on your role, current level, and goals. Here's a practical framework.

Match Your Role to Your First Skill

Your RoleLikely Weakest DimensionRecommended First SkillWhy
EngineeringSafety (47.5)Critical ThinkingYou already prompt well (51.1); learn to evaluate output before shipping it
ProductTechnical Understanding (46.9)PromptingYour prompting is strong (53.4); deepen it to Proficient level for PRDs and specs
DesignSafety (35.3)PromptingAt 47.9, your prompting has room to grow; it'll improve every design workflow
DataTechnical Understanding (42.2)Critical ThinkingYou need to verify AI-generated analyses before they reach stakeholders
FoundersSafety (50.4)Critical ThinkingYou're strong across the board; critical thinking prevents costly AI-driven mistakes
StudentsAll dimensions (~30-36)PromptingStart with the gateway skill; everything else builds on it

The Two-Question Filter

If the table above doesn't match your situation, use this:

  1. Can you reliably get useful output from AI on your first or second attempt? If no → start with prompting. If yes → move to question 2.
  2. Do you routinely verify AI output before using it? If no → start with critical thinking. If yes → pick the dimension most relevant to your current work (workflow for efficiency gains, technical understanding for tool selection, safety for compliance).

Don't Ignore the Overconfidence Factor

Across 1,617 people who predicted their score before taking AISA, the average predicted score was 62.3 against an actual of 44.5 — an overconfidence gap of 17.8 points. Students showed the largest gap: predicted 66.9, actual 35.0 — a 31.9-point overestimation.

This means your self-assessment of which skill to learn first is probably wrong. You likely think you're better at prompting than you are. You almost certainly think you're better at critical thinking than you are. The only reliable way to know where you actually stand is to measure it.

How to Test Which AI Skill to Prioritise

The fastest way to identify your starting point is to take a structured assessment that scores each dimension independently. Self-reflection is unreliable — the data on overconfidence makes that clear.

What a Good Assessment Tells You

A useful AI fluency assessment does three things:

  1. Scores each dimension separately so you can see your actual profile, not just an aggregate number
  2. Provides evidence-linked feedback — not just "you scored 40 in critical thinking" but "here's what you said that demonstrated (or didn't demonstrate) this skill"
  3. Gives you a baseline you can measure against later so you know whether your learning efforts are working

AISA produces a free report covering all five dimensions with specific, evidence-linked scores. 98% of people who retake land in the same or adjacent tier, which means the baseline is stable enough to track real improvement over time.

Build a 30-Day Learning Sprint

Once you know your starting dimension, structure your first month:

  • Week 1: Read the relevant AI Fluency Academy guide for your chosen dimension. Complete the exercises.
  • Week 2: Apply one new technique per day in your actual work. Document what works.
  • Week 3: Deliberately practice the hardest sub-skill within your dimension (e.g., iterative refinement within prompting, or hallucination detection within critical thinking).
  • Week 4: Retake the assessment. Compare your dimension scores to your baseline.

This isn't about becoming an expert in 30 days. It's about moving from Developing to Competent in one dimension — a shift that, based on the score distributions we see, puts you ahead of roughly half the professionals currently using AI.

Stack Skills in the Right Order

After your first dimension reaches Competent (50+), the recommended stacking order for most roles is:

  1. Prompting & Communication → the gateway
  2. Critical Thinking → the quality filter
  3. Workflow & Application → the efficiency multiplier
  4. Technical Understanding → the depth layer
  5. Safety & Responsibility → the governance layer

This isn't rigid. If you work in a regulated industry, safety might need to come earlier. If you're building AI products, technical understanding moves up. But for the median professional, this sequence maximises cumulative ROI.

For a broader view of which AI skills matter most across roles, see our ranked breakdown of 7 AI skills by impact. And if you're looking for ways to accelerate your progress, our guide on how to improve AI skills fast covers the most effective practice patterns we've observed.


Related reading: Should I Learn AI Now or Wait? [2026] — Data on why waiting costs more than starting imperfectly.

Related reading: AI Fluency Levels: 5 Tiers Explained [2026] — What each AISA tier means in practice and how to move up.

Related reading: Am I Falling Behind on AI? [2026 Data] — Where you stand relative to 2,527 assessed professionals.


Frequently Asked Questions

Is prompting the most important AI skill?

Prompting is the most foundational AI skill, not necessarily the most important for every person. Across 2,527 AISA assessments, prompting scores correlate strongly with overall composite scores — roles with the highest prompting averages (Product at 53.4, Founders at 52.1) also have the highest composites. For most people starting out, prompting is the highest-ROI first skill because it directly improves the quality of every AI interaction. But if you already prompt well, critical thinking or workflow skills may deliver more marginal value.

Should I learn technical AI concepts or practical skills first?

Start with practical skills. Technical understanding (how models work, tokenization, context windows) scores lowest across AISA assessments at 39.2, but it's also the dimension with the least immediate payoff for most roles. Prompting (44.5) and workflow application (47.2) produce visible results faster and build the experiential foundation that makes technical concepts easier to absorb later. The exception: if you're building AI products or managing AI infrastructure, technical understanding should move up your priority list.

How long does it take to improve one AI skill?

Based on patterns in AISA retake data, most people can move one tier within a single dimension (e.g., Developing to Competent) in 2–4 weeks of focused practice. The key is deliberate, daily application — not passive reading. AISA's retake stability (98% land in the same or adjacent tier) means the assessment reliably detects real improvement when it happens, so you can use it as a before-and-after benchmark for a focused learning sprint.

Ozan Dagdeviren

Ozan Dagdeviren

Founder of AISA — the AI skills assessment platform used by professionals worldwide to measure, certify, and develop their AI fluency. More about AISA

AISA

Curious about your AI Fluency?

AISA helps you measure, prove and improve your AI skills — free report in a 20–40 minute chat.

The Science Behind AISA

Metropolitan PoliceHarvard UniversityCrowdboticsE.S.E.

In 2026, Anthropic published the AI Fluency Index — the largest empirical study of AI fluency to date, analysing nearly 10,000 conversations. AISA covers 93% of the behaviours Anthropic identified as markers of AI fluency and goes even deeper with 4 additional criteria. The U.S. Department of Labor's AI Literacy Framework (TEN 07-25) defines what every worker needs to know about AI — AISA covers 100% of its 25 sub-competencies.Read our analysis: Anthropic's AI Fluency Study & AISA · DOL AI Literacy Framework & AISA

AISA's framework is developed by a team with deep roots in tech, behavioural science, and AI product leadership — the rubric is informed by backgrounds spanning the Metropolitan Police, Harvard, Crowdbotics (Silicon Valley), and the European School of Economics.