Over-Reliance on AI at Work [2026 Data]

Over reliance on AI in the workplace is no longer a theoretical concern. When AISA asked 81 people starting the AISA AI fluency assessment "When you get a task, how often is your first instinct to reach for an AI tool?", 39.5% said almost always and another 23.5% said often. That means 63.0% of respondents default to AI as their first move on a new task. Only 11.1% said rarely, and not a single person said never.

1. Never0.0%0
2. Rarely11.1%9
3. Sometimes25.9%21
4. Often23.5%19
5. Almost always39.5%32

81 people answered, 24 July to 30 July 2026. Source: AISA research, aisa.to

These numbers raise a practical question for engineering managers, L&D leads, and anyone responsible for team output: when does reaching for AI first sharpen work, and when does it dull the thinking behind it?

How Common Is Reaching for AI First at Work?

The instinct to open an AI tool before doing anything else is now the majority behaviour, not the exception. Among the 81 people AISA asked, 63.0% chose "often" or "almost always," while 25.9% said "sometimes" and 11.1% said "rarely."

This is not a niche habit confined to power users. The spread shows that AI-first behaviour sits on a spectrum, but the centre of gravity has shifted firmly toward frequent, reflexive use.

What the Spread Tells Us

The largest single group, 39.5%, selected "almost always." That is more than the "sometimes" and "rarely" groups combined (37.0%). The data does not tell us what kinds of tasks people reach for AI on, but the sheer frequency suggests AI is becoming a default starting point across a wide range of work.

Why Frequency Alone Is Not the Problem

Reaching for AI often is not inherently a risk. The risk emerges when frequency is paired with a lack of evaluation. A 2025 global study by KPMG and the University of Melbourne found that 66% of respondents rely on AI output without evaluating its accuracy. That gap between use and verification is where ai dependency at work becomes dangerous.

When Is AI-First a Good Habit?

Reaching for AI first is a strong habit when the task is well-defined, the output is verifiable, and the human stays in the decision seat. There are real categories of work where AI-first saves time without sacrificing quality.

Structured, Repeatable Tasks

Drafting boilerplate, reformatting data, generating test cases, summarising meeting notes: these are tasks where the expected output is predictable and easy to check. Using AI first here is not cognitive surrender; it is sensible task decomposition. The person still decides what to do with the output.

Exploration and Ideation

When you need a first pass on a problem space, a list of edge cases, or a rough outline, AI is a useful thinking partner. The key distinction is that the person treats the output as raw material, not as a finished answer. Skilled users iterate, challenge, and reshape what the model returns.

When Verification Is Built In

Teams that have review steps, code review gates, editorial passes, or peer checks can afford to let individuals start with AI. The system catches errors downstream. The habit is safe because the workflow compensates for it.

When Does It Become Over-Reliance?

Over-reliance starts when someone accepts AI output without applying their own judgement, and it compounds when a team normalises that behaviour. The line is not about how often you use AI. It is about whether you are still thinking.

The Verification Gap

The same KPMG and University of Melbourne study (2025) found that 56% of respondents are making mistakes in their work due to AI. That figure connects directly to the verification gap: people use AI output, skip the check, and errors propagate. When 63.0% of the people AISA asked reach for AI often or almost always, the volume of unchecked output in any team could be substantial.

Cognitive Offloading Without a Safety Net

Cognitive offloading AI describes the process of delegating mental effort to a tool. It is not new; calculators, spell-checkers, and search engines all offload cognition. The difference with generative AI is that the output looks authoritative even when it is wrong. A calculator either gives you the right answer or an error. An LLM gives you a confident, fluent, plausible answer that may contain fabricated facts, flawed logic, or subtle bias.

When someone offloads the thinking and also offloads the checking, they have moved from using a tool to depending on one.

Team-Level Drift

Over-reliance is harder to spot at the team level. If everyone on a team starts with AI and nobody questions the outputs, the team's collective judgement atrophies. New joiners learn the norm. The standard of "good enough" shifts. This is not a single bad decision; it is a slow drift that shows up months later in quality metrics, customer complaints, or audit findings.

What Are the Risks of Relying on AI at Work?

The risks of relying on AI at work fall into three categories: output quality, skill erosion, and organisational blind spots. Each one compounds over time.

Output Quality Degrades Silently

AI-generated errors are harder to catch than human errors because they are fluent and well-structured. A hallucinated citation looks like a real citation. A subtly wrong code path passes a cursory review. When teams rely on AI without verification checklists, error rates climb in ways that are invisible until something breaks.

Individual Skills Atrophy

Skills that are not exercised weaken. If a developer always asks AI to write the first draft of a function, their ability to reason through the logic from scratch diminishes. If a product manager always asks AI to draft the competitive analysis, their market intuition narrows. This is not speculation about the future; it is a well-documented pattern in automation research across aviation, medicine, and manufacturing.

The question for managers is not whether their team uses AI. It is whether their team can still do the work without it.

Organisational Blind Spots

When AI mediates most of the thinking, the organisation loses visibility into how decisions are actually made. If a recommendation came from an AI and nobody flagged it, the decision audit trail is incomplete. If a risk assessment was AI-generated and accepted without challenge, the organisation owns a risk it did not consciously evaluate.

For regulated industries, this is not just a quality issue. It is a compliance issue. Understanding AI governance frameworks becomes essential when AI output feeds into consequential decisions.

How Do You Keep Judgement in the Loop?

Keeping judgement in the loop means designing work so that humans evaluate AI output before it matters. This is a workflow problem, not a willpower problem.

Separate Generation from Evaluation

The simplest intervention is to make generation and evaluation two distinct steps. Let people use AI to generate drafts, options, or analyses. Then require a separate, deliberate evaluation step before the output moves forward. This can be as lightweight as a self-review checklist or as structured as a peer review process.

The AISA rubric calls this the difference between using AI as a tool and treating AI as an authority. The critical thinking with AI guide breaks down what effective evaluation looks like in practice.

Build Verification Into the Workflow

Do not rely on individuals to remember to check. Build the check into the process. Code review gates, editorial sign-offs, data validation steps, and stakes-based review protocols all serve this purpose. The higher the stakes of the output, the more rigorous the verification should be.

Practise Without AI Periodically

Some teams run "unplugged" exercises: a sprint planning session without AI summaries, a design review where the designer presents their own reasoning, a debugging session where the developer traces the logic manually. These are not anti-AI gestures. They are skill-maintenance routines, the cognitive equivalent of a fire drill.

Measure AI Fluency, Not Just AI Usage

Usage metrics tell you how often people use AI. They do not tell you whether people use it well. A team AI assessment measures whether people can prompt effectively, evaluate output critically, and apply AI appropriately to different task types. That is the difference between tracking activity and tracking capability.

The post Am I Too Reliant on AI? [2026 Survey] explores this question from the individual's perspective and is worth sharing with team members who want to self-assess.

How Should Teams Set Norms for AI Use?

Teams should set explicit, written norms that distinguish between tasks where AI-first is encouraged and tasks where human-first thinking is required. Implicit norms drift; explicit norms hold.

Define AI-Appropriate and AI-Inappropriate Tasks

Not every task benefits from AI. Start by categorising your team's recurring work into three buckets:

1. AI-first tasks: Low-stakes, verifiable, repeatable. AI drafts, human checks.
2. AI-assisted tasks: Medium-stakes, requires domain knowledge. Human leads, AI supports.
3. Human-first tasks: High-stakes, novel, or requiring original judgement. Human thinks first, may use AI to pressure-test.

This is not a permanent classification. It should be reviewed quarterly as tools improve and team skills develop. The AI readiness assessment checklist provides a structured framework for this kind of review.

Make the Norm Visible

Write it down. Put it in the team wiki, the onboarding doc, the project kickoff template. A norm that lives only in a manager's head is not a norm; it is a preference. Teams that document their AI use guidelines report fewer quality incidents and more consistent output.

Review and Adapt

AI tools change. Team skills change. The norms should change too. Schedule a quarterly review of your AI use guidelines. Look at where AI helped, where it hurt, and where the team's judgement was the deciding factor. Use that evidence to update the norms.

Understanding where your team sits on the AI fluency spectrum helps you set norms that match actual capability rather than assumed capability. The AI personas data shows how different usage patterns map to different skill profiles, which is useful context for any team-level conversation about AI norms.

Invest in the Right Skills

Norms without skills are just rules people cannot follow. If you want your team to evaluate AI output critically, invest in building that capability. If you want people to know when AI is the wrong tool, they need enough technical understanding to make that call. The AI fluency levels breakdown explains what competence looks like at each stage, from basic awareness to expert-level orchestration.

For teams building AI upskilling into their development plans, the AI upskilling and measurement strategy for 2027 lays out a practical roadmap.

About this research

This finding comes from AISA's own research. AISA asks a short research question of people as they start the AISA AI fluency assessment, one question at a time. The question in the chart above was asked from 24 July to 30 July 2026, worded exactly as shown and with the answers in the order shown, and 81 people answered it. Each person counts once: when someone answered more than once, their latest answer is the one counted.

More findings from this research are collected in AISA research and data.

How to cite this research: AISA (2026). "Over-Reliance on AI at Work [2026 Data]". AISA research. https://aisa.to/blog/over-reliance-on-ai-at-work

---

Related reading: Am I Too Reliant on AI? [2026 Survey], the individual side of the same research question.

Related reading: AI Readiness Assessment Checklist [2026], a structured checklist for teams evaluating their AI maturity.

Related reading: AI Fluency Levels: 5 Tiers Explained [2026], what each fluency tier means and how to move up.

Frequently Asked Questions

What is over-reliance on AI?

Over-reliance on AI is the pattern of accepting AI-generated output without applying independent judgement or verification. It is not defined by how often someone uses AI, but by whether they evaluate what the AI produces before acting on it. When verification drops out of the workflow, errors compound and skills erode.

How do you prevent over-reliance on AI at work?

Prevent over-reliance by building verification into your workflows rather than relying on individual discipline. Separate the generation step from the evaluation step, use review gates for high-stakes output, and measure AI fluency rather than just AI usage. Periodic "unplugged" exercises help teams maintain the skills they need to check AI work effectively.

Is it bad to depend on AI?

Depending on AI is not inherently bad. It becomes a problem when the dependency removes human judgement from the process. Using AI to draft, explore, and accelerate is productive. Accepting AI output without checking it, or losing the ability to do the work without AI, is where dependency turns into risk. The goal is skilled use with maintained judgement, not avoidance.