The AI Fatigue Severity Index

Why a 5-Axis Framework Changes the Recovery Equation

Most self-assessment tools give you a single number. You answer 10 questions, you get a score, you get a label. Tier 1, Tier 2, Tier 3, Tier 4. That's useful for triage, but it's not useful for recovery. Recovery requires specificity.

AI fatigue is not one thing. It shows up as cognitive overload, skill erosion, decision fatigue, attention fragmentation, and identity erosion — often all at once, but rarely in equal measure. You might be drowning in attention residue while your skills are actually fine. Or your identity might be in crisis while your cognitive load is manageable. A single composite score hides the pattern that determines which recovery practices actually work for you.

This index measures all five dimensions separately. The result is not just a number — it's a map. And the map is what tells you where to start.

The Five Axes Explained

Cognitive Load — The Mental Bandwidth Crisis

Every AI tool you use adds to your cognitive load — not just the tool itself, but the work of evaluating its output, catching its errors, maintaining context across conversations, and deciding when to trust it versus when to push back. John Sweller's cognitive load theory, developed for educational psychology in the 1980s, identified three types: intrinsic load (the inherent difficulty of the material), extraneous load (the unnecessary difficulty imposed by poor design), and germane load (the cognitive effort that actually produces learning and skill development).

AI tools have a unique effect on all three simultaneously. They reduce intrinsic load by making complex tasks feel simple — which sounds good until you realize that the struggle required for genuine learning is also reduced. They introduce extraneous load through the constant evaluation and verification work. And they suppress germane load by bypassing the productive struggle that would otherwise create lasting neural connections.

Your cognitive load score reflects how much mental bandwidth you're spending on the overhead of using AI tools, rather than on the actual work. If that ratio is high, you're processing more than you're producing. That's the signature of a system under pressure.

Skill Erosion — The Quiet Hollowing Out

Robert Bjork's work on "desirable difficulties" established that genuine learning requires the friction of retrieval practice — the effort of pulling information from memory rather than having it provided. When you use AI to skip that retrieval, you don't just save time in the short term; you eliminate the cognitive process that would have strengthened the underlying memory trace. The skill doesn't just plateau. It erodes.

Skill erosion from AI use is distinct from ordinary learning decay. With normal disuse, you forget because you haven't accessed the information in a while. With AI-driven skill erosion, the problem is more fundamental: you never built the full representation in the first place, because AI handled the hardest parts on your behalf. The gap isn't just that you forgot — it's that you never fully learned. See the full skill atrophy breakdown

The skill erosion axis measures how much of your core coding ability has been compromised by outsourcing to AI. Not just "are you using AI" but "what has AI done to what you can actually do without it?"

Decision Fatigue — The Depletion of the Decision-Making Reservoir

Roy Baumeister's research on ego depletion established that decision-making is a limited resource that depletes with use. Every significant decision you make reduces the quality of subsequent decisions. This is why software engineers who spend their entire day making micro-decisions — which approach to take, which tool to use, whether to accept an AI's suggestion, how to name a variable — often end up making worse decisions by the afternoon than they would have made with a fresh mind.

AI tools have made the decision problem significantly worse. Each AI output requires a decision: Is this right? Should I use it? Should I modify it? Do I understand it well enough to accept it? The volume of these micro-decisions per unit of output is much higher with AI-assisted work than with solo work. You're making ten decisions to ship one piece of code that you could have written yourself in fifteen decisions. The math is relentless.

Your decision fatigue score reflects the extent to which AI tool usage has depleted your decision-making capacity. High scores here often show up as "I spent all day in meetings and got nothing done" — except the meetings were mostly conversations with AI systems, and the output was low quality because the decisions that shaped it were made by a depleted mind.

Attention Residue — The Inability to Fully Arrive

Sophie Leroy's 2009 research introduced the concept of attention residue: when you switch tasks, a portion of your cognitive attention remains on the previous task rather than fully transitioning to the new one. This residue accumulates. In a work environment with frequent context switches, Leroy found that people operated with a significant portion of their cognitive capacity already occupied — not by any task, but by the ghost of tasks they'd left behind.

AI tools are specifically designed to encourage switching. Each conversation is a new context. Each prompt is a new task initiation. The model promises quick answers, which means you start a new thread every time you have a question rather than sitting with the hard problem until it's solved. The result is an extreme form of attention residue — your cognitive capacity is perpetually distributed across multiple incompletely processed AI interactions rather than concentrated on any one thing.

Gloria Mark's field research at a tech company found that after an interruption, it takes an average of 23 minutes and 15 seconds to fully return to a task. AI tools generate interruptions by design — not just externally, but through the constant generation of new prompts and new threads. Read the attention residue research in full. Your attention residue score measures how much of your cognitive capacity is locked up in incomplete AI interactions rather than available for deep work.

Identity Health — The Erosion of the Engineer You Believe You Are

Software engineering has always been more than a job. The engineers who go into this field tend to have a strong occupational identity — they are not just people who write code, but people who define themselves through their technical skill, their problem-solving ability, their capacity to understand complex systems. That identity is not vanity. It's functional. It gives you the confidence to take on hard problems, the resilience to persist through difficulty, and the sense of purpose that makes sustained effort sustainable.

AI tools are creating a specific kind of identity threat that previous technological shifts did not. When you use a compiler, you don't feel like the compiler wrote your code — you feel like you wrote code and the compiler processed it. When you use AI to generate a solution, the authorship boundary is much blurrier. The code works. You didn't build it alone. You may not fully understand it. And if you don't fully understand it, what does that say about your competence as an engineer?

This is not imposter syndrome. Imposter syndrome is the feeling that you are fraudulent despite competence. Identity erosion from AI is the feeling that your competence is genuinely compromised — that you have contributed to a real decline in what you can do independently. Both are painful, but they require different responses. Read the full developer identity analysis. Your identity health score tells you where you actually are on this dimension, separate from the psychological noise of classic imposter syndrome.

Why Severity Index Matters More Than the Quiz

The AI Fatigue Quiz on this site gives you a tier — a category that tells you whether you're holding up, mildly fatigued, significantly fatigued, or in crisis. That's the right tool for initial self-assessment. It tells you whether to take this seriously.

The Severity Index is for after you've taken the quiz. It tells you specifically which dimensions are most compromised. That specificity is what makes recovery efficient. If your cognitive load is manageable but your skill erosion is severe, the recovery practices that work for cognitive overload will waste your time. You need skill-retention work, not cognitive load management.

Taking both tools — the quiz for initial triage, the Severity Index for detailed mapping — gives you the most complete picture of where you stand and where to start.