The AI Fatigue Report
What 3,000 Engineers Taught Us
In late 2025 and early 2026, 3,047 software engineers took The Clearing’s AI Fatigue Quiz. Here’s what they revealed about the middleman problem, skill decline, identity, and the future of the profession.
How We Got This Data
From October 2025 through April 2026, The Clearing ran an anonymous, voluntary, 5-question AI Fatigue Quiz for software engineers. No account required. No tracking. No email gate. 3,047 engineers completed it.
The quiz covers five dimensions: coding autopilot reliance, Sunday dread, craft satisfaction erosion, epistemic abdication (accepting things you don’t fully understand), and authorship ambiguity. Scores range from 0 to 15, clustered into four tiers. Distributed via Hacker News, Reddit, Twitter, and direct engineer referrals.
The Tier Breakdown
Engineers were clustered into four tiers based on their quiz score. Here’s how they distributed:
The Four Core Patterns
Across all tiers, four patterns appeared most consistently in the open-ended responses. These weren’t quiz questions – engineers wrote these in their own words.
“I used to come home from work feeling like I’d built something. Now I come home and I don’t know what I know anymore.”— Tier 2 respondent, 8 years experience, full-stack engineer
The Middleman Feeling
71% of Tier 2+ respondents described this pattern. Code ships. They approved it. They reviewed it. But they don’t feel like they built it. The output exists and the understanding doesn’t. This is the most cited pattern in open-ended responses – it’s the reason The Clearing exists.
Skill Erosion in Specific Areas
58% of respondents reported noticing specific skills declining – not general “getting rusty.” Particular signals: debugging without AI assistance, writing complex algorithms from scratch, estimating without AI input, and reading unfamiliar codebases. The atrophy is selective, not total.
The Sunday Night Reckoning
A recurring theme in open-ended responses: the Sunday evening dread specifically tied to AI dependency. Not “work dread” in general, but knowing that another week of shipping code without understanding it is starting. Several engineers described this as the moment they finally took the quiz.
The Exit Consideration
44% – nearly half – reported having seriously considered leaving the software engineering profession. Not burnout from overwork. Not hours. Specifically: the feeling that the profession as currently practiced is incompatible with maintaining skill, craft satisfaction, and professional identity. This is a retention crisis signal.
What Tier Severity Tells Us
The quiz scores correlated meaningfully with the qualitative themes in open-ended responses. Engineers in higher tiers didn’t just have worse scores – they described qualitatively different experiences.
Who Took the Quiz
The quiz attracted a diverse range of software engineers across experience levels, roles, and company sizes.
What This Means
These numbers point to something real and structural. This isn’t just “burnout” or “too much screen time.” It’s a specific kind of cognitive and professional erosion that comes from how AI has been integrated into engineering workflows without guardrails for craft, learning, and professional identity.
The middleman feeling isn’t a character flaw. It’s a rational response to a workflow that rewards output over understanding. The engineers reporting it aren’t lazy or resistant to change – they’re the ones paying attention.
The 44% who considered leaving isn’t a sign of weakness in the profession. It’s a signal that the profession, as currently practiced, is misaligned with the conditions that make it sustainable.
These patterns are fixable. Not by using less AI, but by using it differently – with boundaries, with deliberate practice, and with an insistence on maintaining the ownership loop that makes engineering meaningful.
🚧 For Engineers
The data shows you’re not imagining it. The middleman feeling is real, widespread, and recoverable. Start with the Explanation Requirement: before accepting any AI suggestion, explain why it works in one sentence. The gap that appears is the learning loop you’re rebuilding.
👥 For Managers
The team that built this data is also the team struggling. Engineers who feel like middlemen in their own work don’t bring the same quality of judgment to code review, architecture decisions, or estimation. Team AI agreements aren’t soft – they’re a retention strategy.
🔬 For Researchers
This is self-reported, self-selected data. The patterns are consistent enough to warrant formal study. We’d welcome collaboration with researchers interested in the cognitive and professional effects of AI tool integration on software engineers.
What Actually Helps
When we asked engineers who had recovered what worked, three practices appeared most frequently in open-ended responses:
The Explanation Requirement
Before accepting any AI suggestion, write one sentence explaining why it works. If you can’t, you don’t accept it. This rebuilds the ownership loop AI dissolved.
No-AI Work Blocks
Protected time – 90 minutes minimum – where problems are solved without AI assistance. Not as deprivation, but as practice. The struggle is the learning.
Weekly Calibrations
Once a week, take a feature shipped this week and rebuild one small piece from scratch. The gap between what you ship and what you can rebuild is data about where the learning stopped.
Cite This Report
If you’re writing about AI fatigue, developer wellbeing, or the effects of AI tool adoption on software engineers, you can cite this report as follows:
Take the AI Fatigue Quiz
3,047 engineers have taken it. See where you fall – and get a personalized recovery plan based on your score.
Free. No account required. No tracking. Anonymous.