6 Things That Failed in 2025
77% less productive. 88% burned out. 95% got zero ROI. Here's what didn't work.
👋🏽 Hey, it’s Anita. Welcome to AI@Scale — I write for leaders trying to make AI work at scale, where the hardest part isn’t the tech, it’s the humans.
💗 It’s the last post of 2025! I’ve linked to former posts throughout this series. I hope they help you close out the year with clarity and start 2026 with conviction
We spent 2025 learning what doesn’t work.
88% of organizations now use AI regularly. But…most are stuck in pilots that won’t scale, tools nobody uses, productivity promises that became burnout reality.
The bill came due: $9 million annually per organization in low-quality AI output (According to Harvard Business Review). 88% higher burnout among AI users. 77% saying AI actually decreased their productivity.
Technology and AI Budget were definitely NOT lacking in 2025, but we did spend most of 2025 celebrating the wrong things, measuring the wrong metrics, and moving too fast to notice we’re breaking people.
Here are the six practices that failed us most in 2025 and why they cost more than we realized.
1. Training Completion Rates
❌ What failed: Celebrating 90% training completion rates
Why it failed: 81% of all workers are non-AI users (Pew Research)
Completion measures compliance, not change. We spent all year measuring attendance while adoption flatlined.
💰 The real cost: Wasted training investment + false sense of progress + leadership credibility when “trained” teams still can’t adopt AI.
What we learned: Completion is the starting line, not the finish line. If people complete training but don’t change their work, the training failed.
2. Efficiency-Only Thinking
❌ What failed: Asking only “Does it work?” and “Does it save money?”
Why it failed: 56% say “AI with no empathy” is the biggest obstacle to adoption (Consultany.uk)
We optimized for Effective and Efficient. We forgot Empathetic.
The result: Tools that technically work but feel terrible to use. Workflows that save time but destroy trust. Deployments that check the ROI box but crater morale.
💰The real cost: Low adoption despite “successful” pilots + high turnover in AI-enabled roles + resistance labeled as “change fatigue” when it’s actually shit*%y deployment .
What we learned: 87% say empathy drives better leadership. 88% say it increases efficiency. Empathy isn’t soft, it’s strategic. And its absence is expensive.
3. Unsupported Middle Managers
❌ What failed: Blaming middle managers for “blocking” AI adoption
Why it failed: Only about 25% of frontline employees say they receive strong leadership support for AI adoption (BCG, 2025)
They’re not blockers. They’re caught between executive vision and terrified teams with no playbook.
We spent 2025 giving managers AI mandates without any training, deadlines without support and accountability for adoption without tools to help their teams navigate fear, workflow disruption, or identity threats.
Then we called them “resistant to change” when they couldn’t magically make it work.
💰The real cost: Manager burnout + team confusion + adoption delays blamed on “middle management resistance” when the real problem is we set them up to fail.
What we learned: Managers can’t enable what they haven’t been enabled for. Train them first on leading humans through change (no, not the tools!!)
4. Pretend Productivity
❌ What failed: Filling AI-saved time with more work
Why it failed: 88% of AI-enabled workers show higher burnout. They’re 2x more likely to quit (Upwork Annual Report, 2024)
AI was supposed to reduce workload. Instead, we raised the productivity bar.
People got faster, so we gave them more tasks. Reports that took 8 hours now take 5 and so we expect 3 more reports. The work multiplied. The hours didn’t shrink.
We called it “maximizing AI ROI.” but it’s actually, “the new normal that’s burning me out.”
💰 The real cost: Doubled turnover in AI-enabled roles + productivity gains erased by rehiring and retraining + reputation damage when “AI adoption” becomes code for “work more.”
What we learned: AI should create capacity, not just multiply output. Protect the freed time or lose the people who freed it.
5. Ignoring Identity Threats
❌ What failed: Calling people “resistant to change”
Why it failed: When you ask someone to adopt AI, you’re not asking them to learn a tool. You’re asking them to redefine who they are at work.
“I’ve spent 15 years becoming an expert. AI does it in 15 minutes. What does that make me?”
That’s not resistance. That’s someone in fear of their future. And we responded with “just try it” and “embrace the change” without addressing the underlying fear.
💰 The real cost: Quiet quitting from your best people + brain drain as experts leave rather than “compete with AI” + institutional knowledge lost because we didn’t help people see their new value.
What we learned: Identity protection isn’t resistance, it’s survival. Help people redefine their value before asking them to change their work.
6. Ignoring “Workslop” (My Favorite)
❌ What failed: Celebrating AI-generated output without measuring quality
Why it failed: “Workslop” = AI-generated content that looks professional but lacks substance.
Cost: $9 million annually for a 10,000-person organization in time spent fixing low-quality AI output.
77% say AI decreased productivity because of time spent correcting mistakes.
We measured volume. We celebrated speed. We didn’t notice that “10 reports in the time it used to take for 3” meant “10 reports that all need significant rework.”
💰 The real cost: $9M annually + credibility damage when clients receive “AI slop” + team frustration spending more time editing than creating + new quality control layer that erases efficiency gains.
What we learned: Speed without quality isn’t efficiency = waste. Quality-adjusted productivity matters more than raw output.
What’s Next
These six practices cost us 2025. Tomorrow, we cover what high-performing organizations are doing instead.
The 4 E’s framework that includes empathy. Manager enablement strategies that work. How to measure behavior change instead of completion. Ways to protect AI-freed time from productivity creep. And how to help people redefine their value instead of defending their old roles (Just to name a few things!)
See you tomorrow with what we’re doing instead.