March 13, 2026

AI Over 40 Series - Week 20: What MIT and Wharton Found

And what they’re still missing

I recently sat down to skim two research reports that had been collecting dust in my reading pile: MIT’s The GenAI Divide: State of AI in Business 2025 and Wharton’s Accountable Acceleration: Gen AI Fast-Tracks Into the Enterprise. I expected to spend thirty minutes. I spent hours.

Not because they revealed something new about AI. But because they invested enormous resources in measuring what many of us have been living — and still missed the one thing that matters most.

The shadow AI economy is real

MIT identified what they call the “shadow AI economy”: 90% of employees use AI tools personally for work tasks, yet only 40% of companies provide official LLM subscriptions.

That 90% figure confirms what I’ve seen firsthand. Professionals are building AI literacy faster than their organizations can react.

Wharton shows a similar shift. Early AI adoption was employee-led. Now it’s consolidating at the executive level. They frame that as strategic maturity. I see organizations catching up to what individuals already discovered.

Personal transformation is happening. Institutional clarity is lagging.

The 95% failure rate

MIT’s headline finding is stark: despite tens of billions invested, 95% of organizations report no meaningful AI return. Only 5% of custom tools reach production.

Their explanation is a “learning gap.” Tools lack context. Feedback loops are weak.

That aligns directly with what we explored in earlier weeks: real business problems are trapped behind unclear ownership, normalized inefficiencies, and exception-heavy workflows. Adding AI doesn’t remove those barriers.

MIT studied failure from the outside. We’ve been living it from the inside.

What the research confirms

Across this series, we’ve documented patterns the research now validates:

• Leaders are more optimistic than frontline managers.
• Training investment is declining even as skill gaps widen.
• Most professionals see AI as enhancing, not replacing, human capability.
• Back-office use cases often deliver stronger ROI than headline-grabbing sales initiatives.

None of this is surprising. It’s measurable proof of what practitioners already know.

“Buy, don’t build” — and what that misses

MIT recommends buying instead of building. External vendor partnerships outperform internal enterprise builds.

That advice makes sense for enterprise-scale deployment. But this series isn’t about IT departments shipping production systems. It’s about leaders building AI literacy.

You cannot intelligently buy what you do not understand.

Many of the 95% failures come from organizations trying to leap from zero literacy to full deployment. They’re buying solutions to poorly defined problems from vendors they can’t properly evaluate.

That’s not a technology problem. It’s a literacy problem.

The internal failure that built literacy

One of my early AI attempts would qualify as a failed internal build. The agent didn’t work.

But that “failure” revealed the real issue wasn’t automation — it was process ownership. The final solution was simpler and more effective.

The research measures project outcomes. It doesn’t measure literacy gains.

Literacy compounds. And that compounding effect is what ultimately drives meaningful ROI.

The ROI trap

Wharton highlights that most enterprises now formally measure AI ROI. That sounds responsible. It can also be paralyzing.

Small experiments rarely survive formal ROI review. Yet those small explorations lay the foundation for later gains.

If you demand proof before practice, you train yourself to be a spectator.

Literacy first. ROI second.

What both reports miss

Both studies precisely describe the problem: learning gaps, training failures, skill shortages, stalled deployments.

Neither provides a practical path for individuals. There’s no clear progression from personal experimentation to organizational competence. No acknowledgment that literacy cannot be purchased.

That’s the missing piece.

Start small. Solve real problems. Let AI be your collaborator. Build pattern recognition. Organize your learning so it compounds.

Research can measure the divide. It can’t show you how to cross it.

The sequence that actually works

1. Leaders build personal AI literacy.
2. Teams reach critical mass.
3. Leaders can intelligently evaluate vendors.
4. Enterprise deployment succeeds.

Skip the literacy phase, and you risk becoming part of the 95%.

Your Week 20 challenge: find yourself in the data

Where are you?

• Haven’t started? Spend thirty minutes exploring one frustrating problem.
• Tried and stopped? Re-engage with a structured diagnostic approach.
• Using AI, but plateaued? Identify the barrier blocking your next step.
• Seeing results? Help someone else begin.

The research measured the divide. Only you can decide to cross it.

The bottom line

MIT and Wharton confirmed the shadow AI economy, the 95% failure rate, the training crisis, and the enhancement-over-replacement reality. They validated the patterns.

What they couldn’t provide is the lived path forward.

The choice isn’t buy versus build. The choice is to build literacy first, then buy intelligently — or buy blindly and risk becoming part of the statistic.

Nineteen weeks ago, I was waiting for AI to prove itself. Now I’m building with it.

The real question isn’t whether the research is right.

It’s what you’ll do in the next thirty minutes.

This post is part of my “AI Over 40” series. It first appeared on LinkedIn: AI for the Over 40 [Week 20]: What MIT and Wharton Found – And What They’re Still Missing.

Next Up: What is an MCP server and why I’m building my own.

Read more AI and Copilot blogs.

Trending Posts