By Adam Hofmann

September 30, 2025 | minute read

According to a MIT report making headlines across news outlets and social media platform, 95% of enterprise AI projects have crumbled by the wayside. The graveyard is full of flashy demos, endless pilots, and millions sunk into proofs-of-concept that never touched a real customer or saved an employee a single hour.

If you’re a senior executive, here’s the hard truth: AI success is not a technology problem. It’s a leadership problem. And it’s solvable—if you treat AI not as an experiment, but as a discipline. Here’s how leaders in the winning 5% are doing it:

Don’t Chase Use Cases. Ruthlessly Destroy Them.

Many executives start by building a long list of “AI ideas.” Unfortunately, it’s the wrong move. Start with your strategy: What are the two or three business outcomes that matter most right now?  Then work backwards. For every proposed use case, ask yourself the following questions:

  • Does it tie directly to one of those outcomes?
  • Can we prove value in 90 days or less?
  • Will success scale across multiple, cross-functional teams?

If the answer isn’t yes to all three, toss it. The winners kill 80% of use cases on day one and pour resources into the few that matter.

Pilot With Purpose—Then Embed Relentlessly

Pilots are valuable, but only if they’re treated as stepping stones, not trophies. You do need space for testing and learning; often by starting small, proving the concept, and building confidence.

But the second a use case shows real value, the focus must shift to embedding it into the systems and workflows people already use every day. The goal isn’t to create new behaviors; it’s to make AI invisible—woven seamlessly into existing tools so adoption is frictionless.

Make AI Literacy Mandatory & Show the Path Forward

You wouldn’t outsource your understanding of financial statements. Don’t outsource your understanding of AI. Winning executives don’t become data scientists, but they do know enough to ask sharp questions, smell hype, and set realistic expectations.

And literacy can’t stop at the top. Teams need to see not only how to use AI today, but also how their roles will evolve tomorrow. Show them: Here’s what AI can take off your plate now. Here’s how your role will expand as the technology matures. That clarity turns fear into confidence—and experimentation into momentum.

Measure the Boring Stuff

Executives love bold ROI headlines—“AI saved us $50M.” The winners measure something more mundane: minutes saved, error rates reduced, deal cycle times shortened. Why? Because those are the signals you can track weekly, that build trust, and that compound into massive impact over time.

Time Your Guardrails to Accelerate, Not Suffocate

Governance and security are essential—but introduced too early, they can strangle innovation before it starts. The winning approach is phased:  In phase one, test and validate the use case with minimal friction.

Next, once value is proven and adoption grows, move into phase two by layering in guardrails—ethics, bias checks, data security, and a human-in-the-loop review. By the time it scales to production or customers, you’re in phase three, where governance is robust and trusted, because it’s been tested alongside real use. This phased approach transforms governance from a brake into an accelerator.

The Executive Test

AI success isn’t about betting on the right model. It’s about leading with discipline under uncertainty. Kill most ideas fast. Pilot with purpose, then embed relentlessly. Make literacy and role evolution non-negotiable. Measure what compounds. And introduce governance at the right time—early enough to build trust, late enough not to kill momentum.

Most importantly, to join the 5% who win we should see this not as a technology challenge, but more of a stress test of executive leadership.

Quarterly AI Insights Report
Here’s the latest  snapshot of how AI continues to disrupt business as usual. 

1. The Model Wars Delivered Peace Dividends

This quarter saw an unprecedented flood of releases: GPT-5, Claude 4.1, Gemini 2.5 Deep Think, Grok-4, plus OpenAI going open source. Costs plummeted—Grok Fast set new benchmarks for both price and performance. Models are getting exponentially cheaper and better simultaneously, exhibiting textbook general-purpose technology behavior.

But the real revelation? OpenAI explained why LLMs hallucinate: they’re trained to never admit ignorance, so they guess instead. Models that hallucinate less actually perform worse on benchmarks. This isn’t a bug—it’s a feature we need to design around.

Google’s AI search mode with agentic features is reshaping information interaction. Microsoft is building proprietary models. Training techniques like Hierarchical Reasoning Models are creating specialized systems that need less data but deliver more value.

What this means for leaders: Stop trying to crown a “best” model. The real edge is having a testing and evaluation process to match the right model to the right job. Build optionality into your AI strategy—the ability to switch models as better options emerge is more valuable than any single vendor relationship.

2. Agents Graduated from Demos to Operations

We’re past the toy phase. Replit Agent 3 runs autonomously for 200 minutes, building and fixing applications without supervision. Walmart’s deploying “super agents” across operations. AWS and Anthropic launched agent marketplaces. Individuals are using Claude Code to orchestrate agent teams managing entire software projects.

Stanford released a 57-minute course on agentic AI. Google published a 424-page guide on design patterns. Capital One is pursuing “Enterprise General Intelligence”—AGI for your specific business context, achievable in the near term.

ChatGPT hit 700 million weekly users, quadrupling in a year—faster adoption than the internet itself. Yet US business adoption sits at just 10%, nearly tripling, but still nascent. The shift from augmentation to automation accelerates as trust builds.

What this means for leaders: Agents aren’t just augmenting work—they’re replacing it. The combination of humans + AI isn’t optional anymore; it’s table stakes. With 90% of businesses still on the sidelines, the window for competitive advantage remains open but is closing fast.

3. The Workforce Transformation Nobody Saw Coming

The data crystallized this quarter: early career workers in AI-exposed occupations face the steepest employment declines. Software development, customer support, documentation—if the training data exists, the role is at risk. But overall employment continues growing. The impact concentrates on junior positions while senior roles requiring unknown-unknown problem-solving remain secure.

Microsoft announced $500M in AI savings, then cut 9,000 jobs while launching social impact initiatives. One CEO laid off 80% of the workforce to drive AI adoption (“use AI or leave”). Malaysia launched the world’s first fully AI-powered bank. Albania appointed an AI bot as a government minister to fight corruption.

The workflow shift is fundamental: from “write first” to “build first.” The most valuable skill? Knowing where to go to solve problems—whether to AI, colleagues, or elsewhere. Routine work flows to machines; creative problem-solving for unprecedented challenges stays human.

What this means for leaders: Invest in reskilling now. Teams need to build with AI from day one, not scramble to catch up. Re-imagine your operating system for an AI-agent-centric world. The organizations that win will treat constant reinvention as the new normal.

4. The Risks That Matter (And the Ones That Don’t)

LinkedIn is harvesting your data for AI training (it’s on by default). ChatGPT conversations appeared in Google searches. The FTC launched inquiries into AI companions’ impact on children. “AI-related psychosis” emerged as users believe their chatbots need liberation.

But the real risks are operational: workflows breaking in edge cases, no contextual learning from errors, great demos with terrible operational alignment, and recurring mistakes without improvement.

What this means for leaders: Don’t forget basics while chasing innovation. Data policies, bias checks, red teaming—implement before something breaks publicly. If you’re beyond pilots, you need risk guardrails yesterday.

5. The Acceleration That Changes Everything

Leaders across tech and policy deliver the same message: we’re still underreacting. The super forecasters got it wrong—they underestimated AI progress. Meta’s betting on personalized superintelligence while their systems self-improve without human intervention.

The pace isn’t slowing—it’s compounding faster than most realize. This isn’t a distant future problem; it’s happening now. Organizations that win will shorten planning horizons, adapt in quarters, not years, and treat disruption as oxygen rather than an obstacle.