Why GenAI Leadership Requires an Alien Mindset
GenAI is a leadership test and it looks like most organizations are failing
To lead in a GenAI world, you need an alien mindset — one that can put you in conflict with you organization.
“What’s the ROI on our GenAI?”
The CFO looks up. The CTO looks down. Eyes shift toward the person who championed AI six months ago.
“Zero.”
A recent report from MIT found that while companies are pouring tens of billions into generative AI (GenAI) projects, only 5% of them are seeing a measurable return. (See the footnotes* for a summary of the study and its findings.)
"We've seen dozens of demos this year. Maybe one or two are genuinely useful. The rest are wrappers or science projects." - Anonymous CIO
How is it possible that enterprises with world-class engineering teams and a dragon’s hoard of data can find themselves struggling to find value in the extraordinary new GenAI capabilities being announced seemingly every day — despite a world full of urgent problems to solve?
The problem here isn’t with foundation models, it’s with the foundations of leadership.
No “measurable” returns
The costs of adopting AI are easy to calculate: cloud subscription fees, software integration, training budgets, etc. Those numbers fit neatly in a spreadsheet.
But the benefits? Those don’t appear as cleanly. How do you measure the value of ten marketing drafts instead of one? Or a legal contract that’s faster to produce but needs more careful review? Or an employee who feels less burned out because an AI tool alleviated some of the drudgery?
The scales are lopsided. So is this really an ROI problem, a measurement problem — or both?
The benefits of GenAI are often downstream and qualitative: improved customer experiences, faster ideation, reduced cognitive load on employees. But unlike cost metrics, which are immediate and tangible, benefit metrics often require time, context, and experimentation to reveal themselves. This is where the conversation shifts from technology to leadership.
Most organizations aren’t struggling with the technology. They’re struggling with the mindset.
It takes special skills to define and measure the value of a GenAI project. The MIT report* shows budgets skewed toward highly visible, easy-to-measure use cases. But the biggest opportunities — the transformative ones — usually hide beyond these safe, board-pleasing choices.
The problem of endless right answers
A big part of the problem is that most leaders are using a one-right-answer measurement playbook for a many-right-answers technology.
Traditional AI is for automating tasks where there’s one right answer.** It’s a bullseye machine: for each input, there’s a best output. Scoring it is like scoring target practice.
Generative AI is for automating tasks where there are endless right answers… and, to pilfer shamelessly from Tolstoy,*** each right answer is right in its own way. When you ask ChatGPT to email-ify a to-do list, you get a fairly solid result. When you repeat the same prompt, you get a different perfectly adequate email. Both are right answers, but which one is right-er?
GenAI plays like jazz: countless riffs, each with its own tone and spark. It’s less a crossword than an improvisation. The measure isn’t “Was it correct?” but “Was it useful?”
But “usefulness” is a can of worms. It raises the stakes for leaders, demanding sharper judgment and broader skill.
What business school won’t teach you
If you’re comfortable judging the performance of systems that can churn out different endless right answers to the same question (at scale!) you’re either a rare outlier or from the future. For most leaders, this way of thinking feels downright unnatural.
Yet it’s revolutionary. And for visionaries, the opportunities are immense.
So let’s unpack the mindset shift GenAI demands — and why no leader can sidestep it.
Some tasks resist automation for the same reason they resist measurement: they are soaked in complexity and ambiguity. The more cleanly an enterprise-scale automation problem can be specified and measured, the less you need GenAI for it — and the more likely a conventional solution already exists.
Consider the difference between “calculate this sum” and “make this more elegant.” The first is checklist-friendly; the second depends on taste, judgment, and shifting context. Tools that tackle such vague requests resist easy scoring. Without a human in the loop, their outputs are brittle.
Generative AI doesn’t work like a calculator that always gives a single, fixed answer. Instead, it produces distributions of possible outputs, each shaped by randomness, context, and the framing of the prompt. Think of it less like retrieving the one “correct” number and more like rolling a loaded die that can land in different ways depending on what you ask.
When success depends on judgment, taste, or context, you can’t grade those outputs with a single metric the way you’d score a math test. Accuracy works if the question has one right answer — but for a piece of writing, a design concept, or a strategy draft, the “best” choice depends on novelty, fit, tone, timing, and audience. No single yardstick captures all of that. So measurement lags behind, chasing shifting value it can never quite pin down to one definitive score.
Wired for certainty
That’s a tectonic shift in mindset. Most organizations still expect crisp, provable accuracy and get rattled when GenAI feels subjective, ambiguous, or slippery to measure.
When we grasp that GenAI is about useful answers, not right ones, we’ll likely find ourselves at odds with our organization. Why? Because institutions are wired for certainty. They’re built on KPIs, compliance checklists, and standardized metrics, all of which assume a single, provable truth.
GenAI thrives on multiplicity, judgment, ambiguity, and context. It asks leaders to value usefulness over correctness, adaptability over standardization. That’s not how most org charts, incentive systems, or governance processes were designed.
So when leaders start embracing the jazz of GenAI, they often clash with organizations and boards still demanding bullseyes.
The modern leader’s imperative
Generative AI isn’t just another technological wave, it’s a leadership revolution. It requires leaders who:
Embrace ambiguity rather than fear it.
Align diverse voices rather than defer to a single metric.
Focus relentlessly on value rather than outputs.
Create clarity around decision rights and accountability.
As many-right-answers thinking moves from human to machine output, scale unlocks unprecedented opportunities. But as we leave behind the single-best-answer mindset, easy wins grow scarce. If you and your organization are ready to embrace the change, the opportunities are enormous. It falls to you to frame opportunities so your organization can harness them.
That’s why, before jumping into any generative AI project, think in terms of what would be valuable for your organization. Scattered pilots rarely scale — they seldom align with what truly matters to the business. Instead, start with a vision and value, then work backwards from there.
As the leader, creating meaning out of ambiguity will fall squarely on your shoulders. There will be risks, but also great rewards. Today’s models and tools are the worst they’ll ever be. Tomorrow’s will be better. The sooner you can adapt to the coming wave, the further it will take you.
Thank you for reading — and sharing!
I’d be much obliged if you could share this post with the smartest leader you know.
In other news, the first few cohorts of my Agentic AI for Leaders course were a triumph and we’ve opened enrollment for an additional cohort to meet demand.
Enroll here: bit.ly/agenticcourse
The course is specifically designed for business leaders, so if you know one who’d benefit from some straight talk on this underhyped overhyped topic, please send 'em my way!
Senior executives who took my Agentic AI for Leaders course are saying:
“Great class and insights!”
“Thank you for teaching AI in an interesting way.”
“…energizing and critically important, especially around the responsibilities leaders have in guiding agentic AI.”
“Found the course very helpful!”
🎤 MakeCassieTalk.com
Yup, that’s the URL for my public speaking. “makecassietalk.com” Couldn’t resist. 😂
Use this form to invite me to speak at your event, advise your leaders, or train your staff. Got AI mandates and not sure what to do about them? Let me help. I’ve been helping companies go AI-First for a long time, starting with Google in 2016. If your company wants the very best, invite me to visit you in person.
🦶Footnotes
* Here’s a quick summary of the MIT Media Lab / Project NANDA report titled The GenAI Divide: State of AI in Business 2025:
95% of organizations fail to see measurable ROI from GenAI.
Winning AI tools require minimal setup and deliver immediate, visible value.
Big firms lead in pilot volume but lag in scale-up. Successful rollouts by mid-market firms averaged 90 days; enterprises took nine months or more.
The highest AI budgets were allocated to (board-friendly) marketing/sales use cases, while the best ROI came from automating back-office tasks.
Internal builds failed at twice the rate of external partnerships… yet most enterprises still attempted to reinvent the wheel with their own AI tools.
Just 40% of companies report official LLM purchases, while 90% of employees use personal AI tools daily. The ROI measured from sanctioned deployments is likely to miss this “shadow AI” productivity.**
Methods: The report (Jan–Jun 2025) draws on a review of 300+ public AI initiatives, interviews with 52 organizations, and survey responses from 153 senior leaders at 4 major industry conferences.
5 tips for running a successful GenAI pilot:
Keep up: Adaptive tools evolve with workflows; static ones flatline.
Go narrow: Start with a high-value, bounded use case, then expand.
Embed: Integrate into existing systems (Salesforce, ERP, CRM) with minimal friction.
Win trust: Show deep process understanding, protect data, and deliver results fast.
Show quick wins: Prove value in weeks, not quarters; land small, then scale.
** But what incentive do workers have to share their productivity gains with their employer, especially when that employer doesn’t pay for the tools? We might see the work becoming shorter/easier instead of more profitable for firms.
*** Where do unsupervised learning and reinforcement learning fit in? Each of these is going to blow a leader’s mind(set) a lot less than GenAI with its many right answers. Unsupervised learning is less about direct automation and more about surfacing hidden patterns —raw material for model-building rather than business-ready engines. Reinforcement learning, by contrast, plays a different game: there are many decent sequences of actions but only one best sequence (whether we find it or not) that optimizes an unbounded objective… but it’s very different from GenAI from the builder’s point of view because reinforcement learning has a crystal clear objective function and a limited set of actions the system can take at each step in the sequence. From the adopter’s point of view, GenAI is stranger terrain: many right answers, but none is definitively the “best.” That open-endedness unsettles leaders far more than the tidy optimization of older approaches.
**** The opening line of Tolstoy’s Anna Karenina is “All happy families are alike; each unhappy family is unhappy in its own way."
🗞️ AI News Roundup!
In recent news:
1. Anthropic overtakes OpenAI in enterprise market share
Enterprise LLM API spending jumped to $8.4 billion in six months, with Anthropic now leading at 32% market share ahead of OpenAI (25%) and Google (20%). Anthropic also raised $13B at a $183B valuation. As if to celebrate, the Claude data policy changed so you’ll need to manually opt out of chat training.
2. China tightens AI rules with mandatory content labelling on social platforms
WeChat, Douyin, Weibo, and Xiaohongshu have introduced new systems to meet Beijing’s law requiring visible labels and metadata watermarks on all AI-generated text, images, audio, and video. The regulation, part of the 2025 “clear and bright” campaign, reflects heightened concern over misinformation, copyright abuse, and online fraud.
3. AI model outperforms WHO in predicting flu vaccine strains
A Nature Medicine study finds that the machine-learning platform VaxSeer more accurately selected influenza vaccine strains than the WHO in most of the past decade. Using genetic and antigenicity data, the model outperformed WHO picks in six of ten years for H1N1 and nine of ten for H3N2, with predictions correlating well with real-world vaccine effectiveness and disease burden reduction.
4. Nvidia says two unnamed customers made up 39% of record Q2 revenue
The chipmaker posted $46.7 billion in quarterly revenue, up 56% year over year, with one customer accounting for 23% and another for 16%. While Nvidia only identified them as direct buyers such as OEMs or distributors, analysts warn the concentration poses risks even as cash-rich clients continue heavy AI data center spending.