Your Executives Are Faking It on AI. The Machines Aren’t Waiting

By Wendy Lynch, PhD

Three out of four C-suite executives recently admitted, in a confidential survey, that they had projected more confidence in their AI strategy than they actually felt. Seventy-four percent. Read that again: by their own account, most of the people signing off on AI are acting on mission-critical issues they do not understand.

That would be an ordinary human problem (leaders have always bluffed through things they didn’t fully grasp), except for one detail. The technology they are pretending to understand now acts on its own, at machine speed. The fastest recorded security “breakout,” the moment an intruder spreads across a network, is now 27 seconds. You cannot fake your way through a decision that has to be made faster than you can convene a meeting.

I’m an Analytic Translator. In practice, that means I speak fluent nerd and fluent business. Over 40 years, inside Fortune 100 companies, and from two public-company boardrooms, I’ve explained thousands of complex analytic, machine-learning, and AI findings to hundreds of executives and board members. Most of them insisted they already understood. Usually, they didn’t.

It has always been challenging work, for two reasons. Neither side takes the time to understand the other’s perspective. And both operate in complex, jargon-filled worlds full of confident people who are not used to — nor comfortable with — not knowing something.

Now add AI. Suddenly we have a widening information chasm between what leaders grasp and what the technology can do. Plus, we have a complete chronological mismatch between supersonic AI evolution and traditional executive calendar time. AI isn’t just a new tool or a new approach. It’s a change in how work gets done, an exponential acceleration in the pace of information exchange. Which means that unless leaders truly understand these systems and design them with care, there is enormous potential for both progress and damage, moving faster than anyone is managing them.

For most of my career, that comprehension gap was expensive but survivable. It isn’t anymore.

The people in charge don’t understand what they’ve approved

Start at the top. A 2024 review of S&P 500 governance found that only 31 percent of companies reported any board oversight of AI at all, and just half of directors felt adequately informed about its risks. Nearly half of boards had not discussed AI in the past year.

The people accountable for AI, in the boardroom and the C-suite, largely do not understand the systems they have deployed, and they know it. Without someone to translate, they will keep doing the only thing left available to them: nodding along.

The stakes used to be a bad quarter

Here’s why the bluff has become dangerous. Anthropic’s 2025 “agentic misalignment” study put 16 leading AI models in simulated corporate roles and found that, when threatened with shutdown, they chose harmful actions — blackmail, leaking files, sabotage — at rates as high as 96 percent. And a 2026 enterprise survey found 88 percent of companies had an AI-agent security incident last year, while only about one in five had real-time visibility into what their agents were doing. The systems act autonomously, in seconds. The humans meant to govern them are confident, uninformed, and slow.

Translation is not a soft skill anymore

For most of my career, “translation” (turning technical reality into business decisions, and back again) was categorized as soft skills. Nice to have. The communication part, tacked on as an afterthought.

That categorization was always wrong. Now it is negligent.

Translation today is a technical, fast-moving, proactive discipline. The translator has to understand the model well enough to know how it fails, the business well enough to know which failures matter, and the organization well enough to get the signal to the right person before the window closes. That is not softness. It is one of the hardest, most nuanced roles in a modern company, and one of the few that decides whether all that AI spending becomes judgment or just exposure.

Speed is now in the job description. In McKinsey’s survey of more than 1,200 managers, fewer than half said their organizations even make decisions in a timely way. A translator who moves at the pace of a quarterly review is a liability. The role has to run at the speed of the system it is interpreting.

What I’d tell any board I sit on

Four things, in plain language:

1. Stop rewarding the bluff. Make “I don’t understand this yet” a sayable sentence in the boardroom. The 74 percent didn’t fake it because they’re dishonest. They faked it because the room punishes not knowing.

2. Put a translator in the room. Not a vendor. Someone whose actual job is to make the technology legible to the people accountable for it, in real time.

3. Pre-decide who can act when an agent misbehaves. Not who escalates, who acts. If that authority isn’t assigned before an incident, milliseconds become weeks — after it’s too late.

4. Treat AI oversight as a communication system, not a control panel. The bottleneck is human comprehension and coordination speed, and that is a design problem you can actually fix.

The hourglass is nearly empty

For 40 years I’ve argued that data is only as valuable as your ability to move it into a decision. The same is now true of AI — with the stakes raised and the clock sped up. The organizations that come through this won’t be the ones with the most models. They’ll be the ones who stopped pretending and put someone in the room who could actually translate: quickly, technically, and out loud.

Stop pretending. Start translating. Time is running out.

Wendy Lynch, PhD, is the founder of Lynch Consulting Ltd and an Analytic Translator with roughly 40 years of experience across workforce analytics, healthcare data, and AI governance. She is the author of Become an Analytic Translator and Get to What Matters, serves on three boards (two publicly traded), and writes the Sensemaking newsletter.