Are Most Enterprises Getting AI Adoption Wrong
From Day One? An NTU Professor Names the Blind Spot

A small short-term bubble, but the long-term trend holds

Humans tend to imagine a vision first, before they get a chance to reshape the future. That means a gap between expectation and the details of implementation is inevitable in the short term — the tools are powerful, but an organization isn't just about tools; it also involves people, culture, and even the supply chain and satellite factories.

That's why Professor Lee believes a "small bubble" will happen in the short term, but over the long run, AI will change daily life and work habits as profoundly as the internet did, becoming indispensable infrastructure.

Why AI transformation is harder than digital transformation: a missing shared language

Digital transformation can be broken into four stages: digitization, digital optimization, digital transformation, and digital reinvention. The core value of digitization is really about reducing information asymmetry; digital optimization requires eliminating waste in the process first — otherwise automation just "automatically manufactures waste at scale."

What sets this wave of AI transformation apart from the past is that LLMs let decision-making become more than just "gathering information" — they let people co-create a knowledge system with AI and eliminate blind spots in human decision-making, with AI acting like an experienced mentor giving critical feedback at every stage.

Even so, most enterprises still find AI adoption harder than expected — and the key isn't technology, but the fact that users and AI experts often share no common language. Just clarifying what each side means by the terms they use eats up an enormous amount of communication effort.

The truly scarce talent isn't an AI expert — it's a "bridge"

Enterprises often mistakenly believe that hiring an AI expert or consultant will solve the problem, but AI teams frequently can't communicate directly with domain experts. The truly critical talent — and the hardest to develop — is the bridge role (the PM) who understands both domain knowledge and the fundamentals of AI. That person has to speak both languages and have the right communication skills.

This kind of talent can't be developed overnight, and most enterprises don't have the resources to cultivate it directly — which is exactly why many organizations still "suffer for a while" during implementation even after hiring AI experts.

Culture clash: review culture vs. a learning organization

Taiwanese companies broadly share a "review culture" — checking progress and KPIs every quarter. This culture helps sustain performance, but it clashes with the "learning organization" culture that AI transformation requires: the latter needs time for dialogue, reflection, and internalization, and results may not show up in the short term.

Leaders need to balance both bottom-up and top-down forces: on one hand, they must embrace the different thinking and working styles the new generation of employees brings; on the other, leaders themselves must lead by example, personally using new tools and understanding where the technology stands, so they're equipped to imagine a vision and lead the organization forward.

Managing morale is equally critical: AI projects often hit a stretch where "six months in, results are underwhelming, and a few more months later, still mediocre." Once morale takes a hit, talent walks. Keeping the team's confidence in the outcome intact — especially giving junior engineers visible wins — is a piece of the process that can't be overlooked.

Action items for enterprises and leaders

Drawing the interview together, a few concrete recommendations stand out:

1. Leaders lead by example first — Personally subscribe to and use LLM tools. Only hands-on experience lets you truly understand the technology's limits, and from there, imagine the vision.
2. Cultivate "bridge" roles, not just AI experts — Invest in the intermediary talent who understands both the business and the fundamentals of AI — they're the ones who decide whether implementation succeeds or fails.
3. Use POC pilots to lower risk — Test first in a simulated environment or a small-scale pilot, which costs less and makes it easier to confirm whether the deployment clashes with real-world conditions on the ground.
4. Define the "boundary of autonomy" — Clearly separate which repetitive tasks go to AI or robots, and which value-adding judgment calls should stay with humans — and expect that boundary to need repeated adjustment in practice.
5. Manage team morale — When results aren't visible early in a project, deliberately design checkpoints that let team members see concrete outcomes, to avoid the "strong start, fading resolve, exhausted by the third push" pattern.

This article is compiled from the interview content and views shared on the YouTube channel "Leadership Impact Academy," in the episode "Dialogue on Leadership" EP23 (source: https://youtu.be/hDwt_gfJ2EQ). It does not constitute a guarantee or commitment of any kind — please evaluate your organization's actual adoption strategy based on your own circumstances.