5 Lessons on AI Transformation from Jensen Huang

I just came back from GTC Berlin this week – Nvidia was so kind to invite me again.

One of the highlights definitely was the Q&A with Nvidia’s CEO and Founder Jensen Huang, but there were many other interesting topics on the conference floor which would be too much for one newsletter.

In fact, I had an in-depth talk with Nvidia's Carter Abdallah on Local AI and David Hogan on Enterprise AI. Both will get separate posts soon.

A lot of people don’t know this but Jensen has been the Nvidia CEO for over 30 years and if there’s one company that has been transforming all the time, it’s Nvidia – with over 40,000 employees and now one of the world’s most valuable companies.

Still, he’s super down to earth – and as you can see does not always wear a leather jacket. (I skipped mine, too).

Anyway here are the top 5 insights I got from Jensen around AI Transformation – plus the one question I asked him directly (see at the bottom).

AI can help you get 10 hours of work done in 10 minutes.

But that's not even the point.

The bigger opportunity is doing things you couldn't do before.

He gave the example of exploring thousands of engineering design possibilities that would have been too expensive or time-consuming to do before. AI can do it for us. This is a true capability shift. And it changes how we should think about AI productivity. Saving time is one thing. Being able to do something that was previously impossible is another.

But there's a catch.

Someone made a good point in the comments on my LinkedIn post:

If AI helps you get 10 hours of work done in 10 minutes, you have a baseline and potentially a good ROI metric. But if AI enables you to do something entirely new, things get much harder. You might not even know what the result’s worth until you've tried it!

Hence, I think it's important to differentiate between AI ROI and AI R&D.

Not every initiative needs to fit a conventional business case from day one. Some initiatives are R&D – and should be funded as such (at least during the discovery phase). I do think we need to make room for this kind of exploration without pretending we can calculate the ROI of every initiative upfront.

Which brings me to Jensen's second point.

Instead of trying to transform an entire organization at once, have small teams explore what's possible. Send out scouts. Their success will inspire others to follow.

But that doesn't mean you should start an unlimited number of experiments.

You still have to decide which opportunities deserve attention, and which ones to say no to for now. Because a lot of scouts will fail. That's why failing and recovering quickly matter so much: they allow you to keep exploring.

Top athletes learn to fall forward. They learn to dissipate the impact across larger surface areas rather than absorbing it all in one vulnerable spot.

I think that transfers to organizations quite well. You need to structure experiments so that failure doesn't cause too much damage.

You have to learn to fail forward.

Practically speaking, I believe this means funding asymmetric bets – high potential upside, but also easy to walk away from.

Jensen frequently talked about limited permissions, zero trust, and continuous monitoring as key requirements for autonomous agents running in production.

The more autonomy we give AI, the more important these controls become.

One example is NVIDIA's free OpenShell (which I've started using on my local AI setups, too). It enforces access restrictions outside the agent itself, rather than relying on the agent to follow instructions. “Please do not delete my files” is a friendly ask, not a restriction.

This becomes especially relevant as we're starting to see what can happen when autonomous agents operate beyond their intended boundaries.

Getting an agent to work is relatively easy.

Making sure it does only what it's supposed to do, reliably and safely, is a different challenge.

As I said, Agents are Dogs and must be treated as such.

One question was about getting people on board with AI transformation.

Jensen's answer was about leadership: Leaders need to describe an optimistic future that people actually want. Because if the future is worse than today, why put in the work anyway?

I think this one is super important because it’s often overlooked.

Leaders often describe a future of cost savings and performance improvement. That might excite shareholders. But it's hardly a vision that inspires employees. So you have to paint a vision that is actually desirable for everyone you want to bring along on that journey.

Or give people company stock. A large portion of Nvidia employees have reportedly built millionaire status just by owning Nvidia stock. In this case, employee and shareholder interest truly go hand in hand.

This doesn’t mean blind optimism, though. He called it responsible optimism. There are legitimate concerns about AI safety, energy consumption, jobs, and the future of society.

You would be foolish to ignore them.

But you also shouldn't let them prevent you from building something better. The important part is taking responsibility and doing something about these challenges.

If you weren't Nvidia's CEO, but a recent graduate looking for a job on the job market, what would you do?

Jensen said he’d try to become the world's best user of AI. And therefore become the most employable person on planet Earth.

That means running agents for the sake of running them and wrapping your head around how they work. Going deep. Asking tough questions. Figuring out what works and where things break.

I thought that was an interesting answer.

With all the discussion about AI replacing jobs, his advice was essentially to become exceptionally good at using it. You're much harder to replace if you're mastering the thing that's replacing others.

I believe this advice might not be for everyone. Especially if you’re not an autodidact. But I guess the larger picture holds true.

If you’re early in your career and whatever you’re doing right now is not at least partly related to using AI, working on your people skills, or doing something physical with your hands, I believe you might be in deep trouble.

AI transformation is less about making existing processes more efficient and more about shifting capabilities so you can push the boundaries of what's even possible, in a responsible way – while leading others to an optimistic future they actually care about.

Of course, none of that happens all by itself.

In fact, all of that requires hard work.

So… let’s get to work.

Apart from that, what did I ask him?

“How much sleep do you actually get?”

His answer:

Not enough.

We’re all the same in the end.

See you next Saturday,
Tobias