🤿 Deep Dive:🗽 AI Summit New York 2025 - AI Grows Up

Hello AInauts,

Welcome to Deep Dive! Last week, we attended the AI Summit in New York. We thought we knew what to expect: agentic AI, foundation models, the usual buzzword bingo. Nope. Not this time. Instead: focused activity.

It's as if someone turned down the hype dial and turned up the reality check. Less excitement, more substance. Fewer promises, more questions. And answers we didn't expect …

This is how the post is structured:

• What the AI Summit New York 2025 promises

• The end of proof-of-concepts

• The agent revolution: Why there's more to it this time

• The inconvenient truth about enterprise AI

• AI vs. AI: The new arms race

• What GenAI really means now

• The notable absentees

• The questions that remain

• Our take: What you should take away

• And then there's NYC in December...

Let's get started—with a few impressions to set the mood 😁.

Last Time the Market Was This Expensive, Investors Waited 14 Years to Break Even

In 1999, the S&P 500 peaked. Then it took 14 years to gradually recover by 2013.

Today? Goldman Sachs sounds crazy forecasting 3% returns for 2024 to 2034.

But we’re currently seeing the highest price for the S&P 500 compared to earnings since the dot-com boom.

So, maybe that’s why they’re not alone; Vanguard projects about 5%.

In fact, now just about everything seems priced near all time highs. Equities, gold, crypto, etc.

But billionaires have long diversified a slice of their portfolios with one asset class that is poised to rebound.

It’s post war and contemporary art.

Sounds crazy, but over 70,000 investors have followed suit since 2019—with Masterworks.

You can invest in shares of artworks featuring Banksy, Basquiat, Picasso, and more.

24 exits later, results speak for themselves: net annualized returns like 14.6%, 17.6%, and 17.8%.*

My subscribers can skip the waitlist.

Skip waitlist

_{*Investing involves risk. Past performance is not indicative of future returns. Important Reg A disclosures: masterworks.com/cd.}

What the AI Summit New York 2025 promises

December 10, 2025. Ten years of AI Summit. It's freezing outside, and the biting wind is almost blowing my hat off. The Hudson River is steaming in the December air.

And I want to get right into the thick of things... Right into the crowd of 5,000 people who all want the same thing—to understand where the AI journey is headed.

But after the first few sessions and conversations, I quickly realize that this year feels different than last year. Not because the presentations and panels are better. Or because the demos are more impressive. But because no one is asking, "When will AI really change things?"

Instead, the question is: "How quickly can we keep up?"

The AI Summit New York has grown from a niche gathering for researchers and early adopters to a stage where companies, providers, and innovators come together.

It's no longer a nerdy tech conference, but a business event where people talk about technology. That's a difference, and a big one at that.

When the AI Summit 2016 took place for the first time, AlphaGo had just shocked the world (documentary here) by defeating the world champion of Go. ChatGPT didn't even exist as a concept. GPT-3 was five years away.

Since then? First commercial chatbots (2018). GPT-2 (2019). The pandemic as an accelerator (2020). GPT-3 with 175 billion parameters (2020). Then the big bang: ChatGPT in November 2022 – 100 million users in two months, the fastest-growing consumer app ever. Launch of the AInauten newsletter (December 2022). And now: Agentic AI.

The end of proof-of-concepts

Last year, the focus was on "AI in Production." The big challenge was: How do we get the models out of the lab and into real workflows? How do we convince management? How do we measure ROI?

All valid questions. But this year, we've taken it a step further.

One of the sessions was called "Cool Demo. Now What?" And honestly, that sums up the mood perfectly. The wow phase is over. Now it's time to get down to business.

No one is discussing whether AI should be used anymore. The conversation has shifted. It's now about scaling. About governance. About what happens when not one model is working, but several—and they are orchestrating each other.

"If one story defines 2025, it's the dawn of the AI Agent."

"If there is one story that will shape the year 2025, it is AI agents." That's how the organizers put it. According to AI analysis of the presentations, "agent" ranks third among the most frequently mentioned terms, with over 2,500 mentions.

A year ago, companies were still asking: "Should we test ChatGPT for customer service?" Today, they are asking: "How do we build a system in which one agent categorizes the request, a second searches the knowledge database, a third formulates the response, and a fourth checks the quality—all without human intervention?"

That's a completely different discussion.

The agent revolution: Why there's more to it this time

"AI agent" is still being used as a buzzword. Every vendor suddenly has "agentic AI" in their portfolio. Every press release promises "autonomous workflows."

And although this sounds suspiciously like the next hype cycle, there is more to it than that.

The basic idea behind an agent is simple: instead of having an AI model perform a task on demand, you build systems in which several specialized agents work together. One researches. One analyzes. One writes. One checks. They pass tasks to each other, just like a team of employees.

What distinguishes agents from previous AI applications is their autonomy. A classic chatbot waits for your input and responds. An agent can act independently.

It recognizes that it needs more information, obtains it, decides what its next step will be, and executes it. This may sound trivial, but it represents a paradigm shift.

Think about it: Until now, AI was a tool. You used it. With agents, AI becomes a colleague. One that takes on tasks, not just assists.

The exciting question that will be discussed at the summit: What happens when AI tools start working together instead of performing tasks in isolation?

The answers are both fascinating and disturbing.

Fascinating, because multi-agent systems can do things that individual models cannot. They can break down complex tasks into smaller steps, combine specialized knowledge, and correct each other.

It's unsettling because no one really knows what will happen when these systems start making decisions that we can no longer understand.

And the even more intriguing question, which is best not asked too loudly: What happens to the people in these workflows?

In many licensed and regulated professions (doctors, nursing, lawyers, finance, security, etc.), people still have the final say. But responsibility is shifting nonetheless.

A new concept is doing the rounds: "Human above the loop" instead of "human in the loop." The difference? Until now, humans were the controllers who approved AI outputs. Now they are becoming strategists who set the direction—while AI takes care of the execution. It sounds like a subtle difference, but it's more than that.

We feel that this issue can no longer be ignored in 2026.

The inconvenient truth about enterprise AI

From experiments to enterprise. From models to systems. From hype to implementation. From "Will AI change anything?" to "How quickly can we keep up?"

The figures speak for themselves: more than 60 countries have now published national AI strategies. 65% of companies already use generative AI in at least one function.

Deloitte predicts that by the end of 2025, a quarter of all enterprises will have agentic AI systems in use—and by 2027, half will. At the same time, 87% of cybersecurity experts experienced AI-powered attacks last year.

The AI Summit is not an academic gathering either. These are the people who actually implement AI in companies. CTOs of large corporations. AI teams from banks. Strategists from the pharmaceutical industry. CIOs of global conglomerates.

And, of course, many practitioners from smaller companies who want to learn from the big players—and are delighted to see how agile you can be when you don't have to go through twelve committees and seven levels horizontally and vertically.

And if you listen to them—really listen, not just to the polished keynotes, but to the conversations over coffee, the questions after the sessions—then you hear things that you won't find in any press release.

Firstly, many AI projects are not yet delivering the desired results and are even failing.

Not because the technology is bad. But because organizations are not ready. The reasons are now well documented: lack of alignment with business goals. Legacy systems. Data chaos. Change management. Skills. Governance. Culture. Seven hurdles—and technology is not one of them.

We know this from our own experience. The initial enthusiasm. The harsh reality that follows. Introducing an AI tool is easy. Scaling AI cleanly is incredibly difficult.

Many AI projects never make it past the pilot phase. Not because they fail technically, but because the organization doesn't get on board. Because the sponsor changes. Because the budget is cut after the proof of concept. Because employees don't want to use the new tools.

Secondly, the real winners are not those who experiment the most in the glass house.

They are the ones who implement most consistently.

The dominance of this topic is once again demonstrated by the organizers' word analysis: "Data" ranks second among the most frequently mentioned terms—with almost 4,000 mentions—directly behind "AI" itself.

It's about the tedious work: cleaning up data, standardizing processes, ensuring governance, training people. Less "Wow, look what AI can do!" and more "Okay, how do we integrate this into SAP?"

There are companies that have been experimenting with AI for years and still have no measurable business impact. And there are others that started with a single, focused use case and now use AI in hundreds of processes. The difference? The latter have understood that the technology is the easy part.

An example that sticks: Northwell Health is rolling out "ambient listening"—AI that listens in on doctor-patient conversations and automatically handles documentation. The goal? To reduce provider burnout. This is not a hype project. It is a concrete response to a real, complex problem (there are many requirements in this area).

Sexy? No. Effective? Absolutely.

Thirdly: Governance becomes a real competitive advantage.

Also unsexy, but true: The companies that are building their AI governance frameworks now will be the ones that can scale quickly in the future.

Early adopters who invested in governance early on can roll out new use cases in weeks. The others need months—because the legal department has to start from scratch every time, because no one knows which data can be used where, because compliance processes do not exist.

AI governance is like good infrastructure: you don't notice it when it works, but you notice it immediately when it's missing. Practical tip: include a question in the procurement process asking whether the procurement is "AI-relevant" – and if so, initiate appropriate follow-up measures.

Recommended reading on this topic: “Governing the Machine” by Ray Eitel-Porter, Paul Dongha, and Miriam Vogel.

And fourthly – AI-ready versus AI-native ...

Ein Satz, der auf dem Summit für Stirnrunzeln sorgte: "The only way forward is not to modernize the business to be AI ready. It is to reimagine the business and to be AI native." 

AI-ready means sprinkling AI over the same old processes and calling it transformation. AI-native means the opposite: rethinking your business as if AI had existed from day one. Different architecture. Different assumptions. Different processes. Everything.

In other words, it is not enough to simply embellish existing processes with AI tools. Instead, the process or even the company itself needs to be rethought from the ground up. AI-native companies play a different game, with different rules.

Most companies choose the AI-ready approach because it is "easier" and because that is exactly what consultants sell.

AI-ready vs. AI-native—that's the difference between a band-aid and surgery. Both have their place. But only one solves the problem permanently.

AI vs. AI: The new arms race

An entire stage at the summit was dedicated to the topic of cybersecurity. The title: "AI vs. AI: The New Cybersecurity Arms Race."

The same technologies that make companies more productive also make attackers more productive. Gone are the days when phishing emails could be spotted from a mile away. Today, it is almost impossible to distinguish AI-generated phishing emails from real ones. Behind this is social engineering on steroids, with automated attacks that adapt in real time.

What this means in concrete terms is that attacks are becoming more personalized, faster, and harder to detect. AI can analyze in seconds which employees are most likely to fall for a particular scam. It can comb through social media profiles, imitate communication styles, and send perfectly timed messages. This is not theory. It is already happening.

On the other side is AI-based defense. Systems that detect anomalies before a human could. That analyze millions of log entries in real time. That predict attack patterns.

It's an arms race. And both sides have access to the same weapons.

So here's a reminder: if you introduce AI without thinking about security at the same time, you're building a house without locks. Attackers can use AI to attack your AI systems. The question then is whether your AI defense learns faster than your attacker's AI.

Welcome to the future. It's complicated...

What GenAI really means now

It was also interesting to note what was NOT the focus: generative AI.

Last year, GenAI was the central topic. ChatGPT here, Midjourney there, everyone was talking about it. This year? GenAI is one of ten tracks. No longer the main attraction.

However, this is not a sign of declining importance, but rather a sign of maturity.

Generative AI has become a commodity. Just like cloud computing ten years ago. Instead of "Wow, AI can write texts!", the question now is "How can I use it to build products that people really want to use?"

The shift is noticeable. A year ago, the focus was on what was possible. Now, the focus is on what is useful. Beth Roth, AI strategist, sums it up: "Impressive technology is not inherently useful. Evaluate AI actions based on necessity, not capability."

That sums up the shift we are currently experiencing. A year ago, "Wow" was enough. Now everyone is asking: "And what does that mean for me specifically?"

The companies that have understood this shift are winning. Those that are still raving about the possibilities are stuck.

The hype is over. The work begins. And that's actually good news. Because now we can focus on what really matters: creating real value.

Takeaways: What do others say?

We have processed the AI Summit New York 2025 from our own perspective. But an event like this thrives not only on keynotes and slides, but also on the conversations in between.

(Video by The AI Summit Series)

Daniella De Grande (Global Operations and Transformation Executive) provided a pragmatic overview of the summit: even industries that one would not expect are involved in AI and motivated, but they often lack the expertise for integration.

That's why experienced support is needed, not only in the field of AI itself, but also in the modernization of legacy systems and infrastructure. Without this foundation, the result will not go live smoothly.

The key point remains down-to-earth: AI is a tool, and what matters is its benefits for people and its real business value, not technical gimmicks.

Tom Bendien (Founder & CEO of GT Edge AI) adds the competitive perspective: A small, early group is already going full throttle and achieving strong results. Over the next twelve months, this share will grow significantly, and the early pioneers will pull ahead.

For the lower half, however, he sees massive pressure, like in a race where you keep falling further and further behind. He found the large number of "wrapper" companies, which will probably not survive, particularly striking. It brings back memories of the dot-com era.

However, he does not see an infrastructure bubble, as computing infrastructure is needed and will be concentrated among those who have sustainable business models and really need AI power.

Andres Andreu (CEO and Global Technology Executive at Constella Intelligence) echoed this skepticism with an uncomfortable observation: many companies are jumping on the hype bandwagon without any real enterprise opportunities. The industry has seen such false starts many times before.

And Tiago Aragona (Creative Technologist & Designer) brought up what is perhaps the most important change in perspective: we need different benchmarks to measure the success of AI. Speed, profit, or "how human it seems" are not enough when talking about ethical, sustainable, and representative AI.

In between, there were other voices that provided the real litmus test: we are still learning, we are still developing, and we should not stop asking questions.

Cybersecurity was repeatedly cited as an eye-opener, especially for people outside the traditional tech bubble. And how differently a single problem can be solved—not as confusion, but as inspiration.

(Video by The AI Summit Series)

The notable absentees

What is NOT said at conferences is often just as revealing as what is said.

OpenAI? Not represented. Anthropic? Conspicuous by its absence.

Instead, on the sponsor list: enterprise software companies. Consulting firms. Cloud providers. Infrastructure providers.

That's no coincidence. The foundation model companies are busy—with themselves, with their race to find the next big model, with the billions they're burning through. But the real action? That's happening elsewhere.

The message is clear: the AI Summit is not about foundation models. It's about their application. It's about how you can integrate AI into your customer support or marry it with your CRM. It's about how you can turn technology into business value.

Incidentally, this is also why we believe that the "Which model is better?" debate is becoming increasingly irrelevant. Sure, the models differ. But for most business applications, the difference between GPT, Gemini, and Claude is minimal compared to the difference between good and bad implementation.

For us, this is the most important insight: The future will not be built by those who have the best models. It will be built by those who implement them best.

That should be encouraging. You don't need a huge budget to benefit from AI. You need focus, good data, clear processes, and the willingness to do the tedious and time-consuming implementation work.

The questions that remain

After two days of summit meetings, numerous presentations and panels, interesting discussions, and time for reflection, the truly important questions remain unanswered. And that is acceptable. It would be somewhat unusual for a conference to provide answers to the major questions of our time…

Who really benefits from AI? The companies that can invest? The employees who retrain? Or just the tech companies that sell tools? We are currently seeing a massive concentration: the big are getting bigger, the fast are getting faster. What happens to those who can't keep up?

And what do we do with the people? If agents take over workflows, what remains for employees? The figures circulating are less abstract than one might think: Goldman Sachs estimates that AI could automate two-thirds of all knowledge work. That's a quarter of all jobs. This is the working hypothesis that corporations are currently using to write their five-year plans. No wonder, when we all have access to a data center with freely scalable geniuses!

The "AI as assistant" narrative also sounds less convincing as agents become more sophisticated. We hear of companies that want to restructure entire departments. Not in five years, but now, preferably yesterday …

How much control are we relinquishing? When AI systems start to orchestrate themselves, who bears responsibility if something goes wrong? Sure, humans remain in the loop—but that's easier said than done the more autonomous the systems become. After all, what happens when the loop is so fast that no human can keep up?

And can we keep up with the pace? Not just technologically, but also in human, organizational, and social terms. Technology is developing exponentially. People are not. At some point, something has to give. It is clear that things like critical, networked, systemic thinking are becoming increasingly important for us humans.

As I said, these are not the questions that a summit can answer. But they are questions that we should all be thinking about.

Our take: What you should take away

So, now it's getting specific. What does all this mean for you?

AI projects need less wow and more process.

The pilot is easy. Scaling is difficult. If you want to introduce AI into your company, invest 20% in the technology and 80% in change management, data quality, and governance. Sounds boring. Works.

Agents are real—but not yet mature in all respects.

Yes, multi-agent systems are already here and will continue to take their place. Yes, they will fundamentally change the way we work. But not next week. Observe developments, experiment within a defined framework, but don't put all your eggs in one basket with a technology that is still in its infancy and comes with many dependencies.

Foundation model providers are interchangeable—your data is not.

Whether you use GPT, Claude, or Gemini, in the end, the winner will be whoever has the better data and processes and knows how to use them. Invest in your infrastructure. That's the real competitive advantage.

Get started. But get started right. Keep going, but do it right.

Not with the biggest, coolest AI project. But with the one that is most likely to work. Manageable, focused, measurable. Success builds momentum. Momentum enables larger projects.

The most pragmatic advice we heard: crawl, walk, run.

First, AI as an assistant for existing employees. Then, automate individual workflows. Then—and only then—larger agent systems. If you try to run before you can walk, you'll fall flat on your face.

And then there's NYC in December...

When you're deep in the AI bubble, every day feels like a FOMO rocket launch: constant acceleration, destination unknown.

But you know what? Sometimes a conference is more than just its presentations. A few blocks away from the conference is the Christmas tree at Rockefeller Center. 23 meters high, 50,000 lights, a 400-kilogram Swarovski star with three million crystals.

Cheesy? Absolutely. But also somehow perfect after a day like this. And even though I've been living here in the Big Apple for over 15 years now, I still enjoy this cheesy, wonderful Christmas magic.

The legendary Rockettes are also going ahead with their program. They are celebrating their 100th anniversary this year. A whole century of synchronized high kicks performed with perfect precision – as always since 1925.

Some things never change. And that's reassuring.

In Bryant Park, you can go ice skating and drink mulled wine in heated igloos, while Christmas market stalls around you exude European flair.

Okay, they'll never be able to compete with German Christmas markets. But still...

The agents are coming. The questions remain. And in NYC, the Christmas tree is lit up.

Perhaps that is the real lesson: technology is constantly changing. But people? They still want to see lights in winter and celebrate together. They want to warm themselves, marvel, and forget for a moment that the world is spinning faster and faster.

And it is precisely in this gap that the magic happens.

These are exciting times, dear AINAUTS—thank you for joining us on this journey!

See you soon, Reto & Fabian from the AInauten

Your feedback means the world to us. We read every comment and message—just hit reply and tell us what you think!