Amber CEO: A Confession and Rebirth of a "Wayward" Public Company Boss

Author: Michael Wu, Founder of Amber Group

From an outsider's perspective, Amber's past few years might seem like a comeback story: nearly collapsing when FTX went bust, surviving; enduring the bear market of 2023, and finally welcoming the bull market; completing the IPO and ringing the bell on Nasdaq. The plot is perfect.

But only I know that the years 2024 and 2025 were the most agonizing two years of my nine years of entrepreneurship.

The agony was not due to business or market conditions, but because starting in 2024, as the CEO of Amber, I spent 70% of my time and almost 100% of my interest on something that seemed "off track" internally at the time: building AI agents—back then, the term "agents" hadn't even been coined.

In 2024, as the CEO of a company with several hundred employees, I spent most of my time immersed in a very small internal team (this team later faced many challenges and became the prototype of our current AI Enablers). I led them in several AI agent projects (again, they weren't even called agents at that time, and no one knew what to call them). First, we created an amber bot that acted as our sales and customer service representative. At that time, the concept of harness didn't even exist; everything was done manually with loops and tool calls. The results were rough, but I enjoyed it, and the more I worked on it, the more convinced I became: this is the future I envisioned when I founded Amber AI in 2017: within 5 to 10 years, most economic activities would be driven by AI agents.

That year, even while still in the crypto bull market, I launched the http://amber.ac accelerator and boldly proclaimed "crypto for AI" during the opening speech ("Crypto applications are not for humans; the real users are the future AI agents!"). In the audience, very few took it seriously.

In 2025, we undertook an even more forward-thinking project: aiac. We tried many things, including something very cool but too advanced for the models and interactions of that time—our first incubated agent, MIA (who is still the eldest sister of our AMBR agent family today), had a capability that no one had done before: she could observe your actions in real-time—where the mouse moved, what you typed and deleted—while providing suggestions and interactions instantly. To be honest, this project failed again. But it gave birth to MIA. The earliest MIA, as a marketing agent for a crypto company, could only manage a few of our social media accounts, producing a bunch of occasionally great but mostly terrible AI meme images (limited by the models of that time, they were genuinely bad).

Also in 2025, I felt very lonely in the company. In a crypto finance company that had been around for nearly ten years, some wanted to trade, some wanted to invest, and some wanted to leverage the IPO platform for group expansion. All of these were valid and good businesses, but none of them were what I wanted to do. I hesitated, felt soft-hearted, and even felt guilty towards everyone. But as an entrepreneur, I couldn't betray my own heart. During the most intense disagreements, the internal atmosphere was almost torn apart.

At the end of 2025, I made the hardest but also the most resolute decision in nine years: to consolidate all resources and focus on the one thing I believed in the most: Build AI Agents. Many early partners who had been alongside me left; there were no right or wrongs, only gratitude.

In early 2026, while reorganizing the team and returning to management, I also took almost all my spare time (there were days when I even went without sleep for several days; I was already 37!) to do something even more willful: from February to March, in two months, I created 11 specialized agents for myself, researching and manually crafting a bunch of harnesses and various tools, entrusting my work, life, investments, trading, socializing, health, and self-improvement entirely to my personal agents. Project codename: extend me.

This is not a toy; I was using vibe coding to repeatedly verify one thing: can AI agents in various vertical scenarios create real, specific value for a real, concrete person?

The answer is yes! This process also strengthened another judgment: for an AI agent company to succeed, the founder must have a deep understanding of what an AI agent is and how to build AI agents.

Then comes the story I most want to tell.

In March of this year, I brought the "failed" team from aiac, along with the "failed" MIA, into the B2B advertising business that we had acquired during our IPO but had not had the energy to manage: iClick and CMRS. In just a few months, to the surprise of almost everyone (except me): this traditional business, which had nothing to do with Amber's main operations, was revitalized—Fortune 500 clients were more satisfied, internal operations became largely automated, many workflows were completed by agents, and even began to self-iterate. Last month, MIA, which grew from the failure of aiac, became our first vertical AI agent product directly facing users.

A friend who is much more successful and older than me joked, "Michael, the problem with Amber over the years is that you, the founder, are too smart and energetic, wanting to do everything: like being a university professor during the day and selling skewers at the university gate at night."

Thinking about it, it's indeed true. After nine years of entrepreneurship, Amber has not yet reached the height I desire, and for me, the biggest lesson is two words: focus.

So for the next ten years, I will start anew, focusing on just one thing: Build AI Agents.

When I founded this company in September 2017, it was called Amber AI. Over the past nine years, we have engaged in trading, served as a crypto finance gateway, acted as an investor, and rang the bell. Now, I am finally going back to realize my most genuine dream.

On September 1, Amber will debut under a new name (which you might find familiar) as a new company that will focus solely on one thing for the next ten years—Build AI Agents. Our flagship AI agent, along with a brand new version of MIA that has quietly gone live for a month, will also meet everyone on the same day.

Looking back at my first nine years of entrepreneurship, Amber has experienced successes and failures alongside the rise of crypto finance from zero to one.

For the next ten years, I will focus on just one thing: build AI Agents that get you there. Let AI agents create real, specific value for you.

See you in Hong Kong on September 1.