Author:Michael Wu, Founder of Amber Group

To outsiders, Amber's past few years may seem like a comeback story: nearly collapsing during the FTX crisis, surviving; enduring a bear market in 2023, and waiting for the bull market; completing an IPO and ringing the bell at Nasdaq. The narrative is perfect.

But only I know that the years 2024 and 2025 were the most agonizing two years of my nine-year entrepreneurial journey.

The agony was not due to the business or the market, but because starting in 2024, as the CEO of Amber, 70% of my time and nearly 100% of my interests were devoted to something that seemed “unproductive” internally at that time: building AI agents—back then, there wasn't even a term for agents.

In 2024, as the CEO of a company with several hundred employees at Amber, I spent most of my time in a very small internal team (this team later faced many challenges but became the prototype of our current AI Enablers). I led them through several AI agent projects (again, they weren't even called agents back then, 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; all loops and tool calls were done manually. The product was rough, but I enjoyed it, and the more we did, the more convinced I became: this was 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.

It was also that year, in the midst of a bullish crypto market, that I initiated the http://amber.ac accelerator, and loudly proclaimed in my opening speech, “crypto for AI” (“crypto applications are not for humans, the true users are the future AI agents!”). Few people in the audience took it seriously.

In 2025, we undertook a more advanced project: aiac. We experimented with 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 attempted before: she could observe your actions in real time—where the mouse moved, what you typed and deleted—while providing suggestions and interactions. Frankly, 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 mix of occasionally high-quality and mostly poor AI meme images (limited by the models of that time, they were pretty sincerely bad).

It was also 2025 when I felt quite lonely in the company. In a crypto finance firm that had been operating for nearly ten years, some wanted to do trading, some wanted to invest, and others hoped to build a group using the public listing platform. All these were valid and good businesses, but none were what I wanted to pursue. I hesitated, felt soft-hearted, even felt guilty towards others. But as an entrepreneur, I couldn’t betray my inner self. During the most intense disagreements, the internal atmosphere was almost torn apart.

By the end of 2025, I made the hardest but most resolute decision in nine years: to consolidate all resources and do what I believed in the most: Build AI Agents. Many early partners who stood by my side left; there was no right or wrong, only gratitude.

In early 2026, I focused on reorganizing the team and returning to management, while paradoxically sparing almost all my fragmented time (there were days I even went sleepless, and I was 37!) to do something even more daring: from February to March, in two months, I created 11 specialized agents for myself, researched and crafted a bunch of harnesses and various tools, managing my work, life, investments, trading, social interactions, health, and self-improvement, all with the help of my personal agents. Project code name: extend me.

This wasn’t a toy; I was personally verifying through repeated vibe coding whether AI agents in various vertical scenarios could create real, specific value for a real, specific person.

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

Then comes the story I most want to share.

In March of this year, I reinstated the “failed” aiac team, along with the “failed” MIA, into the B2B advertising business of iClick and CMRS, which we had acquired during our IPO but lacked the energy to manage. In just a few months, contrary to the expectations of almost everyone (except me): this traditional business, which has nothing to do with Amber's main operations, rejuvenated—Fortune 500 clients were more satisfied, internal operations became largely automated, and numerous workflows were completed by agents, even starting to self-iterate. Last month, MIA, which had grown out of the failure of aiac, became our first directly user-facing vertical AI agent product.

A friend of mine, who is much more successful in business and older than I am, joked with me: “Michael, the problem with Amber over the years is that you, as the founder, are too smart and energetic, wanting to do everything: it’s like being a university professor by day and selling skewers outside the university at night.”

Thinking about it, it is indeed true. After nine years of entrepreneurship, Amber has not reached the height I wanted; for me, the biggest lesson is just two words: focus.

So for the next ten years, I will set out anew, focusing on doing one thing: Build AI Agents.

When I founded this company in September 2017, it was called Amber AI. Over the nine years, we have done trading, operated as a crypto finance gateway, acted as investors, and rang the bell. Now, I am finally going back to realize my truest dream.

On September 1, Amber will reappear under a (which you might find familiar) new name, as a new company that solely focuses on one thing over the next ten years—Build AI Agents. Our flagship AI agent, alongside a completely new version of MIA that has quietly launched for a month, will also be presented on the same day.

Looking back on my first nine years of entrepreneurship, Amber has experienced both success and failure as crypto finance evolved from zero to one.

In the next ten years, I will focus on doing 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.