Amazon

Amazon can't keep its AI people. And that might matter more than any model release this year.

David Luan, the guy running Amazon's San Francisco AI lab, announced Tuesday that he's leaving. "I'll be leaving Amazon at the end of this week to cook up something new," he wrote on LinkedIn. He added that with "AGI so close," he wanted to spend all his time teaching AI systems "brand new capabilities."

Less than two years. That's how long Luan lasted at Amazon. He came over in 2024 when Amazon essentially acquired his startup Adept by licensing its technology and absorbing the team. His job was to lead Amazon's push toward AGI from the company's San Francisco lab. Now he's gone, and whatever he's cooking next isn't Amazon's problem anymore.

The Adept Backstory

Luan's arrival at Amazon wasn't a normal hire. Adept was building AI agent technology that could interact with software on your behalf. The startup raised $415 million at a $1 billion valuation before Amazon came along with what amounted to an acqui-hire dressed up as a licensing deal. The structure let Amazon avoid the regulatory scrutiny that comes with a full acquisition, which was already becoming a hot topic for Big Tech AI deals.

At Amazon, Luan worked on the Nova AI models, including Nova Act, an AI agent that could complete web searches, make purchases, and answer questions. Nova Act launched as a research preview last March and was integrated into Alexa Plus.

The problem? Alexa Plus hasn't exactly taken the world by storm. Early users reported that the AI version of Alexa felt like a worse version of the original. Amazon's own employees have reportedly nicknamed the company's AI products "Amazon Basics," which is either funny or devastating depending on your relationship to the stock price.

Why People Leave

Luan's departure fits a pattern that should worry Amazon's leadership. When top AI researchers leave big companies to "cook up something new," they're telling you something. They believe they can move faster and do more interesting work outside the building.

Amazon has all the resources in the world. It has AWS, the world's largest cloud infrastructure. It has mountains of consumer data. It has billions in capital. What it doesn't have is a culture that attracts and retains the best AI talent.

Compare Amazon to Google, where DeepMind has operated semi-autonomously for years and attracted world-class researchers who publish foundational papers. Or to OpenAI, which has become a magnet for ambitious AI engineers despite its internal drama. Even Anthropic, which is a fraction of Amazon's size, seems to hold onto its talent better.

The issue isn't money. Amazon can pay whatever it wants. The issue is that AI researchers want to build things that matter, and Amazon keeps directing AI efforts toward Alexa and product recommendations. If you're someone who believes AGI is around the corner and you want to be the one building it, working on a voice assistant that can't reliably set a timer isn't compelling.

What Amazon Loses

Luan wasn't just a researcher. He was a bridge between the startup world and Amazon's corporate machinery. He understood how to build AI products from zero, having done it at Adept. That experience is hard to replace.

More importantly, his departure signals to other AI researchers at Amazon that the grass might be greener elsewhere. Talent retention in AI is a chain reaction. One high-profile departure makes others question whether they should stay. If three or four more senior researchers follow Luan out the door in the next few months, Amazon's AGI lab becomes a revolving door.

The timing is also awkward given Amazon's $50 billion investment in OpenAI announced this same week. If you're an AI researcher at Amazon, watching your employer write a $50 billion check to a competitor while your lab leader walks out, what message does that send? It says Amazon's own AI efforts aren't the bet the company is making.

Amazon's AI Identity Crisis

This is the core of the problem. Amazon doesn't know what it wants to be in AI. Is it a model builder? An infrastructure provider? A product company? An investor?

Right now it's trying to be all four, and that diffusion of focus is showing up in the results. Nova models haven't gained traction with developers. Alexa Plus is struggling with adoption. AWS Bedrock is growing, but mostly because it lets customers run other companies' models. And now Amazon is investing billions in OpenAI while simultaneously trying to build competing technology.

Google solved this problem by letting DeepMind be the research arm while Google Cloud handles enterprise AI and the consumer products team integrates AI into search, maps, and other apps. Microsoft solved it by partnering deeply with OpenAI while building Copilot for its enterprise products. Both companies have a coherent AI strategy.

Amazon's strategy feels like "all of the above, please." That approach works for an e-commerce platform with unlimited shelf space. It doesn't work when you're trying to attract world-class researchers who want clear mission, resources, and autonomy.

What Happens Next

Luan said he wants to spend all his time on teaching AI systems new capabilities. That sounds like another AI startup. If he raises a round in the next few months, watch who follows him. The talent flow will tell you more about Amazon's AI prospects than any earnings call.

For Amazon, the replacement search for the AGI lab lead matters enormously. Get it right and the lab stabilizes. Get it wrong and the departures accelerate.

The $50 billion OpenAI investment is a hedge, but it's also an admission. Amazon isn't sure it can build the best AI models in-house. So it's paying someone else to do it while keeping its own efforts alive just in case. That's a rational strategy, but it's not the kind of strategy that inspires people like David Luan to stick around.