AMD Gave a Developer Award to Someone Who Can't Code

Today I went to AMD's developer conference in Shanghai.

The entrance alone was a shock. The line to get in was long, and the hall was already packed before anything started. They expected just over 1,000 people; more than 2,000 showed up. AMD said it was their biggest recent event.

Lisa Su showed up too. She'd been in Beijing the day before meeting Vice Premier He Lifeng to talk chip cooperation. I'd never seen a chip company pull a crowd like this for a developer conference.

AMD Gave an Award to Someone Who Can't Code

That morning, an AMD senior VP got on stage to hand out two developer awards. When they introduced one winner, the host said:

"He didn't actually know how to code before."

He'd used an AI agent to rewrite an entire system in Rust and optimize performance. AMD figured that was worth an award.

Sitting there, the whole thing felt surreal. A chip company worth hundreds of billions, at a 2,000-person developer conference, handed one of two awards to someone who doesn't code.

I bet next year the award will be even harder to judge. AI-native people like him will only become more common.

The Boundary Between Developers and Users Is Vanishing

Before, if you used a product, you just used it. You couldn't really help build it. Even in open source, you had to code before you could contribute.

Not anymore. Coding agents are getting stronger, and regular users can now tweak, optimize, and push changes back while using a product. The same person is both user and builder.

I wrote before about the split between Builders and Promoters. That was about passion diverging. This is the flip side: the roles of user and contributor now overlap, often in the same person. Users are also investing their tokens across different products, and the ones that earn that investment keep evolving.

Product logic has shifted. You used to focus on making the experience great. Now you also need to make it easy for users to become contributors.

AMD's big Strix Halo push is interesting. The AI Max+ 395 chip can allocate up to 96 GB of unified memory to its integrated GPU for running local models, and my inference engine can run on it too. Domestically, prices have been climbing and it's been out of stock. I have several R&D test machines for performance tuning, and they're also my entry point into the ROCm ecosystem.

AMD is pushing this machine to lower the developer barrier another notch. More developers means stickier stacks.

Industry Winds Did a 180 in Six Months

I attended a similar conference around mid last year. The vibe was completely different.

Back then, people in the compute business were stressed. Early last year, DeepSeek raised expectations for models, and everyone was wondering whether the wave would last. How to move product, how to clear inventory, whether the business could survive. Everyone was scrambling for solutions and partners.

This year, the table talk completely changed. The first thing anyone says is, "Can you get me more supply?" or "I'll take everything you've got."

It's completely flipped. Supply is tight, and whoever holds quality inventory is making money. The shift from demand anxiety to supply anxiety took just six months.

This Isn't Another Bubble

Plenty of people say: "Here we go again. Next metaverse."

This time it really is different. I lived through the metaverse and blockchain cycles too. The difference this time is in the data, specifically paid demand from real users.

Lisa Su said on stage that roughly 1 billion people are already using AI worldwide, and by 2030 that number will hit 5 billion daily active users. ChatGPT came out at the end of 2022, so it's been less than three years. The internet took over 20 years to reach that scale; the PC era took even longer. This is a diffusion speed never seen before in history.

The money is keeping up too. Anthropic's Q1 grew 80x year-over-year. That's annualized revenue, not API calls. Dario himself said they weren't ready to catch a wave this big. Claude Code hit a $1 billion annualized run rate within six months of launch, and by April this year the company's overall ARR surged to $30 billion.

This is nothing like a few years ago, when everyone was in a price war, handing out free tokens, and chasing call volume. Supply can't keep up with paid demand.

"X Is Dead" Is the Cheapest Narrative

A friend recently asked me: "Is Openclaw dead?" "How's Claude Code doing?" "I heard Codex is going to win."

I think that's just inertia.

Last December, every top academic conference and product circle was talking about Gemini. Back then everyone thought Google had it in the bag. A few months later, almost nobody mentioned Gemini. Then it was Cursor, then Claude Code. Pretty soon it'll be Codex. At the top table, players keep rotating.

But the underlying trend runs one way. It hasn't reversed. Paid demand is rising, call volume is rising.

Real information is expensive. You have to use the tools yourself, show up on-site, and talk to people inside. So the audience for that is naturally small. Narratives like "it's dead," "it's a bubble," or "just another cycle" are the cheapest to spin up. They validate sitting on the sidelines and feed the need to believe that not engaging was the right call. They spread the easiest.

Not that nobody believes it. Most people just want to.

Go See for Yourself

Lately when I meet friends, I do one thing: tell them to bring their laptop, and I help them install Claude Code or Codex and get the model connected.

Once you get past that hurdle, you can hand off 95% of computer work. I built my own website from scratch without lifting a finger. Frontend, DNS, SEO, all done by agents. The barrier is small, but once you're past it, the world looks completely different.

If that's still too much, just find a conference this year where people are actually doing this and go. There were quite a few workshops at the event where you brought a laptop and worked hands-on. Only when you sit there do you realize how far AI has already come.