Produced for

Chain of Thought: AMD's AI Stack and Driving Developer Curiosity with Sharon Zhou

AMD's new VP of AI on the happy path for developers, open source as a catch-up strategy, and self-improving models. A Chain of Thought episode recorded live at AMD Advancing AI 2025, produced by Free Agent Media for Galileo and hosted by Conor Bronsdon.

Chain of Thought: AMD's AI Stack and Driving Developer Curiosity with Sharon Zhou: Sharon Zhou, Conor Bronsdon at AMD Advancing AI 2025, San Jose
ClientGalileo
FormatPodcast episode · Video podcast
Runtime20:50
Published2025-06-18
Live atGalileo on YouTube ↗
LocationAMD Advancing AI 2025, San Jose
THE STORY

What the media captures.

Sharon Zhou walked out during the Advancing AI 2025 keynote as AMD's new Vice President of AI, days after selling Lamini, the company she co-founded. Conor Bronsdon caught her the same afternoon for Chain of Thought, Galileo's podcast for developers and leaders building with AI. Free Agent Media recorded and produced the episode on site in San Jose.

Her brief at AMD is research and teaching, and the teaching part is the argument of the episode. Developers now prompt with a single question mark and expect something to work. So AMD needs a happy path: three steps to a first success on AMD hardware, an aha moment before anyone loses patience. That means listening first, through talks, courses, and the ROCm repositories, and then rolling out the yellow brick road rather than a wall of documentation.

Zhou frames AMD's differentiation two ways. Open source lets a challenger accelerate by letting the community catch it up. And AMD owns both GPU and CPU, which matters now that agents run model calls on one and their MCP tools on the other. She sorts the people she has to win into three personas, AI developers, AI researchers, and AI leaders, and describes the courses with Andrew Ng's DeepLearning.AI that already put tens of thousands of developers on AMD GPUs through Lamini.

The back half turns to what could go wrong and what comes next. Open invites everyone to contribute, which can produce a monstrosity with no opinionated direction, so the happy path has to be defended. And her prediction for the next two years is self-improving models that edit their own training data, steered by what she calls vibe-based feedback: natural-language correction, closer to how people teach each other than to a config file.

TC 00:02:00:00 · YOUR TURN

Interested in something similar? Reach out.

hello@freeagentmedia.ai
CHAPTERS

Six key chapters.

  • 00:00Live from AMD Advancing AI 2025
  • 01:31The happy path: getting developers to an aha moment fast
  • 05:29Pre-believers, and why AMD's GPU-plus-CPU fabric matters for agents
  • 08:44Three audiences: developers, researchers, AI leaders
  • 11:35Why MCP had to be open, and who AMD is building for
  • 15:15The risk of open, and self-improving models
ON THE RECORD

Who says what.

  • Sharon ZhouVice President of AI, AMD
  • Conor BronsdonHead of Developer Awareness and host of Chain of Thought, Galileo

“I think my Cursor prompt is literally this: the minimum number of tokens is a question mark. So being able to roll out the red carpet, or show the yellow brick road to follow, is really critical.”

“These agents are LLM calls, they run on GPU, but the tools they use through MCP, many of them are actually running on CPU. AMD is in a really interesting position to balance that effectively.”

“That's tens of thousands of developers already hitting AMD GPUs over the past year. I'd like to see a new model architecture be invented on AMD.”

“It's really important to highlight MCP as an open protocol, and therefore it can be a standard. If there's only a closed option, you don't feel like you can see into it at all.”

“The challenge of open is that you're inviting the whole community to contribute, and you can create almost a monstrosity of way too many features tacked on.”

“Not just vibe coding, but can we vibe tune, vibe train these models? Vibe-based feedback, so we can give our natural language direction, much closer to how we teach each other.”

SCOPE

What we delivered.

Podcast production

Free Agent Media produced 1 video for Galileo at AMD Advancing AI 2025, June 12, 2025. Formats: Podcast episode.

CREDITS

On the record.

Production company
Free Agent Media
Client
Galileo
Format
Podcast episode · Video podcast
Featuring
Sharon Zhou (AMD), Conor Bronsdon (Galileo)
Published
Galileo on YouTube, 2025-06-18
Runtime
20:50
QUICK FACTS

About this project.

Who produced this Chain of Thought episode?

Free Agent Media, a video production company for technology companies based in San Francisco, Seattle, and New York, recorded and produced the episode for Galileo, which publishes Chain of Thought. Galileo released it on YouTube on June 18, 2025.

Who is in it?

Sharon Zhou, Vice President of AI at AMD and former co-founder and CEO of Lamini, interviewed by Conor Bronsdon, Head of Developer Awareness at Galileo and host of Chain of Thought.

Where was it recorded?

On site at AMD Advancing AI 2025 in San Jose on June 12, 2025, the event where AMD announced the Instinct MI350 series and ROCm 7 and where Sharon Zhou appeared during the keynote as AMD's new VP of AI.

What does Sharon Zhou argue?

That AMD wins developers with a happy path to a fast first success, by listening to pre-believers before selling to buyers, and by leaning on open source to accelerate. She also argues MCP matters because it is open, and predicts self-improving models steered by natural-language, vibe-based feedback.

Why is Free Agent Media on this page?

Free Agent Media produces Chain of Thought episodes for Galileo, including remote and on-site recordings at industry events like AMD Advancing AI 2025. This page is the production record for this episode.

Is this the full episode?

No. This video is the Sharon Zhou segment. The full Chain of Thought episode #28 also features Anush Elangovan, AMD's VP of AI Software, and is available on chainofthought.show, Spotify, and Apple Podcasts.

WHERE IT LIVES

Proof it shipped.

MORE WORK

Related pieces.