Seeam Shahid Noor graduated from Harvard University and, a few years into his career, moved to Silicon Valley to join Scale AI, which the episode describes as the company that builds training data for nearly every major AI lab in the world. He was inside Scale AI when Meta bought 49% of the company and took its CEO into Meta Superintelligence Labs, and he found out the same day the rest of the world did. At the time of this episode (July 2026), he was an Investing Partner on a16z's Speedrun program, which writes cheques of up to 1 million dollars to early stage founders.
In this episode Ahmed Fahad talks with Seeam about what that week inside Scale AI felt like and how AI training data went from labelling traffic lights to stumping Stanford professors. They discuss why AI labs are paying doctors 150 to 200 dollars an hour, what data robotics companies need, and why Bangladesh's garment factories could be a real advantage. Seeam explains why he believes 99% of AI startups will not make it, why only 2 to 3 percent of the world has adopted AI, where cost is holding Bangladesh back, and what happens to jobs.
What the conversation covers
- Moving to the West Coast for Scale AI, and the people there
- How Meta acquired 49% of Scale AI, and finding out with the public
- From self driving data to RLHF, and how training data tasks have evolved
- The bull and bear case on data labelling
- What AI labs are actually paying for
- The data robotics companies need, and Bangladesh's unfair advantage
- Joining a16z Speedrun
- Why 99% of AI startups will fail, and what makes enterprise AI hard
- AI and jobs, adoption, and whether AI is too expensive for Bangladesh
- How Seeam uses AI day to day
The conversation is in Bangla.