AI Will Automate 50% of Digital Work in Years | Brendan Foody artwork

AI Will Automate 50% of Digital Work in Years | Brendan Foody

MTS

November 11, 2026

Mercor co-founder and co-CEO Brendan Foody discusses the company’s $750 million Series D at a $20 billion valuation and why post-training evaluations have become the primary bottleneck for frontier AI development.

0:00

Brendan Foody

I more broadly and probably AGI long, but ASI short, and that I think that will automate over 50% of the tasks, definitely in the digital world, in the next few years. But I think there's this long tail of tasks that require a lot of tasks at knowledge or are intrinsically human, where people wanna have human to human conversations or everything in the physical world, where it just feels like there's going to be a much longer road of how we codify all of that human expertise into the models.

0:36

SPEAKER_2

All right, we are back. We are live with Brendan Foody, who is co-founder and co-CEO of Mercor, which just announced a $750 million Series D at a $20 billion valuation. Brendan, welcome and congrats.

0:50

Brendan Foody

Yeah, excited to be back. Thanks for having me on and excited to share more about the journey on all of the progress of the company.

0:59

SPEAKER_2

Yeah. So tell us more about what you plan to do with all this money that you're raising.

1:03

Brendan Foody

So the highest level trend also to set the stage on the business is that we scaled extremely quickly over the last two years in becoming the largest data provider to all of the frontier labs and top application layer companies. But we're seeing this enormous trend where enterprises are also extremely interested in now building out evaluations and post-training their own models. So we're applying all of the technology that we built for the AI native companies in San Francisco to the broader set of enterprises in the economy and raising this capital to accelerate our investments in that.

1:40

SPEAKER_3

Yeah, I'm interested in like the trend that you're seeing with regards to custom like evaluations being developed in house by companies. What do you see there?

1:49

Brendan Foody

Yeah, so we're seeing, of course, them developed both as private and public evals. So a couple of examples on the public eval side is we build Frontier Code with Cognition. We built TauBench with Sierra Lab with Harvey, Apex Accounting with Ramp. But we're also seeing that all of the companies deploying agents need to have evaluations to understand which model they should be using. Is that model working effectively in production and how do they optimize the cost of it? And so for pretty much every major agent deployment that has, call it, over $10 million a year in spend, companies are mobilizing to build evaluations to answer those questions.

2:32

SPEAKER_2

I'm curious, you know, we last talked a few months ago, since then, there's been a lot of discourse around RSI. There's been a lot of discourse around people radically shortening their timelines to AI because they think there will be fewer bottlenecks to recursive self-improvement, including data bottlenecks.

So have you changed the way you think about this at all in the last few months?

2:54

Brendan Foody

It definitely feels like a crazy time as we were talking about where there's so many benchmarks, like Apex Agents we launched in February of this year when Frontier Models were at 23%, and now they're at roughly 80%.

And I think that's sort of the general theme across the board on anything that the RL is becoming so powerful that once we have an eval for something, the models can saturate it. And so a lot of ways, the barrier to applying AI to automate everything in the economy is, how do we just have some evaluation or some way to measure it so that we can improve model performance against it? But I would say that I more broadly and probably AGI long, but ASI short, and that I think that will automate over 50% of the tasks definitely in the digital world in the next few years. But I think there's this long tail of tasks that require a lot of tacit knowledge, or are intrinsically human, where people want to have human to human conversations, or everything in the physical world, where it just feels like there's going to be a much longer road of how we codify all of that human expertise into the models.

4:06

SPEAKER_2

Totally. Yeah, I mean, similarly, there seems to be a big qualitative disconnect between like, the models are extremely smart, obviously. They can saturate pretty much any eval that you can throw at them at this point. But they also make all kinds of weird mistakes, little errors. Like I was just experiencing this with this like project I've been building with Fable 5.2, which is, you know, the, you know, mythical, likely the mythical internal entropic Model 3 And it was making all of these silly errors that it should not be.

Read the other 10 minutes of this episode here.

$15 gets you this episode and 99 more. Credits never expire. After checkout, one link brings you back here to read it.

Not what you expected? Full refund within 14 days.

Thousands of transcripts fetched by people building searchable podcast archives

Fetch transcripts with code

The same credits work from a script or an AI agent. The demo key returns a sample episode in full, no card needed:

request
curl -H "x-api-key: pt_demo" \
  https://spoken.md/transcripts/1000651996090

Markdown with the speakers named, for your notes, your knowledge base, or anything that makes HTTP calls.

From $0.10 per transcript. No subscription. Credits never expire. Prices exclude VAT, added at checkout for EU customers. Not what you expected? Email us within 14 days with 20 or fewer credits used and we refund the pack in full.

Using your own key:

request
curl -H "x-api-key: YOUR_KEY" \
  https://spoken.md/transcripts/1000801132499