The Twenty Minute VCMercor Head of Product on Revenue Concentration from Frontier Labs
At a glance
WHAT IT’S REALLY ABOUT
Mercor CPO on data moat, enterprise ROI, and revenue concentration
- Open-source model gains raise the baseline but don’t eliminate demand for frontier data, because customers still pay for datasets that close specific capability gaps at the edge of performance.
- The popular claim that “90% of enterprise workflows are solved” misses latent demand—especially long-horizon agent workflows—and overuses a binary sufficiency framing where many tasks have uncapped, continuous improvement value.
- Mercor argues there isn’t an enterprise AI ROI crisis yet; most companies are still experimenting while token prices and model capabilities shift, but spend discipline is emerging and must be tied to outcomes.
- AI increases product-management difficulty by accelerating engineering output while making focus, simplification, and experiment quality the true bottlenecks—driving a higher PM-to-engineer ratio over time.
- Mercor’s high revenue concentration in frontier labs is addressed by moving downmarket via self-serve tooling and “AI project managers” that make operationally intense human-data projects easier for enterprises.
IDEAS WORTH REMEMBERING
5 ideasOpen source raises the floor, but frontier customers still buy the edge.
Mercor sees open models as eliminating demand only for capabilities that are already commoditized; the most valuable datasets remain those that push specific frontier gaps for labs with unique objectives.
The “90% of workflows” narrative undercounts what enterprises will attempt next.
Osvald argues many calculations ignore latent demand—especially long-horizon agentic automation (e.g., procurement agents running for weeks)—which expands the addressable market for eval and training data.
Binary ‘can it do it’ thinking breaks down in uncapped-reward work.
For tasks like legal reasoning, medical advice, and cybersecurity, performance can always improve and incentives remain strong, so adoption and spend should be modeled as continuous optimization rather than a completion percentage.
Enterprise AI ROI is not ‘broken’—it’s still in a volatile exploration phase.
Because performance and token costs are changing quickly, companies tolerate ambiguity while testing; Mercor expects more rigorous outcome-linked accounting and differentiated spend profiles by team function over time.
Token spend strategy should match whether usage drives growth or unit economics.
High spend can be rational for compounding productivity (e.g., coding agents) even if it’s not classic COGS, but customer-support agents with token costs exceeding served revenue signal a broken model.
WORDS WORTH SAVING
5 quotesI wouldn't say that open source model improvements cannibalize our core business because data is most valuable on the frontier of model performance.
— Osvald Nitski
I don't think there's an ROI problem right now. I think we're in a period of exploration and experimentation where there's more, uh, tolerance, more patience to get that ROI calculation right now.
— Osvald Nitski
I think this paradigm makes the job of product management a lot harder because we're trying not to build 10X more product surface area. It makes things incredibly chaotic.
— Osvald Nitski
Don't delegate your decision-making, like, your actual job to a model, because you're gonna, you're gonna lose that ability, and then you're gonna get psychosis.
— Osvald Nitski
We end every week with so much more money in the bank. Like, the business is, is very healthy, and we can't, we can't spend money fast enough. So what people, you know, wanna, uh, call it is, you know, up to them, but, like, the, the cash flow is insane.
— Osvald Nitski
High quality AI-generated summary created from speaker-labeled transcript.
