Korva is a scroll of the internet’s best ideas.
Open models raise the floor
Where data is valuable
Each customer has its own goals and buys eval and training sets to fill gaps in what current models can do. When an open model already handles something, nobody buys data for it. As long as customers still have new capabilities they want to get better at, the business keeps growing.
The 90 percent claim
The claim that open models can handle 90 percent of enterprise workflows leaves a shrinking 10 percent at the frontier. That calculation may rest on existing demand, or on whatever comes to mind when today's model users think about what models could do. On Mercor's Apex benchmarks, top models score around 50 percent on long-horizon workflows.
Latent demand
There is also a whole category of latent demand in things people are not even trying with models yet, most commonly long-horizon tasks. Picture a procurement agent set up to automate a procurement team for months on end, with a check-in about once a week. Calculations that say enterprise workflows are being handled leave those tasks out, and the market for data to support them is growing.
Why do easy benchmarks stop mattering?
A benchmark stops separating models once the top ones all score near the ceiling. MMLU, a broad knowledge test introduced in 2020, saw leading models climb above 85 percent within a few years. Researchers responded by writing harder tests, such as Humanity's Last Exam. Data buyers follow the same logic, paying for examples at the point where models still fail.
Why do long tasks trip up agents?
Errors compound across steps. A step that succeeds 99 percent of the time yields a success rate of only about 37 percent across a chain of 100 steps. Long tasks also expose the model to situations that drift from anything seen in training, and a single wrong turn early can ruin every later step.
Why can cheaper capability increase total demand?
In 1865, economist William Stanley Jevons observed that more efficient steam engines led Britain to burn more coal, not less. Cheaper use opened applications that had not been worth attempting before. A 1996 study of lighting by economist William Nordhaus found that the price of light fell enormously over centuries while the amount people used rose steeply. Economists now call the pattern the Jevons paradox.
More from 20VC with Harry Stebbings
