Korva is a scroll of the internet’s best ideas.
Customer closeness is existential
For years, software companies leaned on a business model so good that they lost touch with how customers used the product, how those customers were changing their organizations, and what other technology touched the application.
Now every customer is weighing whether to rebuild its applications with a small or large language model and a consulting company, or to stay with the incumbent vendor and weave agents into the application. A vendor that talks to its customers multiple times a week, and is trying to solve their problems rather than raise prices, is in a strong position.
Incumbents also have a cost opening: inference prices are falling in a price war, and open-source models add pressure. That lets a company point each query at the right model, lowering its cost of goods sold, while keeping five nines execution on a system that touches every employee. The companies have earned the right to do that, and it is no birthright.
How does routing queries to models cut costs?
A routing layer sends easy requests to a small, cheap model and escalates hard ones to a large model. Public price lists from providers such as OpenAI and Anthropic show that models from the same provider can differ by an order of magnitude in price per token. Because inference is part of a software company's cost of goods sold, the savings flow directly into gross margin. The router has to decide which queries are easy before the model answers.
How do open-source models pressure prices?
Open-weight models such as Meta's Llama series can be downloaded and hosted by any cloud or specialist provider. Many hosts then offer the same model, so they compete on price per token. Proprietary labs face a ceiling on what they can charge for comparable capability. Inference prices have fallen steeply since 2023.
Why do subscription businesses lose touch with customers?
Multi-year contracts with high renewal rates let a vendor go years without a hard conversation with a customer. Revenue arrives on a schedule, so a decline in how much customers actually use the product shows up late. Sales and renewal teams manage the account relationship, while product teams often rely on usage dashboards instead of direct contact. That gap grows with every renewal that happens on autopilot.