Why We Started Costfunc
The gap between AI research and real-world products, and our approach to bridging it.
The Gap
There's a widening gap between AI research and products people actually use. Papers get published, models get released, benchmarks get beaten. But the lived experience of most software remains unchanged.
We've seen this firsthand. Before starting Costfunc, we worked at companies where "AI" meant either:
1. A feature checkbox that barely improved user experience
2. An R&D project that never shipped
3. A marketing term applied to simple heuristics
The problem isn't the technology. The problem is the approach.
Our Philosophy
We believe the best AI products come from teams that:
Build first, advise second
Most AI consultancies have a fundamental conflict: they make money from ambiguity and complexity. We make money when products ship and users benefit.
That's why we build our own products. Nix isn't a side project—it's proof of what we can do. When we work with partners, they get the same engineering DNA that powers our own products.
Research that ships
We publish our research, but only research that has shipped in production. No theoretical explorations. No benchmark-chasing. Every insight comes from real users, real products, real problems.
Quality over quantity
We take on 2-3 partners per year. That's not artificial scarcity—it's the reality of doing deep work. Shallow engagements produce shallow results. We'd rather do fewer things exceptionally well.
What We're Building
Costfunc is an applied AI research lab. We build products. Selectively, we build with others.
Our first product is Nix, an ML-powered notification manager. It's a small problem with a big impact—notifications are one of the most pervasive sources of digital friction, and existing solutions are inadequate.
What comes next depends on what we learn. We follow problems, not trends.
Join Us
If this resonates, we'd love to hear from you. Whether you're a potential partner, a future team member, or just someone interested in what we're doing—reach out.
The gap between AI research and real products won't close itself.