One of the lines in our deck is "Built on frontier models." A few clients have asked what we actually mean by that, and why we feel strongly enough to put it on slide four.
The short version.
There are a small number of AI labs in the world building the most advanced AI models available. As of mid 2026, that group is essentially Anthropic, OpenAI, and Google DeepMind, with strong arguments for a few others depending on the task. We call the models from these labs frontier models because they sit at the top of every public benchmark for reasoning, coding, and agentic work.
Underneath that frontier is a much larger market of smaller, cheaper, often open source models. They are useful for specific narrow tasks. They are not what you want running a digital employee that talks to your clients and touches your books.
Why this matters for you.
A digital employee is only as smart as the model running it. If the model cannot follow a long instruction, your employee cannot follow a long instruction. If the model cannot reason about a contract clause, your employee cannot either. The model is the ceiling.
Some AI platforms quietly switch you to a cheaper, smaller model in the background to save on their costs. You will not see this in the marketing. You will see it in the quality of the output. Suddenly the assistant feels duller. Forgets context. Misses things. That is the trade-off being made for you, without your knowledge.
If the model gets watered down, the employee gets watered down. There is no other way for it to go.
What we do.
We run your digital employees on the latest frontier model that fits the task. For most work we use Claude from Anthropic, because in our testing it has been consistently the best at agentic, long-context, business-style tasks. For specific work, we sometimes route to other frontier models that perform better on that exact use case.
You always know which model is running which employee. We tell you. We do not switch quietly. If a better model launches, we test it, we tell you, and you decide whether to upgrade.
The pricing reality.
Frontier models cost more to run than the cheaper alternatives. We absorb most of that in the one-time build cost. The ongoing inference cost is a fraction of what a human in the same role would cost, but it is not free, and we are transparent about that line item.
Said differently: we would rather charge you a bit more and give you the best AI available, than charge you less and give you something that will quietly disappoint your team in three months.
