Palantir's forward deployed architect Chad Wahlquist on ontology, the death of dashboards, and the Evolve tool cutting token costs 60%

Palantir's Chad Wahlquist on ontology, dashboards, and Evolve

Chad Wahlquist, a Forward Deployed Architect at Palantir, makes a living solving a problem most enterprise software companies prefer to ignore: businesses don't actually understand how they make money. Wahlquist says he regularly encounters Fortune 500 companies, and smaller ones, that can't identify their most profitable product, often actively pushing sales toward the wrong one because their data is aggregated up to KPIs before anyone can see the underlying mechanics.

The fix, in Palantir's framing, is the ontology — a semantic layer that models the business as it actually operates, rather than forcing it into rows and columns. Wahlquist contrasts this with dimensional modeling that dates to Kimball in 1996, which still underpins most Tableau deployments in 2026. The ontology can hold a CAD file, an image, a computer vision model, and tabular data inside a single semantic object representing, say, a manufacturing plant. That multimodal structure is what lets Palantir run operational applications, KPIs, and agents off the same foundation without rebuilding data architecture for each use case.

“We built a tool called Evolve that will actually go analyze the logs in production about how these models are operating, what people are doing with them, the architecture over it, and actually be able to swap out different models from different providers. Some of these customers were able to, in two days, eliminate 60% of their token cost by re-architecting, picking a different model, and prompt tuning.”

Kill all dashboards

Wahlquist is direct about dashboards: he wants to eliminate them. KPIs and metrics should be a byproduct of operational applications where decisions are actually made, not standalone artifacts built speculatively. Build a dashboard as the goal, he argues, and you get the field of dreams — nobody shows up. The OODA loop (observe, orient, decide, act) only hits the bottom line when it ends in action inside an application.

Evolve and the 60% token cost cut

Palantir recently launched a tool called Evolve that addresses a pattern Wahlquist sees constantly: teams build agentic workflows on the latest frontier model because it was the easiest way to get something working, then blow through their token budget. Evolve analyzes production logs — how models are performing, what users are doing, what the architecture looks like — and identifies where cheaper, older, or more deterministic models could handle parts of the workflow without quality loss. It also flags cases where cached models are sufficient, or where having a given piece of data in the ontology would eliminate the model call entirely.

One customer, cited by Palantir's McCarthy at a company event, cut token costs by 60% in two days through a combination of model swapping, re-architecture, and prompt tuning.

The enterprise AI edge you won't read about

The companies finding genuine AI advantages aren't posting about it. Wahlquist points to AIG as a public example — moving complex commercial insurance underwriting from months to hours or days creates a trust and speed advantage that competitors without the same tooling simply can't match in a competitive bid. The equivalent moments are happening across industries, he says, but stay quiet until they show up in earnings and margin data.

Palantir for Builders

Wahlquist says Palantir has a free developer tier and a Builders program that already includes two-person startups. There is a live Shopify integration. He describes one small business owner who used Palantir — largely self-taught through YouTube videos — to flip from a negative 10% margin to a positive 9–10% margin in three months. The vision of Palantir as a foundational layer for companies from day one, something you sign up for alongside Stripe Atlas, is where Wahlquist thinks the product is heading, even if the sales motion and storytelling aren't there yet.