Palantir's Danny Lutkus demos AI-driven supply chain consulting: from vague problem to proposal in a day

Summary

Palantir is positioning AI-driven consulting acceleration as a direct wedge against traditional strategy consulting engagements. Danny Lutkus, a 12-year Palantir engineer who now leads Midwest commercial business, demonstrated a live workflow that compresses what he describes as a six-to-nine month problem-to-implementation cycle down to roughly one week.

The core argument is that enterprise customers consistently hit a dead zone between identifying a problem and deploying a solution. That gap is currently filled either by internal subject matter experts or by strategy consultants charging millions of dollars for feasibility decks. Lutkus frames both as unnecessary friction on Palantir's sales cycle and on customer time-to-value.

How the Workflow Operates

The demo used Newark Liberty International Airport's runway capacity constraints as a stand-in for a typical Midwest manufacturing problem. A loosely worded two-sentence problem statement was fed into the system, which then ran a sequential agent pipeline:

• Prompt structuring agents reformatted the vague input into a structured problem definition with objectives, constraints, and requirements
• Research agents used a combination of proprietary internal data sources and Perplexity for open-web search, with all outputs stored back into Foundry's ontology layer
• Idea generation agents produced candidate solutions with associated implementation requirements
• Critique agents evaluated each idea across risk, economic feasibility, and safety and regulatory compliance, including auto-generated code for back-of-envelope financial modeling
• The final output is a structured proposal, designed to feed directly into AIP for Development Environments (AIFDE) for build execution

The orchestration layer runs on AIP Logic, Palantir's low-code tool inside Foundry. The underlying model in the demo was Grok 4, though Lutkus noted Palantir treats models as largely commoditized and is building evaluation frameworks that select the right model dynamically based on task type.

The Differentiation Claim

Lutkus argues the output quality is roughly comparable to a strategy consulting deliverable, with the caveat that consulting firms typically lack access to the proprietary data that Palantir customers already have inside Foundry. The enterprise data advantage, runway schematics, machine telemetry, inventory systems, is what makes the AI-generated proposal more grounded than a generic consulting deck.

The workflow is already deployed with existing customers, primarily large Midwest manufacturers. Named accounts in Lutkus's portfolio include Eaton, Johnson Controls, Molson Coors, and Cummins. He describes these as century-old companies that have shifted from skepticism to active deployment readiness, a posture he says has changed materially in the past two to three years.

Human Oversight and Trust Curve

Lutkus is explicit that the system is not designed for autonomous operation at launch. Human review at each stage is framed as both a quality control mechanism and a trust-building exercise. The stated goal is a progressive handoff, starting with human-in-the-loop validation and moving toward higher automation as organizational confidence in the outputs accumulates. He describes the endpoint as a "Jesus take the wheel" threshold, where trust in the system is sufficient to reduce human checkpoints. He also flags the incorporation of undocumented tribal knowledge as a prerequisite for reaching that threshold.

The commercial logic is straightforward. If Palantir can eliminate the consulting dependency from the pre-sales cycle, it shortens deal timelines, reduces budget that would otherwise leave the customer for a third party, and pulls forward the point at which a customer is ready to build on the platform.