Palantir's April: The Month Power Started Shifting to Users

If March was Palantir's defense month, April was the opposite. The biggest news beats this month had nothing to do with battlefields. They had to do with farms, audits, and a 10-year industrial customer doubling down. The product updates told a similar story: more power moving from FDEs and developers to the people who actually need to use the platform.

Here's what shipped, what was announced, and what I think it all means.

News: Three Deals That Tell a Bigger Story

Darling, I Love You, But Give Me Foundry Analysis Tools

If you've never seen the show, Green Acres ran from 1965 to 1971 and was basically a satire of government bureaucracy meeting actual farming. Oliver Douglas leaves Manhattan to be a farmer. Almost nobody around him helps. Mr. Kimball, the well-meaning but bumbling county extension agent, somehow always made his life harder instead of easier. Three forms to fill out. Lost paperwork. Conflicting advice between visits.

Mr. Kimball isn't going to be needed anymore.

On April 22, Palantir signed a $300 million agreement with the USDA to modernize how the department delivers services to American farmers. It supports the National Farm Security Action Plan and builds on the existing Landmark platform Palantir already runs at USDA. The "One Farmer, One File" initiative is the operational headline. Real-time supply chain visibility. Faster disaster aid. Less fraud and waste in farm programs. In other words, every Mr. Kimball-shaped friction point in the system gets a software upgrade.

This one matters more than the typical defense contract gets credit for. Farmers run individual businesses where nature is their best friend or their biggest threat. Better data means better yield predictions, smarter crop protection, and real margin improvements on operations that usually run thin.

I'd love to see this use case work well not just in the US but across other countries too. Food shortages aren't theoretical, and as trade tensions rise and political friction over embargoes and supply chains gets uglier, food sovereignty is going to matter more, not less. The infrastructure that helps farmers produce more efficiently with fewer wasted inputs is the same infrastructure that helps a nation feed itself when external markets aren't reliable.

Hard to make a sitcom about that. But it's the one that actually matters.

IRS Expands Its Palantir Relationship to $130M

A few days after the USDA news, the IRS confirmed an expansion of its work with Palantir's Lead and Case Analytics platform, bringing total investments past $130 million. The IRS Criminal Investigation Division has been using the platform since 2018 to aggregate and analyze federal databases for financial crimes work.

The interesting part isn't the dollar amount. It's the duration. Palantir landed inside the IRS in 2018 and never left. Now they're expanding the relationship eight years later. That's the pattern I see again and again with their commercial and government customers. Once Palantir proves its worth, organizations don't just renew, they keep finding more to do with it. There's always another use case beyond the original deployment.

Companies that embrace it tend to expand with it. The IRS deal is just the latest example.

Stellantis Renews Its 10-Year Partnership for Another Five

On the commercial side, Stellantis renewed and expanded its partnership with Palantir, originally signed back in 2016. They're broadening Foundry usage and now adopting AIP in select business functions and regions.

Two things stand out here. First, a 10-year customer renewing for 5 more years is the strongest possible buyer signal. Renewals from established enterprise customers don't make headlines the way new logos do, but they tell a more honest story. The ROI is real. The trust is real. The platform delivers enough that they're committing for the rest of the decade.

Second, Stellantis is now expanding from Foundry into AIP. That progression interests me. They might have just been using the classic parts of Foundry up until now, like Code Repositories and Pipeline Builder, and only now feel ready to layer AI on top of a stack they already trust. It could also be a partnership signal, a way of validating Palantir's broader ecosystem play after a decade of working together. Either way, the move from Foundry to AIP after a long Foundry-only relationship is a pattern worth watching.

Product Updates: Three That Actually Matter

AIP Evals in AI FDE Closes the Trust Loop

I wrote about AI FDE last month and flagged the trust problem as the biggest hesitation point for clients rolling it out. On April 14, Palantir shipped AIP Evals directly inside AI FDE. You can author eval suites, run them, and review results without leaving the agent. It even self-loops, automatically diagnosing and rerunning failed tests until the eval passes.

I've used it within AI FDE and it's genuinely cool. Watching it iterate on its own AIP logic until the evals come back clean feels like watching the platform grow up. The realistic use case is testing AIP logic and comparing models against each other to find equal performance at lower cost. The eval suites are written by builders working in the system, not by external QA teams, and that's the right home for them.

That said, this doesn't dramatically change my advice on rolling out AI FDE. The trust problem isn't gone, it just got easier to verify when you're in the loop. If you're standing up AIP logic functions and need to ship them quickly, having AI FDE test its own work before you sign off is a real time-saver. If you're skeptical about AI FDE in general, evals are a tool, not a guarantee.

Email Media Sets Just Unlocked an Entire Data Universe

Foundry can now ingest and transform .eml email files through media sets, with attachment extraction built in. This shipped on April 9 and almost nobody is talking about it.

Anyone who's tried to get email data into Foundry the old way knows why this matters. You'd bring messages in through API calls or data connections, get them as datasets, and then spend forever stitching messages to events to meetings to attachments, hoping you could even see the attachment data on the other side. It was disjointed and brittle.

Now you can just upload .eml files and see everything together. Body, headers, attachments, threads. That's a meaningful unlock.

Use cases that suddenly become realistic: AI sifting through inboxes to highlight what's actually worth your time. Automatic procurement processes built on top of supplier emails and PO docs. Legal teams passing documents around with full context preserved. Construction change orders, status updates, and contractor communications stitched into actual project data.

Industries with heavy email regulation or millions of routine messages are going to jump on this first. Legal. Compliance. Construction. Anywhere communication is the audit trail.

Object and Property Security Policies Just Replaced Restricted Views

On April 23, object and property security policies went generally available. This is the official replacement for restricted views as the way to handle fine-grained security in the ontology.

Restricted views weren't terrible. They worked. But they were another step and another dataset to manage, and when the rules got complicated they became a web to untangle. Permission inheritance across multiple restricted views is one of those things that looks fine in a small deployment and becomes a maintenance project at scale.

Moving security to the ontology layer changes the org chart on a Foundry deployment. Super admins can focus on bringing in the data. Ontology admins can manage permissions on the fly without going through file structure plumbing. Different roles, cleaner separation, less duplicated logic.

Honest take: this is a nice-to-have, not a revolution. I wouldn't rush to migrate existing restricted views just because you can. But for any new permission setup going forward, this is the way to go. Use it on greenfield work, leave the existing views alone unless you have a specific reason to change them.

A Throwback: AIP Analyst Has Been Working

I covered AIP Analyst in detail when it went GA the week of April 13. Quick update after spending more time with it: it's powerful, the results are remarkable on live data, and it's a great selling point for non-technical users who want to see how their data interacts in real time without needing a custom app or dashboard.

The honest caveat is that it's slow. The tradeoff is worth it when you need flexibility and live exploration. It's not worth it when you're running the same Monday morning report every week.

If you missed the original write-up, here's the breakdown on what it does and who it's for.

Everything Else That Shipped in April

A packed month of smaller updates worth knowing about:

• Claude Opus 4.7 added (April 21) with a 1M token context window
• AIP Agent Studio renamed to AIP Chatbot Studio (April 22)
• LLM capacity now autoscales up to 2x by default (April 21)
• Media item version history with retention policies (April 15)
• GPT-5.1 available on Azure OpenAI for IL2, IL4, and IL5 enrollments (April 14)
• Quiver redesigned graph mode (April 14)
• Pipeline Builder managed compute profiles (April 14)
• Grok 4.20 from xAI added (April 7)
• Nvidia Nemotron 3 models added (April 7)
• Incremental media set inputs in Pipeline Builder (April 7)
• Models in Pipeline Builder with no-code inference (April 2)

For the full list, Palantir's April 2026 announcements page has everything.

Watch the Startup Fellowship

One more thing worth flagging. Palantir's Startup Fellowship Cohort 002 kicked off this month and the lineup is genuinely interesting. From what's been publicly identified, the cohort includes:

• Pennant AI (Ryan Nowicki Stewart)
• Àjamais AI (Yufan Chen, a USC student-founder)
• Monq (Duygu Gozeler Porchet, AI-powered procurement)
• Ryniant (Arnav Simha, a 16-year-old AI researcher)
• Antonine (Zachary Wilson)
• BrockBox (Kyle Brockley, sovereign AI infrastructure)
• Inquiro AI Research Labs (Matthew Cook, drone and autonomous systems)
• Terp Design (Maxwell Kole, physical AI companion)

It's going to be cool to see what these teams ship in 8 weeks on AIP and Foundry. Procurement, sovereign infrastructure, drone autonomy, physical AI hardware. The breadth alone tells you something about where Palantir thinks the next wave of AI builders is coming from. Worth watching.

What to Watch in May

Earnings drop May 4. Cohort 002 will start shipping demos soon. And given the trajectory of April's product updates, I'd bet we see more AIP tools aimed at non-developer users over the next month or two. The platform shift from "tool for builders" to "tool for everyone" is real and accelerating.

The biggest takeaway from April: get more of your team building with AI FDE and using AIP Analyst. The features are there. The trust mechanisms are starting to catch up. The orgs that figure out how to deploy these tools across both technical and non-technical roles are going to outrun the ones that keep treating Foundry as a developer-only platform.