Share this article:

Inside Palantir: what the headlines leave out

Share This Article

Palantir is one of the most discussed technology companies in the world, and one of the least understood. In the latest episode of the Unit8 Podcast, our Executive Chairman Dr. Marcin Pietrzyk sat down with Courtney Bowman, Global Director of Privacy & Civil Liberties Engineering at Palantir, to talk about the company behind the headlines.

Courtney Bowman has spent 16 years at Palantir, since the company was about the size Unit8 is today. Over 45 minutes he explains what Palantir builds and what it refuses to do with data, why its forward-deployed engineers were once mocked, how his team designs privacy into the platform, and what AI coding tools mean for the next generation of engineers. Here are the ideas that stayed with us.

What Palantir builds, and what it stays out of

Palantir builds data integration and analysis platforms. They bring together information that sits in different formats and systems, so that companies, agencies and nonprofits can use it and govern it. He is just as clear about what the company does not do.

Palantir is not a data company. We don’t collect data, we don’t retail data, we’re not a data broker.

Courtney Bowman, Palantir

Palantir does not move data between customers, build AI models on customer data or sell surveillance tools. In AI, it supplies the infrastructure that connects third-party, frontier and open-weight models to a customer’s own data.

At the centre sits the ontology, which Courtney Bowman describes as a method for modelling “institutional data, actions and logic”. A manufacturer thinks in tools, parts and supply chains. The ontology maps raw data onto those concepts, so people can work with the business the way they already understand it.

The model people said would never work

For years, Palantir struggled to explain itself. People asked whether a company with so many engineers on customer sites was a software company or a consulting firm. Courtney Bowman answer: these are techno-social systems. Institutions that have been around for hundreds of years have their own ways of working, and technology lands when engineers sit inside them and bring it to the people who use it.

People mocked us for that model, and they said it was ridiculous and would never be successful. And now you see a lot of companies adopting this forward-deployed ethos and hiring FDEs.

Courtney Bowman, Palantir

Courtney Bowman named Unit8 among the companies that have learned from the forward-deployed model.

Privacy engineering starts where the law ends

Courtney Bowman’s team works where technology, law, policy and social norms meet. It brings people from philosophy, ethics and the social sciences together with computer scientists, and they build the controls that keep the platform accountable:

  • Granular access controls by role, classification, purpose or time.
  • Justified steps: points in a workflow where users explain why they move to the next step of an analysis, recorded in an audit trail.
  • Sensitive data discovery across an organisation’s information.
  • Granular encryption, so data can only be decrypted for specific uses.
  • AI governance tooling that maps each AI use case to frameworks such as the EU AI Act.

Technology moves faster than regulation, so the team builds ahead of it.

If you’re focused on building privacy protective technology, you can’t just look at the law. The law defines the floor, it doesn’t present the ceiling of opportunity.

Courtney Bowman, Palantir

The same thinking applies to testing AI. Benchmarks are set in laboratory conditions and describe a toy version of the world. Courtney Bowman argues for testing that matches the risk of each use case and keeps running after launch, from red teaming to perturbation testing, where small changes to a prompt show how much a model’s answers vary.

Engineering with AI still needs engineers

Dr. Marcin Pietrzyk shared what Unit8 engineers tell him: they now write their code through LLMs. At Palantir, engineers use models to move faster and still write production code directly, because a product rolled out across a market needs a level of trust that LLM-generated code cannot yet offer. New engineers may even start without AI coding tools until they have shown proficiency.

Both agree that, today, the best users of LLMs for code are the strongest engineers. Courtney Bowman sums it up with an analogy.

I don’t imagine a world in which you have a conductor who has never played an instrument. I think that’s a problem, and that’s super risky.

Courtney Bowman, Palantir

Keeping a culture at 4,000 people

Palantir hires people with “spikes” in performance or expertise, often before it knows which role they will grow into, and keeps many decisions close to the field. Courtney Bowman joined when the company was about Unit8’s size. What he works hardest to protect as it grows is the culture: the mission, the commitment to privacy and fundamental rights, and room for people to disagree openly.

The conversation covers more than we could fit here, from digital sovereignty in Europe to Palantir’s decision to build AIP almost overnight. What connects it is a practical view of technology: it creates value when it fits the institutions that use it, earns their trust and stays accountable to the people it serves. That view shapes how we work at Unit8 too, from forward-deployed teams to production-grade AI.

Watch the full episode

Want to receive updates from us?

close

This page is only available in english