Former Google Product Manager 'Vibe Coded' a Palantir in a Weekend

A former Google Maps product manager used 3 days and AI agents to build a real-time geolocation intelligence dashboard, and then Palantir's co-founder personally responded.

This project is called WorldView. After being posted on X, it blew up immediately. The original post got 1.9 million views and even hit the trending topics on X.

Bilawal Sidhu posted:

"It's truly crazy what you can build with Gemini 3.1 and Claude 4.6. It feels like Google Earth and Palantir had a baby."

Someone reposted saying "One person just vibe coded what Palantir charges the government millions for," and hedge funds and OSINT (Open Source Intelligence) analysts flocked to ask to participate.

Then Palantir co-founder Joe Lonsdale actually responded in a podcast.

Okay, so what does this thing actually look like?

Spy Movie Vibes

The author of WorldView is Bilawal Sidhu. He spent six years at Google, working on 3D maps and immersive views for Google Maps. He is also a TED Technology Curator and a VC Scout for a16z.

This time, he brought out his geospatial expertise and used an AI code editor to piece together a browser-based real-time global monitoring dashboard.

The foundation is Google Photorealistic 3D Tiles, the technology that lets you see 3D buildings in Google Earth. Bilawal participated in the launch of this project during his time at Google.

He then stacked a bunch of visual effects on top. Pressing number keys switches between CRT monitor mode, Night Vision mode, and FLIR thermal imaging mode. Each mode also allows adjusting parameters like sensitivity and pixelation.

In his own words: It looks like a command center from a spy movie.

How Intense is the Data Layer

The most hardcore part of WorldView is the real-time data layer it connects to. Basically, it brings in all publicly available data from the sky and the ground.

Satellite Tracking

It connected to real-time positions of over 180 in-orbit satellites via CelesTrak TLE data. Every satellite has a NORAD ID. Clicking any satellite allows you to track its orbit and see if it's a geosynchronous orbit or a polar orbit. The sky above the entire earth is densely packed with satellite movement trajectories, all real-time.

Civil Aviation Flights

Real-time flight data was loaded through the OpenSky Network, loading over 6,700 aircraft at once. Clicking any plane allows you to track its position in real-time. Bilawal said: "Real-time tracking of all planes on earth, presented on a 3D globe, that's just crazy."

Military Aircraft

This one is even more intense. Through crowdsourced data from ADS-B Exchange, it displays military aircraft that usually don't appear on standard flight tracking websites. The orange dots on the map are all military planes. He even demonstrated zooming in over the Pentagon to see what military flight missions are nearby.

Street Traffic

Using OpenStreetMap road network data, a particle system was generated to simulate city traffic. The densely packed moving dots on London Bridge represent vehicles, and it looks particularly spectacular in night vision and thermal imaging modes.

Real-time CCTV

Public real-time traffic cameras in Austin were connected, and the footage can be projected onto 3D building geometries. Although the update rate is one frame per minute, it is indeed real footage. He said: "Honestly, this kind of thing feels like it shouldn't be legal, but it indeed is legal."

Earthquake Data

Real-time markers for global earthquakes and crustal activity are also connected.

Stacking these data layers together—satellites, military aircraft, civil aviation, traffic cameras, earthquake activity—you get a fairly comprehensive real-time global situational awareness map.

How It Was Built

Bilawal used three AI coding tools: Gemini 3.1, Claude 4.6 and Codex 5.3.

But the key lies in his workflow. He opened 3 to 4 terminals at the same time, running an AI Agent in each, and let them work in parallel.

For example, one Agent works on visual filters and shader effects, another works on data ingestion and particle systems, and another handles street traffic simulation. At most, he was controlling 8 Agents simultaneously.

He gave AI direct commands in the command line, without using a GUI editor. He specifically mentioned using OpenClaw to manage multiple Agents at once.

Of course, AI makes mistakes too. He mentioned that when making the panoramic view, the Agent generated too many particles at once, crashing the browser directly. He told the AI: "Load main roads first, then side roads," and solved the problem with progressive loading.

Essentially, humans make creative decisions and problem decomposition, and AI does the code implementation.

Palantir Steps In

After the project went viral on X, people from all walks of life showed up.

Min Choi commented:

"An indie dev just vibe coded what Palantir charges the government millions for. Claude 4.6 + Gemini 3.1. The disruption of defense tech is coming."

Of course, some people poured cold water on it.

Idan Beck said:

"The only reason you can't vibe code Palantir is that no one actually knows what Palantir really does."

Then Palantir co-founder Joe Lonsdale actually came. He talked about this specifically on the TBPN podcast. The gist was:

"If you are building low-end SaaS, then yes, you should be worried. But if you've spent over $100 million building software infrastructure, no one can replace you. I am very bullish on AI, but the idea that 'we will disrupt everything, Palantir is doomed' is really something I can't agree with."

Lonsdale's core point is: Visualization capabilities like WorldView have long existed in classified systems. Palantir's value lies in the analysis layer, in connecting various confidential and proprietary data pipelines for deep correlation analysis. Visualization is just the most superficial thing.

Bilawal himself was quite relaxed, posting a sentence on Threads:

"Haha, when a Palantir co-founder responds to your vibe coding project on a podcast. Lonsdale being willing to respond shows that this thing is at least real enough to warrant his rebuttal, and that itself is enough."

Six Years + Three Days

Bilawal wrote a passage on LinkedIn that is perhaps worth pondering:

"I built WorldView in three days, but I also spent six years at Google studying geospatial technology."

It could be done in three days because of the six years of prior domain accumulation.

He made the same point at the end of the video.

Back when working at Google Maps, developers who wanted to use 3D Tiles faced an extremely steep learning curve.

Now with AI coding tools, the implementation barrier has been significantly lowered.

But this doesn't mean AI can help everyone make the same thing.

He could do this project so fast and so well because he knew what to do: knowing which data sources could be used, knowing how to project camera footage onto 3D geometries, knowing how to optimize when the particle system crashes.

AI lowers the implementation barrier, but domain knowledge remains key.

He said:

"If you have domain expertise, now is the time for you to get started."

Whether you are in finance, healthcare, or logistics, as long as you have deep enough industry understanding, AI coding tools can help you turn the ideas in your head into products that actually run.

This is arguably the most noteworthy information from this project:

AI is turning domain experts into full-stack developers.