Amazon's secret Burnham project aims to give Astro a brain

The dream of having a robot housekeeper like Rosie from The Jetsons has been stuck in sci-fi limbo for decades. Amazon tried to get us a little closer with its wheeled home robot, Astro, but anyone who tested it knows the truth: without arms and real intelligence, it was basically an expensive tablet on wheels that didn’t do much.

Now, leaked internal documents reveal Amazon is cooking up a secret generative AI project called Burnham to fix exactly that.

A brain for a body that was waiting for one

According to internal documents seen by Insider, Amazon is developing Burnham to add a layer of intelligence and a conversational interface to Astro using large language models. The internal goal is blunt: Amazon writes that its robot already has a strong body, and what it needs next is a brain.

Amazon calls the core capability contextual understanding. Instead of just dodging furniture, an upgraded Astro would remember what it saw and understood throughout the day, answer your questions about the house, and take action.

Some of the scenarios detailed in the documents sound genuinely useful. The robot could spot an open stove flame or a running faucet, track you down in the house, and alert you. It could check on someone who fell and call 911 in an emergency. It could help you find lost keys, check if a window was left open overnight, or monitor if the kids brought friends over after school.

Even more ambitious, Amazon is testing proactive problem-solving. In one example, if Astro sees broken glass on the floor, it realizes the hazard and prioritizes sweeping it up before someone steps on it.

The long road out of the lab

Don't expect your Astro to turn into a full-blown butler tomorrow.

Amazon itself acknowledges in the documents that Burnham has a long way to go before it can ship in a real product. Meanwhile, the current Astro remains invite-only, its price tag recently climbed to $1,600, and Insider reports that Amazon has scrapped plans for a cheaper model.

There is also the creeping creepiness factor. A roving camera that literally records, remembers, and interprets everything happening behind closed doors—including who your kids hang out with—will raise massive privacy red flags.

Why it matters

We think Burnham represents the inevitable next battlefield for generative AI: embodied intelligence in the real world.

For consumers, an Astro powered by Burnham is definitely not something worth spending money on today. At $1,600 and locked behind an invite system, it is an experimental curiosity. But conceptually, it shows what smart homes actually should be. The current ecosystem of fragmented sensors and rigid automations is painful. A single roving agent that understands context and acts autonomously would actually deliver on the smart home promise.

For the industry, Amazon’s move puts pressure on Apple and Google. Both have massive smart home footprints and LLM ambitions, but neither has an embodied hardware platform moving around the home. If Amazon can successfully fuse LLMs with robotics, it could redefine home robotics.

The real danger lies in execution and trust. Generative models hallucinate, and in the physical world, mistakes are dangerous. A robot that misinterprets broken glass or calls 911 on a false alarm is a liability. More importantly, convincing people to let an LLM-powered surveillance machine roam their living rooms might be a harder problem to solve than the AI itself.