DeepMind moves AI research into persistent games

DeepMind wants agents that can remember for years

EVE Online has been running since 2003, with thousands of players sharing one persistent universe shaped by trade, alliances, conflict, and changing rules.

That makes it unusually useful.

Google DeepMind is working with Fenris Creations to study AI agents that can learn continuously, remember past events, plan across long periods, and operate around other players.

The hard part is time.

Most current agents work inside short tasks and limited context windows, while EVE can force decisions whose consequences show up weeks or months later. That makes the game a credible test bed for capabilities AI labs keep talking about but still struggle to demonstrate reliably.

DeepMind is starting inside an offline copy of EVE, away from live players, before considering broader deployment.

You can picture the difference between completing a five-minute quest and remembering why an alliance broke six months ago.

If these agents improve, gaming may be only the first place the results matter.

The US Open is turning match data into live explanations

Every serve at this year’s US Open will generate data from 21 tracked points across the player’s body and racquet, sampled 50 times per second.

That adds up fast.

IBM and the USTA expect roughly 1.2 billion data points across the tournament, feeding a new Serve Quality score alongside upgraded win probabilities, match explanations, and an AI chat feature.

The useful shift is specificity.

Sports apps already drown fans in scores and statistics, so IBM is using AI to explain why a match changed while the ball is still bouncing across the court. That feels more practical than adding another generic chatbot because the system has live, structured data that can answer narrow questions well.

There is real demand here too: 91% of surveyed tennis fans say they use sports apps during events.

The bigger test is whether these explanations stay accurate when a match gets messy.

Cloudflare gives every employee their own software sandbox

Cloudflare employees built more than 4,000 custom business tools in 30 days using a system designed to keep each app isolated from everyone else’s data.

That’s the interesting part.

Cloudflare OS gives each user a separate app instance, while its Gatekeeper system controls exactly which documents, databases, and internal services that instance can touch. Employees can then ask AI to change the app’s code, with approvals required before sensitive actions run.

The security model matters more than the chatbot.

Companies already have people pasting API tokens into scripts and wiring AI into internal systems, often because the approved software cannot handle a specific workflow. Cloudflare is accepting that employees will keep building these tools and putting permission controls around the behavior instead.

You can almost hear the internal support queue shrinking.

The open question is whether companies want another internal platform, especially once Microsoft and Google offer similar controls inside software employees already use every morning.

Prohuman team

Free Guides

Explore our free guides and products to get into AI and master it.

All of them are free to access and would stay free for you.

Feeling generous?

You know someone who loves breakthroughs as much as you do.

Share The Prohuman it’s how smart people stay one update ahead.