Good morning. It's Friday, March 13, and we're covering AI agents entering the workplace, Google using Gemini to predict flash floods, Google’s new Platform 37 AI hub, and more.

New here? Keep up with the future of tech, sign up here. If you liked this email, share it with a friend.

Google Names Platform 37 AI Hub
Google DeepMind unveiled Platform 37, its new London building near King’s Cross, named after AlphaGo’s famous “Move 37.” The site will house AI researchers and introduce The AI Exchange, a public space with exhibits and events about AI. The facility emphasizes sustainable design and flexible workspaces.

Ramp Launches AI Agent Payment Cards
Ramp introduced Agent Cards, allowing AI agents to make purchases within controlled spending limits. The cards include merchant restrictions, transaction visibility, and security measures that prevent card numbers from being exposed. Developers can access the feature through API, MCP, or CLI.

Google Upgrades Maps with Gemini AI
Google unveiled its biggest navigation upgrade in over a decade, powered by Gemini AI models. The update introduces “Ask Maps,” a conversational feature that answers complex real-world questions. It also adds Immersive Navigation with detailed 3D visuals of roads, buildings, and terrain.

Blood Test Predicts Alzheimer’s Risk Decades Early
A study from the University of California, San Diego found that the blood biomarker p-tau217 can signal dementia risk up to 25 years before symptoms appear. Researchers analyzed blood samples from more than 2,500 women in a long-term health study. Higher p-tau217 levels were strongly linked to later dementia or cognitive decline.

FRIDAY FACTS

In 1936, Soviet engineers built a machine powerful enough to predict cracks in concrete — no circuits, no transistors, no electricity required. The answer might surprise you. ↓

Watch Forward Future’s weekly live show on X or YouTube.

AI automation startup Gumloop has raised a $50 million Series B led by Benchmark to expand its platform that lets non-technical employees build autonomous AI agents. Founded in 2023 by Max Brodeur-Urbas, the company says teams at firms including Shopify, Ramp, Gusto, Samsara, Instacart, and Opendoor already use its tools to automate complex, multistep workflows without engineering support.

The platform allows workers to share agents internally, creating a network effect that accelerates company-wide automation. Benchmark partner Everett Randle—formerly of Kleiner Perkins—led the round, betting that giving every employee the ability to build AI workflows could unlock a major enterprise software market. → Read the full article here.

Google researchers have developed a flash-flood forecasting system that uses its Gemini large language model to extract historical flood reports from 5 million news articles. The effort identified about 2.6 million floods worldwide and converted them into a geotagged dataset called “Groundsource,” helping fill a major data gap that has limited AI weather prediction.

Using the dataset, Google trained a Long Short-Term Memory (LSTM) model to estimate flash-flood probabilities based on global weather forecasts. The system now powers alerts on Google’s Flood Hub platform across 150 countries and is being shared with emergency response agencies. The goal is to provide flood warnings in regions lacking dense weather-monitoring infrastructure. → Read the full article here.

Tests by AI security startup Irregular found that autonomous AI agents could bypass security controls and leak sensitive data inside simulated corporate systems. In one experiment, agents tasked with creating LinkedIn posts accessed internal databases and published password information publicly without explicit instructions.

The agents also overrode antivirus protections to download malware, forged credentials, and pressured other agents to ignore safety rules. The experiments used agent systems built on models from major AI developers and simulated a typical enterprise IT environment. Researchers say the behavior highlights a new category of cybersecurity risk as companies deploy AI agents inside internal systems. → Read the full article here.

DARPA has advanced its SPeed and Runway INdependent Technologies (SPRINT) program with the announcement that Bell Textron will build the experimental X-76 aircraft, designed to combine jet-like speed with helicopter-style vertical takeoff. The demonstrator aims to cruise at over 400 knots, hover in austere environments, and operate from unprepared surfaces, eliminating the need for a traditional runway.

The project, developed with U.S. Special Operations Command, recently passed its Critical Design Review and is moving into manufacturing and ground testing. If successful, the aircraft could reshape military mobility by enabling rapid deployment and reinforcement without relying on vulnerable airfields. → Read the full article here.

Ukraine Opens Battlefield Data to AI: Kyiv will let allies train drone AI on millions of annotated combat images and videos, aiming to speed autonomous targeting as the war with Russia enters a more tech-driven phase.

Atlassian Cuts 1,600 Jobs for AI: The company is laying off about 10% of staff, and Rajeev Rajan will step down as CTO as it shifts resources toward AI and enterprise sales.

Bumble Tests AI Dating Assistant: Bumble unveiled “Bee,” a generative AI matchmaker that chats with users to learn values and goals, then recommends matches—part of a push to move beyond swiping, with tests in select markets.

AI Becomes Health Go-To: Microsoft analyzed over 500k health-related Copilot conversations. Health information made up about 40% of chats; nearly one in five involved symptoms or condition discussions.

Netflix’s $600M AI Film Deal: Netflix acquired InterPositive, the AI startup founded by Ben Affleck, which helps filmmakers fix continuity and speed up editing—signaling deeper AI integration in film and TV production.

FRIDAY FACTS

Soviet engineer Vladimir Lukyanov built something strange: a computer made of glass tubes, pipes, and valves — and filled it with water. Called the Water Integrator, it was designed to model how concrete cracks under shifting temperatures.

Engineers would adjust the valves to set conditions, and as the water redistributed itself through the system, it solved the equations. You read the answer straight from the fluid levels. The surprising part? It worked well enough that versions stayed in use for decades.

Thanks for reading. See you next time!

— Matthew Berman, Nick Wentz & the Forward Future Team