Good morning. It's Friday, April 10, and we're covering AI agents replacing software interfaces, breakthrough model efficiency, how AI is reshaping jobs, and more.
If you enjoy this email, share it with a friend. First time reading? Sign up here to keep up with the future of tech.
AI Is Replacing Work, Not Workers
A survey found 20% of full-time U.S. workers say AI has replaced some tasks. However, separate polling indicates that among employed AI users, 27% report replaced tasks, while 21% report the creation of new AI-driven work. Rather than total substitution, researchers project that AI will fundamentally reshape and augment roughly 50% of U.S. jobs over the next three years.
Court Rejects Anthropic Risk Appeal
A federal appeals court denied Anthropic’s emergency request to temporarily pause a “supply chain risk” label from the Pentagon pending a full review. The designation blocks new Defense Department contracts. The dispute stems from a failed $200M deal and limits on AI use. The case continues amid conflicting rulings.
OpenAI Eyes $102B Ad Business
OpenAI expects advertising to become its largest revenue driver, projecting $102B by 2030—about 36% of total revenue. It forecasts $2.4B in 2026 and nearly $11B in 2027, with ad revenue per user rising to $60. Early pilots show high pricing but limited results so far.
DeepMind Targets AI Drug Design
Demis Hassabis says DeepMind is advancing toward AI-driven drug discovery. Alphabet's Isomorphic Labs, an independent subsidiary spun out from DeepMind, is developing an internal drug candidate pipeline focused on oncology and immunology. A new AlphaFold-based model predicts small molecule binding affinities and biomolecular structures more accurately.
FRIDAY FACTS
The Pentagon Once Built a Powerful Supercomputer for $2 Million
A comparable system built the normal way would have cost ~$20 million or more. The hardware they used? You almost certainly own one. Answer ↓
POWERED BY ACCIO WORK
Accio Work: Your Agentic Team for Real Business
Meet Accio Work - the agentic workspace for business owners and solopreneurs. Our specialized agent team manages operations for you—sourcing products, negotiating with suppliers, managing stores, and launching marketing—all on autopilot.
With verified skills and business tool APIs, Accio Work takes action while you stay in control. Powered by Alibaba.com’s 1B+ products and global supplier data, it handles product selection and execution seamlessly. No setup required—just results.
VIDEO
Anthropic Banned OpenClaw…
Anthropic bans third-party Claude tools amid GPU crunch, sparking backlash, confusion, and rapid shift to GPT alternatives.
THE FUTURE LIVE
How do media companies remain editorially independent when they get acquired by large tech companies?@creatine_cycle Host @mots_pod shares his thoughts:"If some company gave me hundreds of millions of dollars, I might say I'd remain editorially independent. But in the back of… pic.twitter.com/oYnbwAgpYp
@ForwardFutureView post on XINTERFACES
Sierra CEO Bret Taylor Bets on AI Replacing Click-Based Software
At the HumanX conference, Bret Taylor, CEO of Sierra, said traditional click-based software interfaces could soon give way to natural language interactions. The company recently launched Ghostwriter, an “agent as a service” platform that lets users describe tasks and automatically generate AI agents to execute them.
Taylor argues this shift could eliminate the need to navigate complex enterprise tools like Workday. Sierra claims rapid adoption, including deploying an AI agent for Nordstrom in four weeks, and reaching a $100M annual revenue run rate within 21 months. Still, industry experts note current AI agents often require ongoing human tuning, limiting full autonomy. → Read the full article here.
HARDWARE
Ex-Apple Engineers Debut AI “Button” Device, Face Purpose Questions
Two former Apple engineers unveiled “Button,” a small AI wearable resembling an iPod Shuffle that activates a chatbot with a physical press. The device, created by ex-Apple Vision Pro team members, aims to offer a privacy-conscious alternative to always-on AI wearables by requiring user input to listen. It can respond via speaker or connect to earbuds and smart glasses over Bluetooth.
However, its creators have not clearly explained why the product should exist as dedicated hardware rather than a smartphone app. The launch follows poorly received AI devices like the Humane AI Pin and Rabbit R1, raising skepticism about the category. → Read the full article here.
NEWS
What Else is Happening
Intel, Google Expand AI CPU Push: The companies deepened ties to deploy Xeon 6 chips and co-develop infrastructure processors, targeting rising demand for AI inference and multi-step agentic systems.
CIA Reportedly Uses “Ghost Murmur” AI: Lockheed Martin system allegedly detected heartbeats across miles to locate a downed U.S. airman in Iran, highlighting emerging AI surveillance capabilities in military operations.
Florida Probes OpenAI Before IPO: Florida’s attorney general launched an investigation into OpenAI over security and misuse concerns, as the company eyes a potential $1 trillion public offering.
CoreWeave Lands $21B Meta Deal: CoreWeave signed a cloud agreement through December 2032 to supply AI compute capacity, underscoring surging infrastructure demand as Meta scales advanced AI workloads.
FRIDAY FACTS
It Was 1,760 PlayStation 3s
In 2010, the U.S. Air Force Research Laboratory wired together 1,760 Sony PS3 gaming consoles in Rome, New York and called it the Condor Cluster. At its peak, it ranked 33rd on the list of the world's most powerful supercomputers.
It could process 500 trillion floating-point operations per second. The Air Force Research Laboratory utilized the cluster for high-resolution satellite imagery analysis, pattern recognition, and early AI research.
Each PS3 ran about $400. Equivalent computing hardware cost $10,000 per unit. The full build came in at $2 million — roughly 10% of what a conventional system would have cost — and consumed 10% of the energy.
The PS3's IBM Cell processor was the key. It was built to handle massive amounts of simple math simultaneously, which made it near-perfect for parallel computing at scale.
