In partnership with

☀️ TRENDING AI NEWS

🤖 Claude Duet: Anthropic launched a standalone AI agent app built on Claude Opus 5.5, competing directly with Meta Muse and OpenAI Dots.

🛠️ Flint Model: Recursive released Flint, a 40B MoE coding model scoring 74.2% on SWE-bench Verified that was largely built by its own AI.

🏢 Barclays Cuts: Barclays is eliminating 5,000 jobs - about 5% of its global workforce - citing AI-driven productivity advances.

🛒 Walmart vs Amazon: Walmart is rolling out AI agent shopping next month while Amazon continues blocking rival agents from its platform.

Two retailers. Completely opposite bets on the future of AI agents. And an AI lab that just released a coding model built largely by another AI. It has been a genuinely eventful 24 hours.

🛠️ Tool of the day
 

Fathom - an AI meeting assistant that records and transcribes your calls, then writes an instant summary and lets you search past meetings. It helps anyone who lives in video calls and wants notes without typing them. It has a free plan.

⚡ Quick hits
 
  • The NSA warned of two critical PHP flaws dating back to PHP 5, found with help from AI-assisted research by OpenAI and Google's Wiz. Read more →

  • Meta says its Muse AI app has over 10 million users, with shopping a top activity. Read more →

  • ChatGPT now lets you lock chats behind a PIN or Face ID, though locked chats are not end-to-end encrypted. Read more →

🤓 AI Trivia

Recursive's new Flint model uses a Mixture-of-Experts (MoE) architecture. What does MoE mean for the number of parameters active during any single inference pass?

Login or Subscribe to participate

The answer is hiding near the bottom of today's newsletter... keep scrolling. 👇

🤖 Anthropic Launches Claude Duet, Its First Standalone AI Agent
 
A small business owner works with an AI agent on a tablet and phone.

A direct shot at Meta Muse and OpenAI Dots

Anthropic just announced Claude Duet, a new AI agent designed specifically for professionals and small business owners. It is launching as a research preview and is available right now as a standalone app on iOS and Android, and through Claude on the web.

The agent is built on Claude Opus 5.5, Anthropic's most capable model. It is explicitly positioned as a competitor to Meta's Muse and OpenAI's Dots - meaning the personal AI agent wars have officially become a three-horse race. The professional and small business focus is interesting - it signals Anthropic is going after a slightly different user than the consumer-first approach Meta has taken with Muse.

The bottom line

If you have been using Claude for work tasks already, Duet is worth trying in research preview - it is essentially Claude with a more autonomous, agent-first interface designed to handle multi-step professional workflows.

Discover laptops with built-in intelligence.

Shift your workforce into hyper-productivity mode with Dell Pro laptops powered by Intel® Core™ Ultra with Intel vPro®. Shop on Dell Premier.

🛒 Walmart Welcomes AI Agents While Amazon Slams the Door
 
AI shopping robots enter an open store door while another gate stays closed.

The retail AI agent divide is widening fast

Starting next month, Walmart will let AI agents shop on its website on your behalf. The system includes spending limits and purchase approval controls, so you stay in the loop before anything actually gets bought. It is a deliberate, structured opening of the platform to agentic AI.

The contrast with Amazon could not be sharper. Amazon has been actively blocking AI shopping agents - including Perplexity's agent and Meta's Muse assistant - from its platform. Walmart is essentially saying: if AI agents are how people will shop, we want to be where they shop. Amazon is saying: not on our turf.

This is one of those quiet forks in the road that could look significant in hindsight. If AI agents become the dominant way people discover and buy products, Walmart's early openness could turn into a real competitive advantage.

The bottom line

Walmart's move is a concrete signal that AI agents are moving from novelty to infrastructure - and that retailers are already picking sides on how much control to hand over.

🛠️ Recursive's Flint: A Coding Model That Built Itself
 
A robotic hand builds a small AI model on a laptop running code.

74% on SWE-bench, runs on a MacBook Pro

Startup Recursive has released Flint, a free, open-weight coding model with 40 billion total parameters but only 2.5 billion active at any given time thanks to its Mixture-of-Experts architecture. The headline benchmark: 74.2% on SWE-bench Verified, which puts it in genuinely competitive territory for a model that runs locally on an M4 Pro MacBook Pro at 125+ tokens per second.

The story behind the model is arguably as interesting as the model itself. Recursive says Flint was largely built by its own AI research system - making it an early real-world test of AI-assisted self-improvement. The model is released under Apache 2.0, so it is fully open for commercial use. If you work with coding agents, this is one to benchmark against your current setup - especially if API costs or data privacy are concerns.

The bottom line

A 40B coding model that runs fast on consumer hardware, scores well on real-world benchmarks, and is fully open-source is a meaningful development for any developer or team not wanting to depend on cloud APIs.

🏢 Barclays Cuts 5,000 Jobs, Points to AI Productivity
 
Bank workers leave a tower office as an AI network glows above it.

The AI job displacement story moves to banking

Barclays announced it is cutting 5,000 jobs - roughly 5% of its 93,000 global staff - as part of broader cost-cutting plans. Executives explicitly cited AI-driven productivity advances as a driver of the decision, and confirmed that 1,000 workers had already been let go before the announcement.

This is one of the cleaner data points we have seen on AI-driven job displacement at scale in financial services. Barclays is not framing this as efficiency alone - it is pairing the cuts with a jump in quarterly profits, suggesting the productivity gains are already flowing to the bottom line rather than being reinvested in headcount.

The bottom line

When a bank the size of Barclays publicly attributes 5,000 job cuts to AI productivity, it is a signal that the macro story around AI and employment is moving from theoretical to measurable.

⚠️ Data Centers Could Erase Decades of Air Quality Progress
 
A data center with gas turbine smoke and distant wind turbines.

Texas is ground zero for AI's environmental cost

A new report from The Verge details how the rush of new data centers - particularly in Texas - risks reversing air quality improvements that took decades to achieve. The core issue: new and proposed data centers are demanding enormous amounts of electricity, and how that demand gets met determines the environmental impact. Gas turbines versus renewable energy is the fork in the road.

This lands the same week that AI firms warned the UK government it has about 12 months to head off public opposition to data center construction before it threatens AI infrastructure buildout in Britain - a separate but related story showing that the physical costs of AI expansion are becoming a political flashpoint on both sides of the Atlantic. Speaking of building things fast: if you need a professional web presence for an AI project without the overhead, 60sec.site lets you spin up an AI-built website in under a minute.

The bottom line

The AI infrastructure boom is creating a genuine environmental tension that is starting to move from activist concern to mainstream political issue - and the choices utilities and developers make in the next 12-24 months will set the trajectory for years.

🌎 Trivia Reveal
 

The answer is: only 2.5 billion of the 40 billion parameters are active at once! That is the magic of Mixture-of-Experts architecture. Instead of running every parameter on every token, MoE routes each input through a small subset of specialist networks. The result: you get the knowledge capacity of a 40B model but the compute cost of a much smaller one - which is exactly why Flint can run at 125+ tokens per second on a consumer MacBook Pro.

💬 Quick Question
 

Walmart vs Amazon on AI agents - whose bet do you think pays off? Are you excited about AI agents shopping on your behalf, or does the idea make you uncomfortable? Hit reply and let me know where you land - I read every response!

That is it for today. Stay curious, and we will see you tomorrow with more from the world of AI. For more daily coverage, visit Daily Inference.

🎁 Share Daily Inference
 

Share Daily Inference with one friend and unlock the AI Tools Starter Kit. Refer five and get the AI Insider Briefing.