Good morning, AI enthusiast.

Get ready for AI assistants to handle more than just browsing—they're starting to get their own wallets. Payment giants like Visa, Mastercard, and PayPal are unveiling new platforms that enable AI agents to securely complete transactions, paving the way for agentic commerce.

This move marks a significant step towards AI managing tasks from start to finish. Will delegating purchasing power to AI fundamentally change how we interact with online shopping, and what hurdles remain for user trust and security?

In today’s AI recap:

• AI agents gain payment power via new platforms

• Midjourney’s Omni-Reference for object consistency

• Apple partners with Anthropic for Xcode AI

AI Agents Get Spending Power

The Recap: Get ready for AI assistants that don't just shop, but actually pay. Visa, Mastercard, and PayPal are rolling out new platforms enabling AI agents to securely complete transactions, ushering in an era of agentic commerce.

Unpacked:

• The systems typically work by giving the AI agent special payment credentials, often using secure methods like digital tokens linked to your primary card, with user-defined spending limits and controls.

• PayPal introduced its PayPal Agent Toolkit, giving developers API access to integrate its payment processes directly into AI agent workflows.

• Major players are partnering with AI leaders like OpenAI and Perplexity, with Visa's CEO indicating agentic commerce capabilities rolling out within the next few quarters.

Bottom line: This marks a significant leap towards AI handling tasks end-to-end, potentially transforming how we shop online by delegating discovery and purchase. Building user trust and ensuring robust security controls will be critical for widespread adoption.

Midjourney Gets Object Recall

The Recap: Midjourney just launched Omni-Reference, a powerful new way to maintain consistency in image generation, alongside major speed and cost improvements for its V7 model.

Unpacked:

• Omni-Reference lets you grab elements like characters, objects, or even artistic styles from a reference image and apply them to new generations, helping you maintain consistency across multiple images.

• The update also includes significant speed and cost boosts for V7: standard jobs are now roughly twice as cheap, and turbo jobs run 30% faster.

• Early users suggest experimenting with the --ow parameter (omni weight) and describing your reference image's key features in the prompt to get the best results.

• This feature opens up huge possibilities for creating consistent characters or objects across different scenes, styles, and narratives.

Bottom line: This is a game-changer for creators needing consistency. Omni-Reference, combined with V7's efficiency gains, significantly boosts Midjourney's power for crafting complex visual narratives and maintaining brand assets or character designs with less hassle.

Apple Taps Anthropic for Xcode AI

The Recap: Apple is reportedly working with AI startup Anthropic to bring Claude Sonnet's coding smarts into Xcode, Bloomberg reports. This partnership aims to create an AI assistant, dubbed a "vibe-coding" platform, for internal Apple developers.

Unpacked:

• The tool is currently rolling out internally at Apple, with no final decision yet on a public release for third-party developers.

• It integrates Anthropic's Claude Sonnet model via a chat interface, letting developers request code generation, user interface testing, and bug fixes.

• This follows Apple's previously announced Swift Assist, an AI coding tool unveiled in 2024 but never released amid internal reports of hallucinations and development slowdowns.

• The move highlights Apple's strategy of blending in-house AI development with key partnerships, already involving OpenAI for Siri features and reportedly exploring Google's Gemini integration.

Bottom line: This Anthropic collaboration could significantly enhance Apple's developer tools and accelerate its push into generative AI capabilities. It signals Apple is getting serious about embedding AI deeply into its ecosystem, even if it means leaning on external expertise to catch up and innovate faster.

The Shortlist

OpenAI detailed the factors behind its recent GPT-4o sycophancy issue, attributing it to combined updates weakening primary reward signals and pledging process improvements.

Google will soon allow kids under 13 with parent-managed accounts to use its Gemini chatbot, adding specific guardrails and stating the data won't be used for training.

Reddit integrating its AI-powered 'Reddit Answers' feature directly into the core search experience to streamline access to community-sourced answers.

Neuralink received FDA Breakthrough Device Designation for its BCI technology aimed at restoring communication for individuals with severe speech impairment.

A federal judge questioned Meta's fair use defense for training Llama on copyrighted books, expressing skepticism about how creating competing products without a license could be fair.

Signing off,

David, Lucas, Mitchell — The Recap editorial team