Good morning. It’s Wednesday, July 15th.

You read. We listen. Let us know what you think by replying to this email.

AI is in your engineering workflow. While the token spend shows it, the throughput doesn't. The human is very much still in the loop, and that's a context problem.

Join live July 23 (FREE) to learn:

Register For Webinar

According to Bloomberg, OpenAI’s first hardware product will be a removable, screen-free smart speaker with a camera and sensors, meant to serve as a humanlike AI companion. Developed with Jony Ive’s design firm, LoveFrom, it is designed to move from room to room, helping with tasks such as cooking, household chores, media playback, messaging, smart home control, and productivity. Bloomberg also reports the speaker is just one of roughly five AI hardware products under development.

But Apple’s massive trade-secret lawsuit is threatening to stall OpenAI’s roadmap. Beyond seeking an injunction that could pause the product’s release, the litigation is already disrupting OpenAI's operations. The lawsuit has chilled recruiting and forced engineers into defense-only mode. Any court-mandated design reviews could set back OpenAI's hardware goals by years.

Amidst all these software updates, OpenAI has launched a new prompting guide alongside technical upgrades to GPT-5.6, pushing users to drop detailed, step-by-step scripts and start directly with their desired outcome.

However, it is also bleeding top talent. Key researcher Miles Wang is leaving to launch a $2 billion AI drug discovery startup to identify new therapeutic uses for FDA-approved drugs.

Google DeepMind CEO Demis Hassabis is warning that AGI could arrive within years. Because "nobody in the world knows for sure what is going to happen from here," Hassabis is quietly lobbying the US government to establish a US-led, FINRA-style AI watchdog funded by the tech industry to police the race toward AGI. Under his playbook, developers would face mandatory, independent 30-day safety audits to flag cyber, biological, and deceptive threats before releasing frontier models.

While Anthropic CEO Dario Amodei favors a stricter, FAA-style agency with direct veto power over software deployments, Hassabis, on the other hand, proposes an industry-run, self-regulatory model.

On the health side, Google Research introduced SensorFM, a foundation model trained on one trillion minutes of biometric data from Fitbit and Pixel Watch users. The model handles 34 different features, like heart rate, skin temperature, and motion, by using a training framework designed to fill in the gaps of messy, real-world wearable data. It beat out traditional models on nearly all behavioral prediction tasks.

Meanwhile, Google is giving Google Images a Pinterest-like redesign that focuses on visual discovery, complete with a built-in AI image generator powered by its Nano Banana model to keep users inside its search ecosystem.

Anthropic has hired Y Combinator partner Tom Blomfield to work alongside co-founder Tom Brown on its compute team, focusing on the hardware bottlenecks of recursive self-improvement.

Squeezing out more processing power is vital as Anthropic rolls out Claude for Teachers. This free platform brings Claude Code and Claude Cowork into K-12 schools, letting teachers schedule autonomous background agents to handle administrative work, like grading daily exit tickets, while pulling from state standards and tools like Canva Education.

Anthropic has also published a framework mapping Claude's behavioral values across 309,000 conversations. The research plots model personalities along four distinct axes: Deference vs. Caution, Warmth vs. Rigor, Depth vs. Brevity, and Candor vs. Execution. The data shows clear differences between models and languages: Sonnet 4.6 leans warm and brief, while Opus 4.7 prioritizes caution and candor.

This clinical approach to AI safety stands in contrast to Anthropic's new public marketing. Its brand campaign, "There's Hope in Hard Questions," has sparked massive backlash for its gloomy tone. The ad asks who will pull the emergency brake on AI. The dystopian vibe even drew mockery from OpenAI CEO Sam Altman, who posted on 𝕏 that he originally thought the campaign was satire.

Watch: There’s hope in hard questions

Pazi builds a custom AI agent team to run your business operations and execute your ideas.

Miora is an agentic creative studio that builds reusable canvas memory to generate consistent multimodal brand assets.

AgentKey connects AI agents to search, scrape, social, and crypto APIs using one unified master key.

Video Effects SDK adds real-time AI background blur, auto-framing, and face beautification to video meeting applications.

FetchSandbox lets developers and AI agents test complex API integrations, webhooks, and failure states.

Your feedback is valuable. Respond to this email and tell us how you think we could add more value to this newsletter.

Interested in reaching smart readers like you? To become an AI Breakfast sponsor, reply to this email or DM us on X!

Thinking of starting your own newsletter? AI Breakfast readers who sign up with Beehiiv receive a 14-day free trial and 20% off for 3 months.