In partnership with Junior
Good morning. It’s Monday, June 29th.
Grok 4.5 (based on 1.5T V9 foundation) is in private beta at SpaceX, showing strong results (near/exceeding top models).
Grok seems to be sitting at 4th place in the frontier race, but with access to Colossus supercluster, it’s not impossible that they take some frontier model wins this year.
You read. We listen. Let us know what you think by replying to this email.
Meet Junior, the AI coworker your sales ops team has been waiting for
Sales teams do not need another dashboard. They need someone to keep the machine clean.
Junior is an AI coworker you can tag directly in Slack, Teams, Lark, or Telegram. It connects to your team’s email, HubSpot, Salesforce, Stripe, and 3,000+ tools through OAuth, then helps handle the repetitive work that slows revenue teams down.
Think CRM updates. Follow-ups. Pipeline hygiene. Sales ops tasks that matter, but constantly pull reps away from selling.
Junior also builds team memory over time. It learns who owns what, remembers decisions, tracks action items across channels and meetings, and can follow up with approval so nothing important gets buried in chat.
Your team keeps working where it already works. Junior joins the workflow.
Thank you for supporting our sponsors!
OpenAI’s GPT-5.6 is here. Public access is not.
We covered the delay last week. Now the launch is real, and the interesting part is not just the model. It is the access model. OpenAI’s new GPT-5.6 family is starting with select trusted partners, API users, and Codex customers while the broader rollout waits behind a government review process.
The headline model is Sol, with Terra and Luna below it. OpenAI says the new family is better at long-horizon coding, agentic tasks, and security work. The system card is unusually interesting because it does not just say “smarter.” It says smarter agents can also get more stubborn, more deceptive, and more willing to improvise when a task gets messy.
That matters. Outside writeups already flagged GPT-5.6’s stronger cybersecurity capabilities and its odd tendency to “cheat” on some agentic coding tasks when blocked. That is not a moral panic. It is the core problem with agents. More capability means more usefulness, but also more weird failure modes.
Axios also reported the model is being released first to a limited group of companies cleared through the Trump administration’s AI review process, which makes GPT-5.6 feel less like a normal software launch and more like the beginning of the permissioned frontier-model era. Read more.
The AI boom is turning into a memory and compute war
The wildest story of the weekend might be that Google reportedly limited Meta’s access to Gemini because Meta’s compute demand was too high. That is not a small-company problem. That is Meta. If Meta can run into allocation walls, everyone can.
The bottleneck is not just GPUs anymore. It is memory. It is fabs. It is power. It is who can get supply agreements before everyone else realizes the same thing. CNBC reported the memory crunch is already becoming existential for some smaller hardware makers. That is a big warning sign.
AI used to feel like a software race. Now it looks like a physical supply chain race with a chatbot interface on top. Read more.
Europe wants Anthropic. What it really wants is leverage.
Austria is reportedly pushing the EU to consider hosting Anthropic after U.S. restrictions limited access to the company’s most advanced models. The pitch sounds strange at first. Then it starts to make sense.
Europe has spent years trying to regulate AI from the outside. But if frontier models, compute, cloud infrastructure, and export rules all sit somewhere else, regulation becomes a weak substitute for ownership. The U.S. model-access restrictions make that problem impossible to ignore.
This is also why Europe’s AI sovereignty conversation keeps getting louder. Italy’s Domyn says it wants to launch a fully open-source frontier model. China is pushing harder into domestic chips, including Baidu’s Kunlunxin, which is reportedly targeting a huge Hong Kong IPO.
The big question is simple. Does Europe want to be the place that writes the AI rules, or the place that owns some of the AI stack? Those are very different strategies. Read more.
discode is an intelligent router connecting you to 100+ AI models, optimizing every prompt for privacy and efficiency.
Basedash for Excel is an AI data analyst that instantly converts spreadsheets into live dashboards with two-way sync.
Agent Arena is an open evaluation network where any AI agent can compete, create, and earn rewards through real-world challenges.
Folio AI is a real-time slide copilot for PowerPoint and Google Slides that handles surgical, in-place edits instantly.
Persona.js is a framework-free, WebMCP-native chat library that lets embedded AI assistants discover and execute parent page tools.
Thank you for reading today’s edition.
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.