Presented by
Good morning, AI enthusiasts. Former OpenAI CTO Mira Murati just launched her first product, and it's a direct shot at how GPT-4o and Gemini handle real-time voice.
Rather than stitching interactivity onto an existing model, Thinking Machines built it in from scratch. Whether that actually matters depends on whether the demos hold up at scale.
In today's recap:
Mira Murati's Thinking Machines previews realtime AI
Unitree ships world's first rideable manned mecha
Run parallel agents with Claude Code Agent View
OpenAI launches Daybreak and Deployment Company
4 new AI tools, prompts, and more
THINKING MACHINES
Mira Murati's lab previews native realtime interaction models
Thinking Machines
Recaply: Thinking Machines, Mira Murati's new AI lab, just unveiled a research preview of interaction models. They work in real time across audio, video, and text, with no external scaffolding.
Key details:
The model stays in a live two-way exchange with the user at all times. Users can interrupt or respond at any moment without waiting for turn changes.
The system has two parts: a fast model for real-time responses and a background model for heavier tasks. Both run at once.
Most AI voice products add interactivity as a layer on top. Thinking Machines trains it into the model directly, so the model gets better at both reasoning and collaboration as it scales, according to the company.
The research preview launched on May 11, 2026. No pricing or general availability date has been announced.
Why it matters: Mira Murati is betting that building interaction into the model rather than adding it on top means it gets better on its own over time. The early demos look and feel different from GPT-4o Realtime and Gemini Live. Latency at scale is still unproven. But the approach is worth watching.
PRESENTED BY VIKTOR
We hired one colleague for every department.
Last Tuesday, marketing asked Viktor to write the weekly campaign recap, pull performance from Google Ads and Meta, and format it as a PDF for the exec team. Done in four minutes.
That same afternoon, engineering asked Viktor to review three open pull requests on GitHub, cross-reference with the Linear sprint board, and flag anything blocking the release. Posted to private channel before standup.
At 9pm, ops asked Viktor to draft a vendor contract summary from three Notion docs and send it to the team. It was in #ops by morning.
None of them knew the others were using it.
Same colleague. Three departments. That's what changes when your AI coworker lives in Slack, where your whole company already works. It's not a tool one person logs into. It's a teammate everyone messages.
5,700+ teams. SOC 2 certified. Your data never trains models.
"Viktor is now an integral team member, and after weeks of use we still feel we haven't uncovered the full potential." - Patrick O'Doherty, Director, Yarra Web
UNITREE
Unitree launches the world's first manned mecha
Unitree
Recaply: Unitree just launched the GD01, the world's first mass-produced manned mecha, a $537K robot that walks on two legs with a human pilot riding inside.
Key details:
The GD01 has a bipedal walking frame with a padded cockpit. A human operator rides inside while the robot navigates terrain and switches between walking modes.
The GD01 costs ¥3.9M ($537K) per unit. Unitree already produces the Go series robot dogs and the G1 and H1 humanoid robots.
The cockpit has been padded for the production model, which suggests early rides are rough, according to Hacker News discussion. Engineers expect gait quality to improve as production feedback comes in.
The GD01 entered mass production in May 2026. Buyers in professional and industrial sectors can purchase at $537K per unit.
Why it matters: Unitree made its name on affordable robot dogs. The GD01 is something else entirely. It shows a company known for cheap robotics can now make something you ride inside of. Piloted robots entering mass production changes what human-machine interfaces could look like over the next decade.
GUIDES
Use Claude Code Agent View for parallel coding tasks
Recaply: In this tutorial, you will learn how to dispatch and monitor multiple Claude Code agents running at the same time, using the new Agent View feature to stay in control without switching between terminal windows.
Step-by-step:
Check your Claude Code version by running
claude --versionin your terminal. Agent View requires v2.1.139 or later. Runnpm install -g @anthropic-ai/claude-codeto update if needed.Open Agent View by running
claude agentsin your terminal. You'll see a dashboard listing all active and background sessions with their current state, such as running, waiting, or done.Dispatch your first background agent by pressing
ninside Agent View and entering your task prompt, or runclaude --background "fix all failing tests in src/api/"from your project directory to start a session that runs without your attention.Monitor progress with the peek panel by pressing
pon any session to open a preview and check its latest output or send a quick reply. PressEnterto fully attach and take control of that session when it needs you.
Pro tip: Use the --worktree flag when dispatching a session to isolate its file edits in a separate git branch, preventing agents from conflicting when they edit overlapping parts of the codebase.
TOGETHER WITH BELAY
Your business has grown. Is your accounting?
If your accounting hasn't kept pace with your business, it's quietly costing you. Outdated financials, no clear view of profitability, and hours lost every week — these are growth bottlenecks, not just bookkeeping headaches. BELAY's Financial Experts handle it all.
TOOLS
Trending AI Tools
🔧 Pareto Code (OpenRouter) - Free experimental coding router that routes requests to the cheapest model that clears your minimum coding score, ranked by coding benchmark
🎨 Lovable Aesthetics - Lovable's new design system for specifying typography, layout, and color preferences while building apps and landing pages
🛡️ OpenAI Daybreak - OpenAI's cyber defense platform powered by GPT-5.5 for vulnerability scanning, patch generation, and secure code review
⚙️ Agent View (Claude Code) - Anthropic's new session dashboard for monitoring and dispatching parallel background agents, available as a research preview
NEWS
What Matters in AI Right Now?
Five publishers and author Scott Turow filed a class-action lawsuit alleging Mark Zuckerberg personally authorized Meta to illegally scrape millions of books to train Llama. Meta said it will fight the suit, citing court precedent on AI training and fair use.
OpenAI launched Daybreak, a cyber defense platform built on GPT-5.5 that helps security teams find vulnerabilities, generate patches, and validate fixes inside development workflows. The platform offers tiered access up to GPT-5.5-Cyber for authorized red teaming and penetration testing.
Startup Andon Labs launched an experimental Stockholm café run by Mona, an AI manager powered by Google's Gemini that handles hiring, supply orders, and daily operations. Human baristas still make the coffee.
Anthropic released Agent View in Claude Code, a research preview feature that shows all active sessions in a single list, making it easier to monitor and manage parallel coding tasks.
Lovable launched Aesthetics, a design upgrade that lets users specify typography, layout, and color preferences while building, generating multiple design concepts for bolder landing pages, apps, and blogs.
OpenAI launched the OpenAI Deployment Company to help businesses build and deploy AI, bringing together 19 investment firms, consultancies, and system integrators. The company is majority-owned and controlled by OpenAI.
🧡 Enjoyed this issue?
🤝 Recommend our newsletter or leave a feedback.
How'd you like today's newsletter?
Cheers, Jason
