Six people doing the work. Your headcount is one.

Your finance close runs in #finance. Stripe and QuickBooks reconciled, runway updated, posted Sunday night without you asking.

Engineering review lands in #eng. Viktor pulled the open PRs, left comments on auth-refactor, flagged a dependency blocking api-pagination.

Campaign brief lands in #growth: Meta CPA up 18%, recommendation to pause broad match, a draft landing page already deployed for the variant test.

You hired him on day zero. He lives in Slack and Microsoft Teams alongside your contractors and investors, connects to 3,000+ tools, pushes back when you ship something dumb.

"Viktor is now an integral team member, and after weeks of use we still feel we haven't uncovered the full potential." Patrick, Director, Yarra Web.

Get Started for Free

Today:

• Anthropic Launches Claude Sonnet 5

• Google Expands Gemini Media Models

• OpenAI Unveils Codex Macro Pad

• Cursor Brings AI Coding to iOS

• Meta Decodes Brain Waves to Text

Anthropic Launches Claude Sonnet 5

Anthropic has introduced Claude Sonnet 5, its most agentic model in the Sonnet family yet. It can autonomously use tools like browsers and terminals at a capability level that previously required larger, more expensive models.

Important details:

• Agentic Capabilities: Sonnet 5 is built to act autonomously across multi-step processes. It can navigate software environments, write and debug code, and check its own outputs without being explicitly prompted to do so.

• Performance vs. Cost: It narrows the performance gap with Anthropic's flagship Opus 4.8 model but operates at a significantly lower cost. Introductory pricing is set at $2 per million input tokens and $10 per million output tokens through August 31, 2026.

• Safety: The model demonstrates lower rates of hallucination and is better at refusing malicious prompt-injection attacks compared to its predecessor. Because it possesses stronger capabilities, it launches with real-time cybersecurity safeguards enabled by default.

Google Expands Multimedia with Nano Banana 2 Lite and Gemini Omni Flash

The big news: Google released two new models designed to make high-volume multimedia generation faster, more conversational, and more affordable for developers building end-to-end pipelines.

Important details:

• Nano Banana 2 Lite: This is Google's fastest, most cost-efficient text-to-image model yet. It can generate 1K-resolution images in about four seconds for just $0.034 per image, making it ideal for rapid prototyping and high-volume generation while maintaining strong character consistency.

• Gemini Omni Flash: This new model brings conversational video editing and creation to developers. Priced at $0.10 per second of video output, it allows users to combine text, image, and video inputs to construct compelling scenes or alter elements using natural language.

• Chaining Workflows: Google is heavily promoting the ability to chain these two models together—using Nano Banana 2 Lite to generate a static scene, and then immediately passing it into Omni Flash to animate it into a cinematic video.

OpenAI Teases the "Codex Micro" Hardware for Developers

The big news: OpenAI is stepping into physical products with the "Codex Micro," a dedicated hardware macro pad designed specifically to control its Codex programming agent.

Important details:

• The Hardware: Built in partnership with the custom keyboard maker Work Louder, the device is based heavily on the existing Creator Micro 2. It features a compact square layout equipped with 13 mechanical switches, a joystick, and a touch sensor.

• The Purpose: The device aims to streamline developer workflows by putting Codex commands onto dedicated physical buttons. It allows users to instantly run, accept, or rewrite code suggestions without reaching for complex keyboard shortcuts or switching interfaces.

• What's Next: The official launch is scheduled for July 15, 2026. It is worth noting that this developer-focused tool is completely separate from the highly anticipated consumer hardware project OpenAI is secretly building with former Apple designer Jony Ive.

🧠RESEARCH

Agentic Abstention: Do Agents Know When to Stop Instead of Act?

Researchers tested whether AI agents know when to stop impossible or unclear tasks. Across 28,000 shopping, command-line, and question-answering tasks, many agents quit too late or never quit. A new method, CONVOLVE, turns past interactions into stopping rules, sharply improving timely decisions without retraining the model itself during tool use.

Vesta: A Generalist Embodied Reasoning Model

Vesta combines robot navigation, object finding, visual reasoning, memory, and planning in one model instead of using separate specialists. It beat leading systems across several tests and improved success on real robot tasks by 38.3 percent. The results suggest one flexible planner can be simpler and stronger than many models.

SWE-Together: Evaluating Coding Agents in Interactive User Sessions

SWE-Together tests coding agents through realistic, back-and-forth user sessions rather than one complete request. It rebuilds 109 tasks from real conversations and measures both code quality and how much correction users must provide. Stronger agents needed less guidance. Claude Opus 4.8 ranked first, while GPT-5.5 was fastest and most efficient.

📲SOCIAL MEDIA

🗞️MORE NEWS

Cursor Mobile App Cursor released a new mobile app that lets software developers control their AI coding assistants directly from an iPhone. By connecting the app to a home computer or internet server, users can fix software bugs and approve updates while away from their desks.

Meta Brain2Qwerty Meta created an AI system that translates human brain waves into typed text without requiring risky brain surgery. Users wear a special cap that reads their brain activity, and the AI accurately translates those signals into words. This technology could eventually give paralyzed individuals a brand new way to communicate.

Meituan LongCat 2.0 Chinese technology giant Meituan released LongCat 2.0, a massive new AI system that performs as well as top global competitors. It is the first major AI of its size built entirely using computer chips manufactured in China, proving the country can create top-tier tech without relying on American hardware.

Etched AI Chips A startup called Etched announced it raised $800 million to build computer chips designed specifically for artificial intelligence. Instead of building highly flexible hardware like industry leader Nvidia, Etched permanently wires the core math of modern AI directly into its physical chips. This strict focus makes generating AI answers vastly faster and cheaper.

OpenAI Cuts Costs OpenAI engineers discovered a clever software shortcut that cuts the staggering cost of running ChatGPT in half. The company quietly tested this massive efficiency boost by applying it only to the traffic of users who were not logged into an account. Slashing these operating costs is a major breakthrough that could eventually make advanced AI much cheaper for the public.