Last week Viktor wrote a brief, built a landing page, and opened a pull request.

Last week, Viktor wrote a campaign brief, built a landing page, opened a pull request, generated a board-ready PDF from live Stripe data, and sent a follow-up email to a churned customer. All from Slack. Same colleague that also pulls your reports and monitors your dashboards. 5,700+ teams. 3,000+ integrations.

Start free. $100 in credits →

Today:

• Anthropic Launches Claude Opus 4.8 with Enhanced Agentic Capabilities

• Microsoft Unveils Major Redesign and Workspace Integration for 365 Copilot

• Google Announces General Availability for Nano Banana 2 and Pro Image Models

• Microsoft Prepares to Launch New AI Coding Model Next Week

• Anthropic Closes Massive Series H Funding Round

Anthropic Unveils Claude Opus 4.8: Dynamic Workflows, Fast Mode, and Effort Control

Anthropic has officially launched Claude Opus 4.8, a major model upgrade that delivers sharper judgment, superior agentic skills, and significantly improved reasoning—all while remaining at the exact same price point as Opus 4.7. While Opus 4.8 boasts higher scores across standard coding and knowledge-work benchmarks, its most prominent improvement is its honesty. Early evaluations show the model is four times less likely to confidently hallucinate progress or let code flaws pass without explicitly flagging them to the user.

Alongside the foundation model upgrade, Anthropic launched three groundbreaking features designed to give users unprecedented control over how the AI operates:

• Dynamic Workflows (Claude Code): This feature marks a massive leap for AI-assisted software engineering. Dynamic Workflows allow Claude to break down massive, codebase-scale tasks and orchestrate tens to hundreds of parallel subagents in a single session. These agents can independently tackle problems, review each other's work, and converge on a final, verified result. In one stunning early test, this feature was used to rewrite the Bun runtime from Zig to Rust (roughly 750,000 lines of code) with a 99.8% test pass rate in just 11 days.

• Fast Mode: For developers who need raw speed without sacrificing intelligence, Fast Mode delivers the full capability of Opus 4.8 at 2.5x the output token speed. It is also three times cheaper than fast modes on previous models. It is currently available as a research preview in Claude Code (for users with extra usage enabled) and via an enterprise waitlist on the Claude API.

• Effort Control: Users on claude.ai now have manual control over how much compute Claude expends on a prompt. By adjusting the effort level, users can choose lower settings for rapid responses that conserve rate limits, or dial it up to "extra" or "max" for complex problems, prompting Claude to think deeper and spend more tokens to guarantee a high-quality result.

What’s Next? While Opus 4.8 sets a new operational standard today, Anthropic also teased the imminent arrival of a new, even higher-intelligence tier. "Mythos-class" models—currently being utilized by a select group of organizations for rigorous cybersecurity tasks under Project Glasswing—are expected to roll out to all customers in the coming weeks once advanced cyber safeguards are finalized.

Nano Banana 2 & Pro Go GA: A Leap Forward for Visual Prompting

Google has officially made Nano Banana 2 (Gemini 3.1 Flash Image) and Nano Banana Pro (Gemini 3 Pro Image) generally available. While the models bring enhanced image generation and editing capabilities backed by robust enterprise infrastructure, the standout feature for the creative pipeline is a powerful new preview capability: Nano Banana 2 now accepts video files as input prompts.

By utilizing deep video understanding, the model can analyze specific subjects, actions, and the overarching visual context of raw footage. This allows for the generation of highly context-aware imagery, such as matching custom thumbnails or rich infographics directly to a video's specific environment. For those engineering prompts for hyper-specific visual aesthetics—whether it is the natural lighting and "perfect imperfections" of UGC style, or the tactile felt stitching and squash-and-stretch physics of custom animation—the ability to feed video directly into the model significantly tightens the loop between creative intent and the final generated asset.

Additionally, 1K and 2K output resolutions are now generally available across both models, with 4K generation currently in preview. The models are already being integrated into major media production platforms, allowing creators to anchor their AI generation within controllable, secure 3D pipelines.

Microsoft 365 Copilot: A Redesign Built for Agentic Workflows

Microsoft has rolled out a sweeping redesign for Microsoft 365 Copilot, shifting the tool from a static text box to a fully integrated, task-aware workspace. The update fundamentally changes how Copilot operates across Word, Excel, PowerPoint, and Outlook, transforming it into an active editing partner that works directly alongside you on the document canvas.

The engine driving this progression is "Work IQ," a new intelligence layer that dynamically draws on your broader ecosystem of emails, files, chats, and meetings. Instead of just answering isolated prompts, Work IQ adapts to the complexity of the task at hand. If you are drafting comprehensive project updates, writing instructional scripts, or managing client correspondence, the model grounds its responses in that broader context, unfolding its support in layers as your initial ideas take shape.

Performance has also seen a massive boost. The standalone Copilot app now loads more than twice as fast, and response times for complex prompts have improved by 10%. With a new progressive disclosure interface that keeps the workspace clean, Copilot reveals advanced tools and formatting options only as the work demands them. The result is a faster, highly responsive system designed to keep you in the flow of your work without constantly switching contexts.

🧠RESEARCH

Gamma-World: Generative Multi-Agent World Modeling Beyond Two Players

This paper introduces Gamma-World, a new generative AI model that simulates interactive virtual environments with multiple agents. Unlike previous systems limited to one or two characters, it uses a novel mathematical method to efficiently track and manage numerous independent players interacting in a shared 3D space at real-time speeds. 

Self-Improving Language Models with Bidirectional Evolutionary Search

This paper presents Bidirectional Evolutionary Search, a new method to improve how AI models solve complex problems. Instead of just guessing the next step forward, it simultaneously breaks the main goal down backward into smaller tasks while combining parts of different potential solutions to find the best overall answer. 

ScientistOne: Towards Human-Level Autonomous Research via Chain-of-Evidence

Autonomous AI researchers often write professional-looking papers containing fake citations and unreproducible results. This paper introduces ScientistOne, an AI system that solves this by using a "Chain-of-Evidence" framework. This forces the AI to trace every written claim back to a verifiable source, ensuring complete scientific integrity from discovery to drafting. 

📲SOCIAL MEDIA

🗞️MORE NEWS

Microsoft's Comeback Attempt Microsoft is launching a new artificial intelligence system designed specifically to write computer instructions, or code. This release is seen as a major effort to catch up to rivals and regain leadership in the technology industry. The tool aims to make software creation significantly faster and easier for human workers.

Anthropic's Historic Funding Anthropic recently secured $65 billion in its eighth major round of corporate funding to expand its operations. This massive investment brings the total estimated value of the artificial intelligence company to nearly one trillion dollars. The funds will be used to build smarter and safer computer systems.

Apple's iOS 27 Update Apple's upcoming iOS 27 update will introduce major changes to the iPhone, including a completely rebuilt Siri voice helper. The software also adds a professional-grade camera application and a smarter system for organizing saved pictures on the device. These upgrades use advanced computer programs to make everyday phone tasks much easier for users.

Andon Labs Tests Opus 4.8 Researchers at Andon Labs recently tested the new Claude Opus 4.8 computer model using a custom evaluation program called Vending Bench. This test measures how quickly and accurately the artificial intelligence can solve difficult problems compared to other systems. The early results highlight the updated model's ability to successfully handle highly complex tasks.

The Cost of AI for Big Business Large businesses are spending massive amounts of money on artificial intelligence, but many are struggling to see a clear financial return. The high, ongoing costs of running these advanced computer systems are causing company leaders to carefully rethink their technology budgets. Business owners now want firm proof that this expensive software will actually increase their profits before they spend more.