In partnership with
Hello from The AI Night,
Today in AI:
Anthropic Launches Dynamic Workflows in Claude Code
Microsoft Redesigns Copilot App Across 365 Apps
Cursor Launches Auto-review Run Mode for Agents
Here's the deal: Anthropic launched dynamic workflows in Claude Code, a research preview that lets Claude break large tasks into subtasks and run tens to hundreds of parallel subagents in one session. It targets work that normally spans weeks, like codebase-wide bug hunts and large migrations.
The Breakdown:
Claude writes orchestration scripts dynamically, fanning subtasks across parallel subagents and verifying each result before folding it in.
Available now in the CLI, Desktop, and VS Code extension for Max, Team, and Enterprise plans.
Also live on the Claude API, Amazon Bedrock, Vertex AI, and Microsoft Foundry.
A new ultracode setting sets effort to xhigh and lets Claude decide when to trigger a workflow.
Progress saves mid-run, so interrupted jobs resume instead of restarting.
Jarred Sumner used it to port Bun from Zig to Rust, 750,000 lines, 99.8% test pass rate, eleven days.
The bigger picture: Seven hundred fifty thousand lines ported from Zig to Rust in eleven days at 99.8% test pass rate. That is not a feature demo. That is a before-and-after that makes engineering managers rethink headcount. Every company selling multi-agent orchestration as a product just watched Anthropic ship it as a setting toggle inside an existing tool.
Here's the deal: Microsoft unveiled a full redesign of Microsoft 365 Copilot, reworking both the standalone Copilot app and how Copilot appears inside Word, Excel, PowerPoint, and Outlook. The update turns the prompt line into a task-aware workspace and adds a single, consistent entry point for Copilot across all Microsoft 365 apps.
The Breakdown:
The Copilot app now loads over 50% faster, with complex chat response times improved by 10% at the 95th percentile.
A new intelligence layer, Work IQ, draws on emails, files, chats, and meetings to adjust reasoning depth and lets users switch between AI models.
Capability-focused agents include Designer, Researcher, Word, Excel, and PowerPoint, each tuned to specific tasks.
Copilot can now edit documents directly from a side pane or be invoked inline within a paragraph, cell, or slide.
In-app usage rose 27% in Word, 33% in Excel, 43% in PowerPoint, and 30% in Outlook after rollout.
The bigger picture: Twenty-seven percent more usage in Word. Forty-three percent in PowerPoint. Microsoft didn't ship a smarter model. It shipped a faster interface with better placement. Users don't abandon AI because it is bad. They abandon it because reaching it takes one click too many. Microsoft just proved that making AI invisible matters more than making it intelligent.
Here's the deal: Cursor shipped Auto-review, a new run mode in version 3.6 that lets its coding agent run longer with fewer approval prompts. It governs how the agent handles Shell, MCP, and Fetch tool calls by routing them through allowlists, a sandbox, or a classifier subagent.
The Breakdown:
Auto-review applies specifically to Shell, MCP, and Fetch tool calls made by the agent.
Allowlisted calls execute immediately without prompting the user.
Calls that can be sandboxed run inside the sandbox for safer execution.
All other actions route to a classifier subagent that decides the next step.
The classifier can allow the call, try a different approach, or ask the user for approval.
Run mode is configured under Settings, Cursor Settings, Agents, Run Mode.
Users can steer the classifier with custom instructions.
The bigger picture: Every approval prompt breaks a developer's flow. Removing approvals entirely is how agents delete production databases. Cursor built the middle ground. Safe commands run freely. Risky ones get sandboxed. Everything else hits a classifier. Claude Code and Codex still force developers to choose between babysitting and blind trust. Cursor just shipped the third option nobody else has.
Your prompts are leaving out 80% of what you're thinking.
When you type a prompt, you summarize. When you speak one, you explain. Wispr Flow captures your full reasoning — constraints, edge cases, examples, tone — and turns it into clean, structured text you paste into ChatGPT, Claude, or any AI tool. The difference shows up immediately. More context in, fewer follow-ups out.
89% of messages sent with zero edits. Used by teams at OpenAI, Vercel, and Clay. Try Wispr Flow free — works on Mac, Windows, and iPhone.
What else you need to know:
Shopify integrated Perplexity's Computer agent into store management, letting merchants run market research, product image generation, and theme design in parallel from a single agentic workflow.
Google made Nano Banana 2 (Gemini 3.1 Flash Image) and Nano Banana Pro (Gemini 3 Pro Image) generally available via Gemini Enterprise Agent Platform, adding preview support for video files as input prompts.
Figma launched new Figma Make capabilities letting designers connect to a production codebase and make direct visual edits, with annotations, Git branching, commits, and PR creation, in a limited beta starting May 28.
Linear launched Diffs, a native code review tool letting developers review GitHub pull requests inside Linear with real-time agent iteration, guided reviews for large PRs, and bidirectional GitHub sync.
Mistral rebranded Le Chat as Vibe, a unified agent handling long-running work tasks and coding across its web app, a new VS Code extension, and the CLI, starting at $14.99 monthly.
That’s it for today’s edition of The AI Night.
Our goal is to cut through the noise, surface what actually changed, and explain why it matters.
2 ways to support us:
Forward this to your AI-curious friend → https://www.theainight.com
Sponsor The AI Night and reach 500+ AI builders daily → passionfroot.me/theainight