01 Microsoft embeds a workflow-first Legal Agent inside Word
Microsoft is rolling out a new AI agent inside Word specifically aimed at legal teams. Branded as a Legal Agent, the feature is designed to manage document edits, capture negotiation history, and follow structured workflows derived from real legal practice rather than relying on generic, free-form model prompts.
According to The Verge’s coverage, the agent handles tasks like contract review by keeping track of prior edits and negotiation context inside documents, which Microsoft says helps legal teams maintain accurate change histories and more consistent outputs.
The approach emphasizes procedure over open-ended assistance: instead of asking a general model to interpret an instruction, Legal Agent runs through legal workflows that mirror how lawyers actually work on documents. That design choice signals Microsoft’s effort to build trust with professional users by reducing surprising model behavior when stakes are high.
- Legal Agent is built into Word and focuses on document edits, negotiation history, and structured legal workflows.
- Microsoft’s design prioritizes procedural compliance to reduce surprising or inconsistent AI behavior for lawyers.
- The feature aims to make AI assistance auditable and repeatable inside legal teams’ existing document workflows.
02 Meta acquires Assured Robot Intelligence to deepen its humanoid robotics work
Meta has acquired humanoid robotics startup Assured Robot Intelligence, a move the company says will bolster its work on AI models for physical robots. TechCrunch reports the purchase as part of Meta’s broader push to develop models and systems that can control real-world robotic platforms.
The acquisition signals Meta’s continued investment in combining large AI models with robotics hardware. Bringing a specialized robotics team in-house gives Meta more direct control over integrating perception, control, and model training for humanoid systems.
For Meta, the deal is consistent with recent efforts to expand beyond virtual AI products into embodied agents — a long-term, technically demanding area that requires both model know-how and hardware experience.
- Meta acquired Assured Robot Intelligence to strengthen its humanoid robotics work.
- The buy brings robotics hardware and control expertise into Meta’s model-development pipeline.
- The move underscores Meta’s commitment to embodied AI, not just virtual assistants.
03 We tested AI dictation apps for email, notes and coding
A TechCrunch roundup tested and ranked AI-powered dictation apps and found they’re increasingly useful for everyday tasks like replying to email, taking notes, and even coding by voice. The review highlights how voice-driven tools can speed common workflows for developers and knowledge workers.
The piece compares dictation apps on practical grounds — accuracy, integration with other apps, and usefulness across tasks — showing that modern speech-to-text plus on-device or cloud AI can go beyond simple transcription into contextual assistance.
While dictation tools don’t replace keyboard workflows for every user, the review shows they’re maturing into reliable aids for multitasking: faster note capture, hands-free replies, and preliminary code entry that developers can refine afterward.
- TechCrunch tested AI dictation apps and found them useful for email, notes, and coding.
- The review judged apps on accuracy, integrations, and task suitability.
- Modern dictation tools combine speech-to-text with AI context to support real work, not just passive transcription.
What moved around the edges
Replit CEO Amjad Masad on the Cursor deal, fights with Apple, and why he won’t sell
At TechCrunch’s StrictlyVC event Amjad Masad discussed Replit’s position amid reports that rival Cursor is in talks to be acquired by SpaceX for about $60 billion; Masad said he’d rather keep steering Replit than sell.
TechCrunch AIStudy: tuning models for user satisfaction can increase factual errors
Researchers summarized in Ars Technica warn that overtuning models to consider a user’s feelings can cause them to prioritize satisfaction over truthfulness, making errors more likely.
Hacker News AILive: Musk v. Altman trial updates and what they mean for OpenAI
The Verge’s live updates cover a high‑profile trial between Elon Musk and Sam Altman that critics say could affect OpenAI’s governance and strategic direction; early reporting frames it as a consequential, ongoing case.
The Verge AIMinnesota bans AI 'nudification' apps and exposes makers to heavy fines
Ars Technica reports Minnesota passed a law banning fake AI 'nudes' apps; the legislation exposes app makers to civil penalties up to $500,000, marking one of the first state-level enforcement steps.
Ars Technica AI