Microsoft Launches Agent Boss AI for Workplaces

Hak

28 Apr 2025 — 3 min read

It's Monday, April 28, 2025, and you're reading the Agentive Daily Report.

Busy People's Section

Microsoft launches powerful AI agents that transform the workplace into "agent boss" structures

xAI's Grok now features real-time visual inputs with Grok Vision on iOS devices

Google's research introduces SWiRL, teaching AI models to reason in steps rather than just answer questions

Anthropic begins studying "model welfare" to explore AI consciousness and moral rights

OpenAI prepares to release its first truly open-source model since GPT-2, expected this summer

Today's Top Stories

Microsoft Transforms Workplace with New Copilot AI Agents

Microsoft has rolled out a new class of workplace AI agents that significantly expand Copilot's capabilities beyond meeting summaries. The new Researcher and Analyst agents, embedded in the Microsoft 365 Copilot suite, can pull data from meetings, emails, CRM systems and more to produce business insights, flag risks, and suggest strategies.

Microsoft is positioning Copilot as the "browser for AI" - a central hub where enterprise tools converge. The launch includes an Agent Store, notebook features, and search capabilities that position Copilot as both a knowledge worker and workflow operating system. This marks the beginning of AI-powered organisational structures where human employees become "agent bosses" directing teams of digital teammates.

Grok Vision Brings Real-Time Visual Capabilities to xAI's Chatbot

xAI's Grok chatbot now supports visual inputs on iOS, allowing users to point their phone cameras at objects and receive instant, natural-language feedback about what they're seeing. With Grok Vision, users can snap pictures of forms, signs, or products and get immediate feedback, blurring the line between traditional chatbot and augmented reality assistant.

This push toward mobile-native, multimodal interaction represents a strategic differentiation. While ChatGPT uses Vision primarily for document and image analysis and Gemini emphasizes visual learning, Grok's approach focuses on ambient awareness, making phones the interface and AI the interpretive layer. This real-time capability may finally provide Grok with a clear, everyday use case.

Google's SWiRL Teaches AI to Think Like Humans

Google and Stanford researchers have introduced Step-Wise Reinforcement Learning (SWiRL), a new training approach that teaches AI how to think more like a human. Where large language models are currently good at first-pass answers, they struggle to combine insights or tackle tasks requiring reasoning across multiple steps.

SWiRL doesn't just give AI new knowledge—it teaches it how to learn. By training models to plan, search, and synthesise iteratively, researchers found improvements across categories like math, question answering, and logic. This points toward a more generalizable form of intelligence that could be crucial for building reliable AI agents that can handle complex reasoning tasks.

Fast Forward

Anthropic Explores Model Welfare: Anthropic has launched research into whether future AI systems could develop consciousness, with lead researcher Kyle Fish estimating a 15% chance models could become conscious. This initiative explores frameworks for assessing consciousness and potential signs of AI distress.
OpenAI's New Open-Source Model: OpenAI is developing its first open-source language model since GPT-2, planning to release it by summer 2025. Led by VP of Research Aidan Clark, the model will have a permissive license for commercial use and be synthesise for consumer hardware.
Deep Research for Free Users: OpenAI has introduced a lightweight version of its Deep Research feature for free ChatGPT users, providing up to five queries monthly using the o4-mini model. This offers comparable depth to the standard o3-powered tool with slightly shorter responses.
GPT Image API Launches: Developers can now access GPT-Image-1, previously only available in ChatGPT, through OpenAI's API. The model supports text-to-image generation with precise prompt alignment at a tiered pricing structure based on usage.
Meta's Smart Glasses Translation: Meta's Ray-Ban smart glasses now feature real-time translation capabilities, demonstrating AI's increasing integration with wearable technologies for practical everyday applications.

New Tools Discovered

Zuzia: AI-powered monitoring platform that watches server uptime and SSL status, notifying users only when something actually matters and reducing alert fatigue for small teams.
Genspark: Search engine that builds custom interactive pages from prompts instead of links, using multiple AI models to answer questions and automate multi-step workflows.
optimisedAlter: AI copilot that talks to your Mac to help you get things done quickly through natural language interaction.
Icon: Self-described as the "world's first AI CMO," able to plan, create, and run thousands of advertising campaigns end-to-end.
Jo: Conversational AI that talks directly to users, gathers insights, and helps craft products they'll love based on feedback.

Discover more tools at Agentive.Directory

That's a wrap for today! Thank you for reading this report.

Have thoughts on today's edition? Hit reply and let us know what you're thinking. Or if you've discovered a cool AI tool we should feature, drop us a line.

Until tomorrow,
Hak from Agentive.Studio