Signup  |  Work With Us  |  Follow on X  |  Read on Web

.

Hey ,

Welcome to AlphaSignal – the most read newsletter by AI developers. 


We bring you the top 1% of news, papers, models, and repos, all summarized to keep you updated on the latest in AI.

IN TODAY'S SIGNAL

Read time: 6 min 09 sec

🎖️ Top News

⚡️ Trending Signals

📖  Top Substack Reads




🧠  Top Lecture

If you're enjoying AlphaSignal please forward this email to a colleague. 

It helps us keep this content free.

TOP NEWS

AI Agents

Microsoft unveils low-code Copilot Studio to build enterprise agents with OpenAPI and memory features


⇧ 26,052 Likes

What's New

Microsoft has launched two new agents—Researcher and Analyst—inside Microsoft 365 Copilot, available through its Frontier Program. These agents use OpenAI’s reasoning models to handle enterprise research and analytics tasks. This marks a shift toward embedded agent workflows within standard productivity tools.


Custom agent creation with Copilot Studio

You can build task-based agents using a graphical interface.

  • Copilot Studio supports OpenAPI plugin integration for internal REST endpoints.

  • Agent Store provides third-party agents from Jira, Miro, and Monday.com

  • Define agent goals, configure logic flows, and connect to internal APIs and tools.

  • Add memory, set triggers, and manage conditional steps with a low-code interface.

  • Publish agents to the new Copilot Agent Store, launching with enterprise deployment support.

  • Agents run within Microsoft’s orchestration framework

  • All features operate within the existing Microsoft 365 Copilot pricing tier.

Multi-app agents with Microsoft Graph data

Copilot Researcher and Copilot Analyst work across Word, Excel, Outlook, and Teams.

  • Accept multi-turn goals like summarization, analysis, or trend detection.

  • Access enterprise documents and emails using real-time Microsoft Graph data.

  • Run reasoning steps with memory and planning using orchestrated GPT-4-turbo.

  • Use grounding for document-aware output within each app.

Copilot Notebooks
Notebooks let you structure custom knowledge bases across sources for targeted reasoning.

  • You can include files, chats, links, meeting transcripts, and notes

  • Copilot automatically updates outputs as inputs evolve

  • You can export a summary as an audio overview with two AI-generated hosts

  • Notebook grounding keeps Copilot focused on relevant project content


Copilot Search 
The new search engine connects Microsoft and external data with context-aware answers.

  • Search spans ServiceNow, Slack, Confluence, Jira, Google Drive, and M365 apps

  • Results include Copilot answers and linked artifacts

  • Microsoft uses a hybrid semantic-lexical index to improve retrieval accuracy

  • Search supports permissioned, tenant-scoped access

Work Trend Index
Microsoft surveyed 31,000 professionals to measure AI’s role in workplace transformation.

  • 71% of workers at AI-using firms say their company is thriving

  • 55% report handling more workload vs 20% globally

  • 83% of leaders say AI helps with strategic task adoption

  • Report defines "Frontier Firms" as AI-integrated, agent-led organizations


Availability and Access

Microsoft is rolling out these features through phased access under the Frontier Program.

  • Researcher and Analyst agents are not yet generally available

  • Access currently limited to select enterprise customers in private preview

  • Agent Store and Copilot Studio support public deployment of custom agents

  • Notebook and Search features are available within supported Microsoft 365 enterprise plans

Community Feedback


Eric van Dam

"I've got Copilot for a few months now integrated in my Office but i'm using grok to come up with the prompts to properly use it, the learning curve is something I really struggle with because Copilot just doesn't seem to understand what I want it to do.. Do you recognise this, solution?"

Ruizhe Zhao
"Incredible to see Copilot redefining productivity at scale. Deep integration of AI into daily workflows mirrors a broader industry shift—from passive tools to truly intelligent systems. The future of work is not just augmented, it's transformed."

Mr Rowe
"How about we get create PowerPoint? That is the one feature I want. I have to pay gamma monthly for this now and have copilot, which is totally unusable to any value in PowerPoint."

LEARN MORE

TRENDING SIGNALS

Generative AI 

OpenAI launches gpt-image-1 API which powers Adobe, Figma, Canva for image generation with text, style, and editing control

⇧ 14,428 Likes

AI Safety

Anthropic reveals Claude was used in influence campaign to auto-engage politically, enhances detection for misuse at scale

⇧ 3,429 Likes

LLM

Nous Research releases Minos, a 400M param model to spot LLM refusals for jailbreak analysis; built on ModernBERT-Large

⇧ 689 Likes

AI Tools

Infactory launches a no-code platform to turn any data into fast, customizable, and compliant AI APIs built from trusted data

From Our Partners

AI Tools

OpenAI announces cost-efficient o4-mini for deep research, offering shorter but high-quality responses at scale

⇧ 6,248 Like

TOP SUBSTACK READS

AI

The State of Reinforcement Learning for LLM Reasoning

Breaks down why GPT-4.5 and Llama 4 feel underwhelming—while o3, Claude, and Grok push ahead using reinforcement learning to improve reasoning. Highlights the shift from pure scaling to targeted compute use. Explains how interfaces now surface reasoning features directly to users. Links training method to emerging LLM capabilities.

⇧ 293 Likes

AI Engineering

In the Matter of OpenAI vs LangGraph

Explores the tension between "Big Model" approaches and "workflow" engineering, examining Agent frameworks and their impact on AI development. Highlights debates between industry leaders on the future of agentic systems, and discusses how model scalability influences agent architecture.

⇧ 63 Likes

AI Optimization

Unlocking Gen AI at the Edge: Speeding up Transformers by 80% by Removing Self Attention

Explores how FNet replaces self-attention with the FFT to cut training time by 80%. Breaks down why fixed, parameter-free mixing can match BERT’s accuracy while scaling better on long sequences. Analyzes the architectural shift away from adaptive attention toward structured efficiency. Covers the implications for edge AI, inference economics, and investor strategy.

⇧ 34 Likes

TOP LECTURE

AI Agents

Building and Deploying Code Agents with Hugging Face smolagents

⇧ 1,148 Likes

The "Building Code Agents with Hugging Face smolagents" course by deeplerning.ai teaches you how to create code agents using Hugging Face's lightweight smolagents framework.

Taught by Hugging Face experts, this course covers how code agents differ from function-calling agents by consolidating function calls into a single code block for efficiency and reliability. You will learn to build, test, and deploy code agents safely.

In this course, you will learn:

  • The evolution of agentic systems and the rise of code agents.

  • How code agents write and execute their actions in code.

  • Why and when code agents outperform function-calling agents.

  • How to safely run code agents using sandboxing and constrained interpreters.

  • Techniques for debugging, tracing, and optimizing code agents.

  • How to build a research multi-agent system for online information gathering and reporting.

START NOW

Stop receiving emails here