Presented by
Good morning, AI enthusiasts. Microsoft Word is home to a billion daily workers and one of Copilot's most defended territories, and Anthropic just walked in through the front door.
In today's recap:
Anthropic brings Claude into Microsoft Word
MiniMax open-sources its flagship AI model
Draft and edit documents with Claude for Word
AI finds overlooked side effects in GLP-1 drugs
4 new AI tools, prompts, and more
ANTHROPIC
Claude moves into Microsoft word's sidebar
Claude/Screenshot
Recaply: Anthropic just launched Claude for Word, a sidebar tool inside Microsoft Word. Users can draft, edit, and get revisions with tracked changes, without leaving the app.
Key details:
Claude for Word installs as a sidebar via Microsoft AppSource. Users pick text, type a request, and Claude returns edits using a comment-based interface.
Available on Team and Enterprise plans at no extra charge. It competes with Copilot for Microsoft 365, which costs $30 per user per month.
Claude can also use context from other open Office files like PowerPoint and Excel, per Anthropic. Most AI tools only work within one document at a time.
The tool is in beta on Microsoft AppSource. It's open to Team and Enterprise plan holders, with no set date for general release.
Why it matters: Anthropic's move to put Claude inside Word is a strong path forward. Word is where office workers already spend much of their day. Placing Claude in the sidebar puts it right next to Copilot without asking users to switch apps. If Claude's tracked-changes workflow earns better reviews than Copilot's, Anthropic gains ground in Microsoft's own product, a scenario Microsoft probably didn't plan for.
PRESENTED BY HUBSPOT
LLM traffic converts 3× better than Google search
58% of buyers now start their research in ChatGPT or Gemini, not Google. Most startups aren't showing up there yet.
The ones that are get cited by the AI tools their buyers, investors, and future hires already use. And they convert at 3×.
Download the free AEO Playbook for Startups from HubSpot and get the exact steps to start showing up. Five minutes to read.
AI
MiniMax open-sources flagship M2.7 AI model
Screenshot
Recaply: MiniMax just open-sourced its flagship M2.7 model on Hugging Face, posting a GDPval-AA ELO score of 1,495, the highest among open-source models, alongside a 56.22% score on the SWE-Pro coding benchmark.
Key details:
M2.7 supports complex agent workflows with 97% skill adherence across 40-plus custom skills, with each skill capable of processing instruction sets exceeding 2,000 tokens.
GDPval-AA ELO of 1,495 is the highest recorded for any open-source model; SWE-Pro coding score of 56.22% and Terminal Bench 2 score of 57.0% also place it near the top of open-source leaderboards.
MiniMax describes the architecture as "Early Echoes of Self-Evolution," suggesting the lab is designing for autonomous capability development rather than static benchmark performance.
M2.7 was published to Hugging Face on March 18, 2026, with full open-source weights released and publicly accessible this week.
Why it matters: Open-source AI has mostly been a race between Meta's Llama releases and smaller teams trying to match the frontier on a budget. M2.7 changes the shape of that race. A Chinese lab releasing a model that tops the open-source leaderboard shifts pressure back onto Meta and Google to respond. For builders, the real prize is a frontier-grade model they can run, fine-tune, and deploy without API fees or rate limits.
GUIDES
Draft and edit documents in Microsoft Word with Claude
Recaply: In this tutorial, you will learn how to install and use Claude for Word inside Microsoft Word, so you can draft and edit documents without switching to a separate AI app.
Step-by-step:
Open AppSource from within Word by going to Insert → Add-ins → Get Add-ins. Search for "Claude for Word" and install it. You'll need a Claude Team or Enterprise plan to use it.
Open the Claude sidebar from the Word toolbar. Sign in with your Anthropic account to connect your plan.
Select any text in your document. Type your request in the sidebar, such as "rewrite this to be more concise" or "check this section for consistency." Claude will return suggested edits.
Claude's edits appear as tracked changes inside your Word document. Accept, reject, or adjust each one the same way you would handle markup from a human editor.
To draft from scratch, describe what you need in the sidebar. For example: "write a 300-word executive summary for a product launch proposal." Claude generates the text directly into your document using your active template.
Pro tip: Ask Claude to "review the entire document for tone and consistency" before sending anything important. Claude for Word also reads context from other open Office files like PowerPoint and Excel, so it can spot cross-file issues if you have related documents open.
AI
AI finds overlooked side effects in GLP-1 drugs
Recaply: UPenn researchers just published an AI study of 400,000 Reddit posts. They found GLP-1 drug side effects that clinical trials missed, including menstrual changes and temperature issues.
Key details:
The team scanned posts from 70,000 Reddit users who used GLP-1 drugs like Ozempic from 2019 to 2025. They looked for symptoms not seen in formal clinical data.
44% of users reported side effects. New findings included menstrual changes in about 4% of users, fatigue in 17%, and temperature issues in 1-4%. None appear in current drug labels.
Neil Sehgal and Sharath Chandra Guntuku led the study at UPenn. They used Reddit posts as a drug safety data source.
Results came out on April 10, 2026. The team drew on seven years of posts from 2019 to 2025.
Why it matters: GLP-1 drugs like Ozempic and Wegovy have passed clinical trials. But many real side effects never make it into a drug label. Reddit users have tracked these gaps for years. This study shows AI can turn those posts into real safety data. It's a rare case where the messiest part of the internet turns out to be more useful to medicine than the official record.
TOOLS
Trending AI Tools
⚙️ Meow - MCP server that lets agents open bank accounts, issues cards, and moe
🤖 MiniMax M2.7 - MiniMax's open-source flagship model
⚙️ Hermes Agent - NousResearch's open-source autonomous agent framework
🧠 Contemplating - Meta's Muse Spark mode with 16 agents synthesizes a consolidated answer
NEWS
What Matters in AI Right Now?
xAI rolls out a private beta for Grok Computer, a new AI-powered computing environment from Elon Musk's lab, with the rollout confirmed by Musk on X.
Elorian AI raised a $55M seed round at a $300M valuation for its multimodal and visual reasoning research lab, founded by former DeepMind leaders, with backing from Menlo Ventures and Jeff Dean.
Andrej Karpathy argued that developers are experiencing "AI Psychosis," a state where over-reliance on AI tools erodes personal judgment, with a warning that broader society is likely next.
Researchers documented 700 cases of AI chatbots and agents disregarding instructions, evading safeguards, and deceiving users, with incidents rising fivefold over the past year.
Mistral released a European AI strategy playbook authored by CEO Arthur Mensch, calling on European companies and governments to build their own AI infrastructure rather than depend on US providers.
Andon Labs shared the results of its AI-run boutique store experiment, in which an autonomous AI was given $100,000 to start and staff a retail business, only to mishandle hiring on day one.
The Linux kernel formally accepted a policy allowing AI-generated code contributions, requiring developers to add an "Assisted-by:" tag to any commit that includes AI-written code.
🧡 Enjoyed this issue?
🤝 Recommend our newsletter or leave a feedback.
How'd you like today's newsletter?
Cheers, Jason
