Hi {{first_name}}!
This week felt like a clear signal that AI is moving deeper into the tools, workflows, and decisions that shape how businesses actually operate. Microsoft brought Claude into Microsoft 365. Google kept expanding Gemini across Workspace. And underneath the product headlines was a bigger question that matters a lot more than any single feature release:
Are you using AI to simply cut costs, or to build a better business?
That tension is showing up everywhere right now. In the tools getting released. In the way companies are restructuring. And in the choices leaders are making about where AI fits in their organization.
Here’s what we’re covering this week:
New and Noteworthy: Anthropic vs. the Pentagon, Claude inside Microsoft 365, Zoom avatars, and the other updates worth paying attention to
Tool Spotlight: How Gems and NotebookLM can turn Gemini into a repeatable workflow inside Google Workspace
A Bigger Question: When AI shrinks the team, and when it actually helps grow the business
Let’s get into it.
New and Noteworthy
Zoom Is Launching AI Avatars This Month: Zoom’s AI-powered avatars are expected later this month, giving users a photorealistic digital version of themselves for meetings and async video updates. The avatars replicate appearance, facial expressions, lip movement, and eye movement. Alongside that launch, Zoom is also adding deepfake risk detection to flag potential audio or video impersonation during calls, expected in April. Zoom also introduced new AI-first productivity tools called AI Docs, AI Slides, and AI Sheets, arriving in preview this spring and pulling from meeting transcripts and connected app data. The avatar feature has been anticipated for a while, but the deepfake detection pairing is the part worth watching.
Anthropic Is Taking the Pentagon to Court: Anthropic filed two federal lawsuits challenging the Defense Department’s decision to label the company a “supply chain risk,” an extraordinary designation for a U.S. AI company. Anthropic is calling the move unprecedented and unlawful, arguing the designation punishes the company for its views and restrictions around military AI use. The financial stakes are real: Anthropic says the designation could put billions in 2026 revenue at risk. The broader signal is that this fight is becoming bigger than one company, with major implications for how AI labs navigate safety, speech, and defense work.
Microsoft Embeds Claude Directly Into Microsoft 365: For the first time, a non-OpenAI model is being integrated directly into the core Office suite. Microsoft is adding Anthropic's Claude to Microsoft 365 Copilot through a feature called Copilot Cowork, allowing users to delegate multi-step tasks across documents, emails, and spreadsheets. The agents run within Microsoft's existing OneDrive and SharePoint security frameworks, so enterprise data protections stay intact. That matters because a lot of teams won't adopt AI through brand-new tools. They'll feel it through the assistants already built into the software they already use every day.
Anthropic Launches the Claude Partner Network: Anthropic just announced a new Claude Partner Network, backed by an initial $100 million investment for 2026. The program is designed to help businesses find trusted firms that can implement Claude with the right training, support, and real-world expertise. The bigger takeaway is that AI is moving past the novelty stage. Better models matter, but what most companies really need is help applying them in practical ways that save time, improve quality, and create momentum. It’s encouraging to see the ecosystem becoming more structured, and it’s something I’m personally excited to explore further at Ampra as we look at what it could mean to become an official partner.
Google Weaves Gemini Into Docs, Sheets, Slides, and Drive: Google is embedding Gemini across its entire productivity suite, and the integrations go deeper than surface-level autocomplete. Docs can now generate first drafts in your writing style, Sheets can auto-populate tables with live data from Google Search, and Drive can answer complex questions across your files, emails, and calendar in plain language. Slides builds editable, theme-matched decks on demand. All of it is rolling out now in beta for Google AI Ultra and Pro subscribers. If your team lives in Google Workspace, this is worth testing sooner rather than later.
OpenAI Acquires AI Security Startup Promptfoo: OpenAI acquired Promptfoo, a security firm that specializes in testing LLMs for vulnerabilities. The technology will be integrated into OpenAI Frontier, its enterprise agent platform, adding automated red-teaming, compliance monitoring, and protection against prompt injection and jailbreak attempts. The bigger signal here: as AI agents take on more sensitive business tasks, security is becoming table stakes, not an afterthought. Labs are moving to build it in before problems surface at scale.
NVIDIA Releases Nemotron 3 Super: NVIDIA launched a new 120-billion-parameter open model that keeps only 12 billion parameters active at any given time, making it far more efficient to run. The company says it delivers five times the throughput of its predecessor with a one-million-token context window, and it's designed specifically for agentic AI workflows. Open models at this capability level continue to close the gap on frontier proprietary ones, which matters for businesses that want flexibility without ongoing API costs.
ElevenLabs Launches ElevenCreative: ElevenLabs released ElevenCreative, a single platform combining voice cloning, text-to-speech, AI video, music, sound effects, and image generation in one browser-based editor with support for over 70 languages. For content teams producing multilingual or audio-heavy assets, having localization and generation in one place removes a lot of the back-and-forth between separate tools. Worth a look if you're currently stitching together multiple platforms to produce that kind of work.
Meta Acquires AI Agent Social Network Moltbook: Meta bought Moltbook, a Reddit-style platform where AI agents can post, comment, and interact with each other autonomously. The co-founders are joining Meta Superintelligence Labs. The acquisition follows a similar move by OpenAI to bring in the creator of the underlying agent framework. The short version: the major labs are quietly acquiring the infrastructure layer that makes large-scale agent coordination possible, and social platforms may be one of the first places you see it show up.
Tool Spotlight: The Gemini Stack: How Gems and NotebookLM Turn a Chat Tool Into a Repeatable Workflow
Most people treat Gemini like a chat tool. Ask a question, get an answer, move on.
That’s fine, but the real shift happens when you stop thinking in single prompts and start thinking in systems. The setup I keep coming back to, and the one I walk through in the AI Training Series I lead, combines Gemini, Gems, and NotebookLM as three distinct layers working together.
If your team is already in Google Workspace, this can be incredibly powerful.
Before getting into how they connect, here’s what each layer actually does.
Gems are Google’s version of custom AI assistants built to support repetitive tasks. You configure a Gem with a specific role, rules, tone, and output structure, and it shows up the same way every session. No re-explaining your brand voice. No resetting context. It’s a persistent, purpose-built assistant you design once and use repeatedly.
NotebookLM is Google’s AI-powered knowledge and research tool, grounded in your source materials. You can upload documents, reports, PDFs, notes, YouTube videos, and websites, and it works from what you’ve given it. It stays anchored, cites sources and does not pull from the wider web or hallucinate. That constraint is exactly what makes it so useful. Think of it as a private knowledge base your AI can actually reason from.
Here’s how the full system works together.
Gemini is your execution layer: reasoning, brainstorming, drafting, and generating.
Gems are your process layer: you configure them once with your standards, and they hold that pattern across repetitive tasks.
NotebookLM is your grounding layer: it works from your materials, which helps keep outputs aligned with what is actually true about your business.
A practical example: say you run a financial advisory practice. Your NotebookLM contains your firm’s positioning documents, service descriptions, approved messaging, and compliance boundaries. Your Gem is configured as a content assistant that knows your tone, your audience, and your preferred output formats. Every time you need a blog post or webinar outline, you are not starting from scratch. You are running a new task through the same trusted system.
Another example: say you run a home services business with a small office team. Your NotebookLM holds your SOPs, service descriptions, pricing guidelines, past proposals, and common customer questions. Your Gem is configured as an operations assistant that knows how to turn rough notes into polished estimates, follow-up emails, onboarding checklists, or internal SOP drafts. Instead of reinventing the wheel every time, your team can work from a repeatable setup that saves time, improves consistency, and reduces bottlenecks.
The knowledge stays consistent. The process stays consistent. Only the output changes.
The real work is deciding what should stay stable and what should stay flexible in your situation. That is the difference between simply using AI and actually designing with it. And if Google Workspace is already where your team works, the infrastructure is already there. You just have to use it.
When AI Shrinks the Team vs. When It Grows the Business
By now you've probably seen Jack Dorsey's note to employees at Block. The company is reducing its workforce by nearly half, from over 10,000 people to just under 6,000. The reason given wasn't financial trouble. The business is growing. The reason was AI.
Dorsey's take is that smaller, flatter teams powered by AI tools represent a new way of building and running a company, and that this shift is accelerating fast.
To his credit, he didn't just cut people and move on. Departing employees are receiving 20 weeks of salary plus one additional week for every year of tenure, equity vested through the end of May, six months of healthcare, their corporate devices, and $5,000 to support their transition.
This isn't the first time we've seen this play out. A few editions back I mentioned that Salesforce publicly admitted to overestimating what AI could handle after laying off 4,000 customer support staff. They reduced their support team from 9,000 to 5,000 expecting AI to cover the gap. It couldn't. Automated systems struggled with complex issues, customer experience took a hit, and the company was left trying to rebuild what it had dismantled. That's the version of this story nobody talks about when they're announcing the cuts.
The businesses getting the most out of AI right now are not the ones treating it as a cost-cutting tool. They're the ones using it to make every single person on their team more capable, more responsive, and more effective at the work that drives the business. The output per person goes up. The quality goes up. The speed goes up. That's a competitive advantage that compounds over time in a way that a leaner headcount simply can't replicate.
Trust is built through relationships. Culture is built through people who feel invested in something. Customers notice when the humans they're dealing with are stretched too thin or disengaged.
AI should be the reason your team feels less burned out and more capable of doing the work that matters most. Not the reason they're looking over their shoulder.
That's the version of this future I'm trying to help organizations build toward, and it's the one I hope wins.
If there's one thing I want you to take away from this edition, it's that the companies winning with AI right now are not the ones using it to cut corners or reduce headcount. They're the ones using it to make every person on their team sharper, faster, and better at the work that actually drives the business forward.
If something in this edition sparked an idea or a question, hit reply and let me know. I read every response and it shapes what we cover next.
See you next week.
Julien
PS: If this was valuable, please forward it to a friend or colleague. They can subscribe and catch up on previous editions at www.ampra.ai/join-our-newsletter.