Hi {{first_name}}!
I know I say this a lot, but this was easily one of the most pivotal weeks in AI history. We are witnessing a massive structural shift: the "Big Three" are no longer just building better chatbots; they are building the infrastructure for autonomous agents and private, local execution.
Anthropic’s revenue just tripled. Google released a model powerful enough to help run your business from a laptop without an internet connection. And the New York Times profiled a founder who built a $400 million business with AI and just two employees... though the full story has some layers most people are skipping.
This week we’re breaking down:
Privacy First: Why Google’s Gemma 4 is a "locked room" for your data.
Agent Scale: Anthropic makes building AI teams plug-and-play.
The Medvi Breakdown: What a $400M "two-person" company actually means for your strategy.
Cowork vs. Code: Which Claude tool should you actually use?
OK, let’s dive in!
New and Noteworthy
Anthropic launches Claude Managed Agents for businesses: Anthropic just made it significantly easier to build and deploy AI agents at scale. Claude Managed Agents, now in public beta, handles all the behind the scenes infrastructure like sandboxed execution, authentication, and error recovery. Early adopters like Notion and Asana are using it to let knowledge workers automate complex workflows in days rather than months. If you have been curious about AI agents but didn't know where to start, the on-ramp just got much shorter.
Google releases Gemma 4: High-performance AI that stays on your hardware: Google launched Gemma 4, which is a family of open models built on the Gemini 3 architecture. The big story here is the move to an Apache 2.0 license. This means you can use, modify, and deploy these models commercially with no restrictions. Why this matters for your business: If you have sensitive client data or internal IP that you are hesitant to send to the cloud, Gemma 4 is a massive win. It is small enough to run on a standard modern laptop but powerful enough to handle 90% of daily tasks like drafting, summarizing and analyzing. You get the intelligence of a frontier model with the security of a locked room. This is a signal that the gap between 'rented' cloud AI and 'owned' local AI is closing fast. This exact topic is something I plan to dive deeper into next week!
Z.AI releases GLM-5.1: The new open source coding king: The Chinese lab Z.AI (formerly Zhipu AI) released GLM-5.1, an open source coding model that scored 58.4 on the SWE-Bench Pro benchmark. This actually outperforms GPT-5.4 and Claude Opus 4.6 on that specific test. What makes it different is endurance: Z.AI built it for long-horizon work, meaning it can stay on a single coding task autonomously for up to eight hours. In one demo, it built a working Linux desktop web app from scratch with no human guidance. The model is available under the MIT license and was trained entirely on Huawei chips.
Anthropic's revenue tripled to $30 billion and it just locked in a massive compute deal: Anthropic announced its run-rate revenue has surpassed $30 billion, up from $9 billion at the end of 2025. To keep up with demand, they signed a deal for 3.5 gigawatts of compute capacity. For context, that is roughly the power output of three nuclear power plants. Whether you use Claude directly or not, this signals that the infrastructure behind AI is scaling fast. The biggest companies in the world are betting that business adoption is about to accelerate even further.
Microsoft launches its own AI models to compete with OpenAI: Microsoft just released three new in-house AI models under its new MAI brand: MAI-Transcribe-1, MAI-Voice-1, and MAI-Image-2. These are separate from their partnership with OpenAI and signal that Microsoft is building its own independent AI stack. The transcription model is particularly impressive: it is priced at just $0.36 per hour and is 2.5 times faster than their previous offerings. If you use Teams or Copilot, expect a massive jump in transcription accuracy for noisy environments very soon.
Meta unveils Muse Spark from its new Superintelligence Labs: Meta launched Muse Spark, the first model built by its new division led by Alexandr Wang. It includes a specialized "shopping mode" that combines AI reasoning with user behavior to act as a personal concierge. Interestingly, Meta is keeping this model closed source, which is a major departure from their previous open-source Llama strategy. If your business relies on Instagram or WhatsApp for customer engagement, the AI assistants inside those apps are about to get significantly more capable.
Anthropic keeps Claude Mythos private for cybersecurity defense: Anthropic just announced Claude Mythos Preview, a new model that possesses a level of coding capability that surpasses all but the most skilled humans. During internal testing, Mythos autonomously discovered thousands of high-severity vulnerabilities across every major operating system and web browser. Rather than releasing it publicly, Anthropic launched Project Glasswing: a $100 million initiative with 12 major partners including AWS, Apple, Google, Microsoft, and CrowdStrike to patch critical software before bad actors find the same flaws.
Why this is a landmark moment: The vulnerabilities Mythos spotted had, in some cases, survived decades of human review and millions of automated security tests. It found a 27 year old flaw in OpenBSD, which is an operating system renowned for being one of the most security-hardened in the world. It also discovered a 16 year old vulnerability in FFmpeg, a media library used by billions of people, in a line of code that traditional testing tools had hit 5 million times without ever catching the problem.
The Bigger Picture: This represents a massive stepchange in capability. Just a few months after we adjusted to the power of the Opus 4.6 and ChatGPT 5.4 model generation”, we are seeing another leap in autonomous reasoning and "cybersecurity" capabilities. We will dive deeper into why this matters later in the newsletter, but for now, recognize that the gap between "AI as a tool" and "AI as an autonomous agent" is closing faster than anyone predicted.
OpenAI proposes a Public Wealth Fund and a four-day workweek: OpenAI published a policy paper calling for a national fund that gives citizens a stake in AI-driven growth. They also proposed shifting the tax burden from labor to capital, including potential taxes on robots, and supporting a subsidized four-day workweek. For business owners, this is a look at the long-term policy shifts the leaders in this space are expecting.
The Writers Guild locks in stronger AI protections: The WGA approved a new agreement that includes strict protections against studios training AI on writers' materials. At the same time, the AFL-CIO reported that AI accounted for a quarter of all layoffs in March. This is a clear signal that the legal and labor frameworks around AI-generated content are tightening. If you use AI for marketing or client-facing materials, staying aware of these boundaries is a smart move.
Google quietly launches an offline AI dictation app on iOS: Google released a free, offline dictation app for iPhone called AI Edge Eloquent. It transcribes speech in real time, automatically filters out filler words, and polishes your text into professional drafts. Because it runs locally on your device, your audio and confidential conversations never leave your phone. It is a fantastic, privacy-first alternative to Wispr Flow, which has been one of my favorite tools. I’ll be testing one out and reporting back soon!
A Two-Person Company Hit $401 Million in Year One.
You have likely seen this story floating around LinkedIn and Twitter this week. The New York Times profiled a founder named Matthew Gallagher who used AI tools and $20,000 to build Medvi, which is a telehealth company selling weight-loss drugs. He claims it will be a $1.8 billion business this year.
Two full-time employees. No investors. No office.
He used ChatGPT, Claude, and Grok to write code. He used Midjourney and Runway to create ad content. ElevenLabs and custom AI agents handled his customer service. The entire medical side: doctors, prescriptions, pharmacy, and shipping was outsourced to partners. He simply pointed AI-generated ads at the checkout page and scaled.
The numbers are hard to ignore. Medvi reportedly brought in $401 million in its first year with a 16.2% net profit margin. For context, Hims and Hers reported $2.4 billion in revenue last year with 2,442 employees. Sam Altman predicted that a one-person billion-dollar company would happen soon. It looks like he may have won that bet.
The part most people are skipping
There is a real lesson in this story, but it is not the one most people are sharing. Six weeks before the profile ran, the FDA sent Medvi a warning letter for misbranding the compounded drugs generating that revenue. The agency found that Medvi’s website implied it was the actual manufacturer of the drugs it sold. Separately, a class action lawsuit was filed alleging deceptive spam emails through affiliate marketers.
Additionally, the company’s AI customer service chatbot fabricated drug prices and hallucinated product lines that did not exist. Both required manual fixes. When you are a two-person operation running AI at scale, you are the only safety net when something breaks.
The Signal for You
AI is exactly why this story was possible. Full stop. Two people do not build a $400 million business in a year without it. The tools Gallagher used are available to all of us right now. They genuinely compressed what used to take years and a full team into months and a few laptops.
But AI alone did not make this a billion-dollar story. Gallagher made strategic decisions:
He picked a high-demand market: Weight-loss medications are currently the most sought-after products in healthcare.
He leveraged a regulatory window: He utilized a temporary allowance for compounded versions of expensive brand-name medications.
He outsourced everything: He focused entirely on growth and rented the rest.
AI is the multiplier. It multiplies whatever you point it at. If you point it at a strong strategy with real demand, you get what Gallagher got. If you point it at a weak strategy, you just get to a bad outcome faster.
The "own nothing, rent everything" approach gives you speed, but you have zero control when a partner gets flagged or a regulation changes. For established business owners, the goal is not necessarily to be a one-person company. The goal is to use AI so your 10-50 person team operates like a 50-250 person team. That is where the most durable, lasting value is created.
Claude Cowork vs. Claude Code: A Quick Breakdown
I have been getting this question a lot lately, so I wanted to clear it up. Claude Cowork and Claude Code are both made by Anthropic. They are both built on the latest Claude 4.6 architecture, but they do very different things.
Claude Cowork is your very smart, very capable digital intern. It works across your desktop and your browser inside emails, Slack, documents, and your apps. It helps you get through knowledge work faster.
To give you a better idea of how this actually looks in a normal workday, here are two examples of what you can ask Cowork to do:
From Transcript to Proposal: You can give Cowork a messy transcript from a discovery call and ask it to use a "Proposal Writer" skill. It will autonomously identify the client's pain points, draft a structured project proposal, apply your brand fonts and logos, and save the final document to your desktop.
Research and Spreadsheet Automation: You can ask Cowork to open your browser, visit a list of competitor websites to find their latest pricing, and then update an Excel spreadsheet on your computer with the new data. It acts just like a human assistant with a mouse and keyboard.
Claude Code is your digital engineer. It lives inside your file system and actually builds things. This is the technical side of the recent stepchange in capability. It can render custom graphics as real video files, build small internal tools for your team, or write automated scripts.
Here is how Claude Code functions as a technical resource for your business:
Automated Video Assets: You can ask Claude Code to create an animated "Product Launch" video using the Remotion framework. It will write the animation code, pull in your product images, and render a high-quality .mp4 file for your social media team without you ever touching video editing software.
Infrastructure and Integrations: You can ask Claude Code to connect your Shopify store to a custom database or Slack channel using an MCP connector. It handles the "plumbing" of the code so your sales data or inventory alerts sync automatically in the background.
The Bottom Line:
Cowork handles your day to day communication and administrative tasks.
Code handles the technical tasks you would normally need a developer for.
Note: If you want more details or if you want to get Claude Code up and running on your computer, I recently put together a full walkthrough with step by step instructions inside the Ampra Circle community.
Quick Win for Claude Users
I want to make sure you do not miss out on some free leverage. Anthropic is currently giving away extra usage credits to all paid subscribers.
The amount is tied to your plan. Pro users get $20, while Max and Team users can see credits of $100 or more. These credits work across the entire ecosystem: Claude.ai, Claude Code, and even third-party apps.
The deadline to claim these is April 17. To grab yours, go to your settings and look under the "Usage" section. It takes about 10 seconds and ensures you have the extra capacity to test out the new agentic features we discussed today.
That's a wrap for this week. Like I said, this was a packed one.
Thanks for reading. What an exciting time to be building!
See you next week,
Julien
PS: If this edition was useful, forward it to someone on your team who's been asking about AI. They can subscribe at www.ampra.ai/join-our-newsletter.