Hi {{first_name}}!
It’s almost five o'clock on a Friday and what a week! So much happened in the world of AI.
I spent a few hours this week training a non-profit team on AI adoption. While we started with prompts and productivity, the conversation was focused on a much more grounded concern: "Security"
Which is a good reminder for everyone… The AI plan or tier you pay for (or don't pay for) is one of the most important security decisions your business will make this year.
In this edition, I’m breaking down the "Security Tier" gap, because as these tools get more powerful and easier to use, the implications and risks of using the wrong version get exponentially higher.
And speaking of “more powerful and easier to use”: this week Anthropic gave Claude the ability to actually use your computer (moving your mouse, opening and controlling apps, and finishing tasks while you're away from your desk).
Here's what's in this edition:
New & Noteworthy: Google lowers the barrier to building apps, Claude keeps getting more capable, ChatGPT leans further into shopping, and a few other AI moves worth paying attention to.
The Deep Dive: Why the gap between free, pro, and business AI plans matters more than most small teams realize.
The Main Event: Claude Cowork can now use your computer, and paired with Dispatch, that makes remote task execution a lot more practical.
OK, let's get into it!
New and Noteworthy
Google’s "Vibe Coding" Suite: Google just overhauled its creator tools. Stitch now uses an infinite canvas where you can literally speak a website design into existence. Meanwhile, Antigravity (in Google AI Studio) has moved beyond prototypes; it can now build full-stack apps with user logins and databases. If you’ve been waiting to build a client portal or internal dashboard, the "technical barrier" to making that happen has been significantly lowered this week.
The Microsoft / Claude Alliance: Microsoft 365 Copilot just introduced a multi-model approach that brings Anthropic’s Claude into the Researcher experience and powers the new Copilot Cowork workflow layer. The bigger takeaway is not that every SMB suddenly needs another subscription, but that Microsoft is making it easier for teams already paying for Copilot to access stronger reasoning inside the tools they already use.
Anthropic’s Relentless Shipping Pace: Anthropic has been shipping at an unusually fast pace lately, with a steady stream of updates across Claude, Cowork, Code, and its model lineup. The latest additions are Claude Dispatch, which lets you kick off or monitor work from your phone, and Computer Use, which gives Claude the ability to interact with apps on your desktop using your keyboard, mouse, and screen when a connector or browser route is not enough. If you already use Claude Cowork, this is another clear sign that Anthropic is pushing hard to make Claude more useful for real workflows, not just chat. And if you already use Cowork, Dispatch and Computer Use are a great way to stay productive while away from your desk.
AI Agents Are Getting Enterprise Guardrails (That's Good News for Your Business): Nvidia announced NemoClaw at its GTC conference, a new platform that is promising to add security, privacy controls, and policy enforcement to AI agents. Here's why that matters: AI agents like OpenClaw can already handle tasks across email, files, and the web on your behalf. The problem was, they were built for individual users with no controls over what data gets accessed or shared. NemoClaw aims to be the missing layer that IT teams and compliance departments need before they'll give the greenlight. In his keynote, Jensen Huang said: “every company now needs an AI agent strategy the same way every company once needed a website strategy.” This is still early, but the direction is clear. AI agents are getting closer to being ready for real business use, not just tech demos.
Google Is Also Making it easy to switch to Gemini: Taking a cue from Anthropic, Google now allows you to migrate your chat history and "memories" from other LLM’s directly into Gemini with ease. This is a clear move to lower the friction for people who want to switch platforms without losing months of personalized context and preferences.
The Text-to-Speech Arms Race: Two major updates hit the voice space this week. Smallest.ai released Lightning V3, which is designed specifically for real-time conversations and includes natural "fillers" like "um" and "uh" to make AI agents sound more human. Meanwhile, Mistral released Voxtral, a high-quality open-weight model that allows businesses to run professional-grade speech generation on their own local servers for better privacy.
OpenAI Cancels Sora: In a surprising move, OpenAI has shut down Sora, its AI video app, just months after launch. The clearest takeaway is that OpenAI is narrowing its focus and stepping back from a consumer video product that brought both excitement and a lot of controversy. For business owners, this is a reminder that not every flashy AI product becomes a lasting platform, and that the companies winning right now are the ones turning AI into everyday workflow and infrastructure, not just viral demos.
OpenAI Launches "Agentic Commerce": ChatGPT is becoming a much more serious product-discovery channel. OpenAI just rolled out richer shopping experiences powered by its Agentic Commerce Protocol (ACP), including more visual product results and side-by-side comparisons. The practical takeaway for SMBs is simple: product search is starting to happen inside AI tools, not just on Google or Amazon. If you sell physical products, this is worth watching closely because visibility in ChatGPT is starting to matter in a more meaningful way.
The #1 AI Fear Is Unreliability: Anthropic published findings from the largest qualitative AI attitudes study ever conducted, covering 80,508 interviews across 159 countries. Unreliability ranked as the top concern at 26.7%, ahead of job displacement at 22.3%. That tracks with what I hear from teams I work with every week. The biggest barrier to adoption is not trusting the output. If you're leading a team through AI adoption, this is useful context. Meeting that with proper training and clear usage guidelines goes a lot further than just handing them a login and hoping for the best.
Meta Is Making a Big Bet on AI Agents: Meta acquired the team behind Dreamer, a startup building a personal agent OS for non-technical users. They are building agents described as "truly personalized and always-on." The practical takeaway for you is that the apps you already use every day are about to get more capable. You won't have to adopt new tools. AI will simply be embedded in the platforms you already run your business on.
OpenAI Is Building a Single Desktop App for Everything: OpenAI is working on a unified desktop application that brings chat, research, and task execution into one workspace. The goal: instead of bouncing between ChatGPT in your browser, a separate coding tool, and web research in another tab, everything lives in one place. For anyone already using ChatGPT as part of their daily workflow, this is heading toward becoming your default work hub. Fewer tools, less friction, faster execution.
Lovable Hits $400M ARR and Is Acquiring Fast: Lovable, one of the leading AI app-building platforms, crossed $400 million in annual recurring revenue with over 200,000 new projects created daily. That growth signals something bigger than one company's success. The barrier to building custom internal tools, client portals, intake forms, and lightweight business apps without hiring a developer is rapidly disappearing.
Midjourney V8 Is Live in Alpha: Midjourney's V8 model is now available for early testing at alpha.midjourney.com. Images generate roughly 5x faster than V7, with native 2K resolution, better prompt understanding, and much-improved text rendering in images. If your team creates proposals, pitch decks, social media content, or marketing materials, V8 is a noticeable step up. The text rendering improvement alone matters: you can now generate visuals with readable headlines, labels, and captions directly, something earlier versions struggled with.
Every Major AI Lab Is Now Chasing Enterprise Distribution: Every major AI lab is now chasing private equity deals to get their tools into more businesses. OpenAI is in talks with firms like Bain Capital, while Anthropic is reportedly courting Blackstone. The labs that win enterprise distribution will win the market. This means AI is coming to your vendors and tools whether you seek it out or not.
Free, Pro, or Business? Why Your AI Plan Matters More Than Most Small Teams Realize
This week, a big part of my conversation (during one of my training sessions) wasn’t just about productivity. It was about security.
A lot of small businesses are already using ChatGPT, Claude, or Gemini for real work, but they’re doing it through personal or lightly managed accounts. That creates a problem fast. The models may be powerful, but the privacy, admin control, and data protections behind the account can vary a lot depending on which plan you’re actually on. And the pricing gap is often smaller than people think. ChatGPT Plus is $20/user/month, while ChatGPT Business is $25/user/month billed annually or $30 month-to-month. Claude Pro is $20/month, while Claude Team starts at $20 per seat/month billed annually or $25 billed monthly.
Here’s the simple way to think about it:
Free plans are fine for experimentation, but they are not the right default for business use. Privacy settings, data handling, and admin controls vary by platform, and you generally do not get the safeguards a business needs.
Paid personal plans like ChatGPT Plus or Claude Pro are a step up in capability, but they still are not the same thing as a true business setup. You may get access to stronger models and higher usage limits, but you are still operating in a personal-account environment, usually without centralized admin controls, shared billing, or consistent oversight. And once you reimburse a few employees for their own $20/month accounts, you are often very close to business-tier pricing anyway. With OpenAI, the difference between Plus and Business can be as little as $5 more per user per month on annual billing.
Business plans are where this starts to get much cleaner for businesses. That is really the point. Not enterprise, not massive compliance overhead, just a better setup for a real team. OpenAI’s business plan includes admin controls, centralized billing, and business-focused security commitments, while Claude Team is designed for shared work across growing teams. That is why my advice is pretty simple: if AI is becoming part of how your team actually works, skip the patchwork of personal or pro plans and move to a business-grade tier early. The added cost is usually modest. The added clarity, security, and control are not.
The bottom line: if you’re using AI at work, treat it like business software, not a personal app. For most small businesses, this is less about getting fancy and more about getting aligned. A slightly higher monthly cost usually buys you a much better foundation.
Sidenote: The on-prem option: Local LLMs
If you want the most private setup, the next step is running an open-source model locally, on a machine you control in your office or private environment. That can reduce or eliminate the need to send sensitive prompts and files to a third-party cloud service, which is why some teams look at tools like Ollama and other self-hosted setups. But this route also comes with tradeoffs: more technical setup, hardware requirements, maintenance, and usually a different level of model performance than the top cloud tools. For most SMBs, this is not the first step. But for teams handling especially sensitive data, it can be a very compelling option to explore. If you’d like to learn more about that option you can do so here.
The Main Event: Your Phone Is Now a Remote Control for Your Desktop
If you have been reading the last few editions, you already know about Claude Cowork. It is the desktop Agent I now use to turn a meeting transcript into a summary, a Google Doc, an email follow-up, and a series of ClickUp tasks without writing a single prompt.
Last week, Anthropic introduced a capability in the Claude Mobile app called Dispatch. It gives you the ability to prompt Claude Cowork on your desktop or check in on a task from your phone. But this week, they added a massive upgrade to Claude Cowork allowing the tool to actually navigate your computer, use apps, and control your mouse. It is called Computer Use, and it changes how Claude handles the "gaps" in your workflow.
The New Three-Step Execution Before this week, if Cowork did not have a direct "connector" for a tool (like Gmail or Slack), it had to rely on a browser to get things done. If the task required an app that lived only on your desktop, such as a specific accounting tool or a legacy CRM, it was stuck.
Now, Cowork follows a smart priority order:
Connectors First: It uses high-speed integrations (like Google Drive) because they are the most reliable.
Browser Second: If no connector exists, it opens a browser to find what it needs.
Computer Use Third: This is the new part. If the first two fail, Claude can now "see" your actual desktop. It can move your cursor, click buttons, and type in any application on your screen, just like a human would. I tested it and it was a bit slow but it does work and it will only get faster from here.
Why This Matters for Your "Dispatch" Workflows The real win here is for the "in-between" moments when you are away from your desk. In the past (as in last week), you might have sent a Dispatch command from your phone only to have it fail because Claude could not find a web-based way to finish the task. Now, that same command has a much higher success rate because Claude can simply "drive" the desktop app you already have open.
It makes your remote commands via Dispatch significantly more robust. You can send a task from your phone and trust that even if there isn't a fancy "AI integration" for your software, Claude can still click the buttons and get the job done.
A Few Important Guardrails Because this new feature interacts with your actual screen (and not just a sandbox or a browser tab), there are some common-sense rules to follow:
The Permission Layer: Claude will ask for permission before it touches a new app.
Keep It Awake: For Dispatch and Computer Use to work, your desktop has to stay awake with the Claude app running.
System Requirements: Computer Use is currently a research preview and is available for macOS users on Pro and Max plans.
Sensitivity: Anthropic has blocked access to sensitive categories like banking, but I still recommend closing any windows with private client data before you leave your desk.
The Bottom Line We aren't just giving Claude "access" to our files anymore. We are giving it the ability to navigate our entire digital workspace the same way we do. If you have been hesitant to try Cowork because your specific software "doesn't have an AI version," this is the week that barrier started to fall.
That's it for this week. If you want help with Claude Cowork, feel free to reach out with questions. And for those that want to go deeper, we offer working sessions to help you set up your Claude Co-work account and fully leverage its capabilities. If that’s of interest please feel free to reach out!
Talk soon,
Julien
P.S. If you found this useful, forward it to someone on your team who's making decisions about AI tools. They can subscribe at www.ampra.ai/join-our-newsletter