AI Vendor Hidden Costs and the Latest Practical AI News
Dear Friend,
Happy Wednesday. In our Applied AI class last week, we spent time working through generative image tools like Canva, Leonardo, and Firefly. The big takeaway from the lab was that you do not need to know complex photography terms to get good results anymore. You can simply upload a reference photo and tell the system to apply your image or branding guidelines. It is all about finding the tool that fits naturally into the process you already use.
That theme of simplified implementation carries right into a major topic we need to discuss: vendor contracts. The gap between a polished demo and a deployed workflow is where your budget disappears, and you have to protect your downside.
Anthropic Clashes With The Pentagon Over AI Risk
Anthropic recently briefed the U.S. government on its unreleased ‘Mythos’ model, which features highly advanced cybersecurity capabilities. At the exact same time, the company is actively suing the Department of Defense over a supply-chain risk designation. The lawsuit follows Anthropic’s refusal to allow the military unrestricted access to its systems for mass surveillance and autonomous weapons.
My Take: This is a massive legal and regulatory reality check for the industry. You cannot build boundary pushing tech and expect to avoid government oversight. For business leaders, this signals that the regulatory environment is tightening rapidly. The tools you rely on could face sudden compliance hurdles or legal injunctions. Legal and regulatory changes matter because careless use has real consequences, and we are seeing the groundwork for serious governance right now.
MIT CSAIL Solves the “Machine Unlearning” Problem
Researchers at MIT have successfully demonstrated a framework allowing large language models to selectively “forget” specific pieces of data without requiring a full system retrain. This technical breakthrough addresses a massive compliance headache for enterprises trying to adhere to data privacy laws like GDPR or CCPA after proprietary or protected information accidentally leaks into a training dataset.
My Take: This is a massive relief for corporate risk teams. Until now, the fear of accidentally baking sensitive customer data into a permanent model kept many legal departments from greenlighting internal AI initiatives. If we can reliably extract data post-training, businesses will adopt custom models much faster. It removes the permanent penalty of a data slip.
Source: MIT News – Machine Unlearning Framework
SanDisk Joins Nasdaq 100 As Data Demands Surge
SanDisk is officially joining the Nasdaq 100 index, replacing Atlassian. This shift reflects a massive surge in demand for data storage and memory chips driven by the rapid expansion of data centers. Because complex models require vast amounts of data processing, infrastructure companies that provide the physical hardware are seeing their market value skyrocket.
My Take: Everyone focuses on the software layer, but the physical infrastructure is where the real bottlenecks exist. If you want to understand where the market is heading, follow the hardware and the money. This is a clear reminder that data is heavy and expensive. If you are planning large scale deployments, you need to budget for the underlying storage costs, or you will quickly get caught off guard.
Science Corp Prepares First Human Brain Sensor Implant
Biotech startup Science Corp is preparing to implant its first sensor into a human brain, marking a significant step forward for neural interface technology. Led by Max Hodak, the company aims to use these advanced implants to treat severe medical conditions and eventually restore lost physical capabilities for patients.
My Take: This sounds like science fiction, but it is a very real medical advancement happening right now. The companies that succeed in this space will be the ones that navigate the massive ethical and privacy guardrails successfully. Would you try this???? I’m pretty biocurious, I wouldn’t be the last person to do it.
Apple Pivots to Smart Glasses
Apple is reportedly testing four different designs for upcoming smart glasses, targeting a launch in 2027. Unlike the heavy Vision Pro headsets, these glasses will not feature displays. Instead, they will focus on everyday wearability, incorporating audio, cameras, and integration with an upgraded Siri for hands-free assistance.
My Take: This is exactly what I mean when I say businesses need workflows, not just flashy demos. Apple is pivoting away from heavy mixed reality headsets toward lightweight, practical wearables. Imagine field workers, retail staff, or logistics teams accessing real-time data without needing to look down at a screen. You know I’ll be in line for these. I’ve had my Meta RayBans for a few years. AND I mostly use them for taking pics, taking calls, and listening to podcasts on my walks.
What Your AI Vendor Isn’t Telling You: 10 Questions Before Signing
AI vendor evaluation requires asking the hard questions that salespeople hope you skip. I recently published a detailed guide on this because the reality is stark. Industry research shows that the total cost of ownership often exceeds initial quotes by 300 to 400 percent. Sales presentations are optimized for impact, not accuracy, and case studies feature massive enterprises with resources you simply do not have.
Before you sign a contract, you must ask about the true total cost, including data preparation and integration. You need to know exactly who owns your data and whether it is being used to train outside models. You also need ironclad performance guarantees and clear exit terms to avoid expensive vendor lock-in. Perhaps most importantly, check the liability caps. Most vendors limit their financial exposure to a single month of subscription fees, leaving your business to carry the risk when the system inevitably makes a mistake. Do not accept vague verbal promises. Get the error handling and liability terms in writing.
Read the full Article
Google Labs 🧪
🚀How to get start to Opal
What Is Google Opal?
Google Opalis an experimental, no-code platform from Google Labs that lets you build, edit, and share AI-powered mini-apps and automated workflows using plain English instead of traditional programming.
Often referred to as a “vibe coding” tool, Opal takes a natural language description of an app idea and automatically translates it into a functional visual workflow.
When you first open Opal, your dashboard might be empty. Here is how to kick things off:
- The Gallery: If you need inspiration, start here to see examples of what others have built.
- Create New: Click this button to start building. Simply type in a natural language idea (e.g., “An app that plans a weekly meal plan based on the number of people and cooking frequency”).
🎨 The Visual Editor
Once you input an idea, Opal generates a visual workflow. This canvas represents the “logic” of your app.
Understanding the Blocks
Opal uses a color-coded system to show how data flows:
| Color | Block Type | Description |
| Yellow | User Input | Fields where the user enters data (e.g., “Number of people”). |
| Green | Generate | The AI “brain.” It uses models like Gemini 1.5 Flash to process info. |
| Blue | Output | The final result, such as a landing page, recipe list, or image. |
Pro Tip: You can manually add these blocks from the sidebar or simply type instructions in natural language to let Opal update the canvas for you.
🛠️ Customizing Your App
1. Refining the AI
In any Generate block, you can:
- Change the Model: Use the dropdown to select different models from the Google ecosystem (e.g., switching to Imagen for photos).
- Expand Prompts: Opal automatically expands your simple ideas into detailed instructions for the AI.
2. Adding Assets
You can make your app more specific by clicking Add Asset.
- Sources: Upload files, or pull from Google Drive and YouTube.
- Reference: You can tell the AI to “mimic the style” of an uploaded image or use a specific document as a template for its output.
3. Personalizing the Theme
Click the Theme section to change the cover image of your app.
- Randomize: Generates an AI image relevant to your app’s purpose.
- Upload: Use your own custom branding or photography.
🏃 Running and Testing
To see your app in action, click the Start button to enter Preview Mode.
- Console: While the app runs, open the Console tab to see the “interim” steps and technical details of what the AI is doing.
- App Toggle: Switch to the “App” view for a more immersive, full-screen experience of your creation.
- Undo/Redo: Use the canvas controls to revert changes or zoom in/out on complex workflows.
📤 Sharing and Exporting
Once you are happy with your Opal, you have two main ways to share it:
- Share the App: Click Share App and then Publish. This gives you a link where others can run the app themselves.
- Share the Output: If you just want to show someone a specific result (like a specific meal plan you generated), click Share Output to get a unique URL for that HTML page.
Change the Output Format
By default, Opal creates a web page. However, you can change the output block to save directly to:
- Google Docs
- Google Slides
- Google Sheets
Note: If you choose Google Docs, every time you run the Opal, it will append the new content to that same document!
AI Tools to Explore This Week
- Jasper.ai: Great for teams creating consistent, high-quality marketing content while keeping brand voice aligned.
- HeyGen: Create professional videos with AI avatars and voiceovers. Perfect for ads, training, or social media content without a full production team.
- Opus Clip – Chop up long form videos into short form content for social media.
Join the Conversation
Got a burning question, a fresh take, or just want to share your latest AI wins? Hit us up at [email protected]. Your insights keep this community growing and thriving!
See you in the Lab,
-Nicole A. Donnelly
Founder, AI Smart Ventures
AI Strategy – AI Training – AI Consulting – AI Implementation
