What Kind of AI is Your Vendor Offering?
If you’re looking to integrate AI tools into your workflow, consider three different types of AI tools:
- Your AI assistant. The assistant you talk to directly, by text or voice. ChatGPT, Claude, Gemini, whatever you have open in a tab or on your phone. You chose it, you pay for it, and it works for you.
- Their AI. An assistant a product hands you inside its own walls. The chat bubble on a software vendor’s site, the “Ask AI” panel in a document tool, the support bot that answers before a human does.
- Invisible AI. Models working behind the scenes with no text interface at all. Classifying, extracting, routing, proofreading, filling in fields. You see the result but the model is invisible.
All three have real uses. Fenix runs on all three. But they are not all equally important.
Type 1 is understood but still undervalued
Everyone gets the personal assistant. It is the AI most people have actually used. What is undervalued is where it is going.
Right now your assistant mostly lives in a box. You paste things in and copy things out. That is a temporary state. The interesting version of your AI is the one connected to your other software: your email, your calendar, your documents, your billing system, your docket. It reads from them and writes to them. You give an instruction once and the work happens across four or five applications you never open.
That is the future. Your AI assistant is your primary interface. You don’t primarily work in MS Word or even the browser. The chat interface is on its way to replacing a large share of the interfaces we use today. Nobody wants to learn where a setting lives in the menu of twelve different applications. They want to say what they want and have it done. A personal assistant with real access to your tools is the first interface that can deliver that.
So the question to ask of any software you use is simple: can my AI reach it? If the answer is no, the product is holding your data at arm’s length.
Type 2 is overestimated and overdone
The embedded assistant is where most of the marketing money goes, and most of the hype. Every product now has a sparkle icon. Click it and a chat window opens that knows about this one product and nothing else (or do they?).
Chat apps are useful. A support bot that answers a question in ten seconds beats a ticket queue. A document tool that summarizes the file you are looking at saves a minute.
But consider the limits:
- It only knows what the vendor lets it know. It cannot see your email, your other tools, or the context that makes the task make sense.
- It runs on a model the vendor picked, at a quality level the vendor picked, and it will not get better until they ship an update.
- It cannot hand the task off. When the job crosses into another product, you are back to copy and paste.
- You are on the vendor’s side of the glass. The assistant works for them first, not you.
The result is that you are drinking AI through a straw. A little at a time, only the flavor they chose, only while you are inside their product.
Beware of any company whose AI strategy is to keep you in that chat window. When a vendor pushes hard on “use our AI” and offers no way for your own assistant to connect, the AI is window dressing, not a serous capability.
Type 3 is important and invisible
Behind-the-scenes AI gets the least attention because it has no interface to demo. That is fine. It is doing much of the real work.
In a patent practice, that looks like converting images to line drawings, reading an incoming office action and populating the docket, pulling cited references, flagging antecedent basis errors in a claim set, or checking a filing for missing parts before it goes out. No one wants to chat with any of that. They want it done correctly and silently, with a human reviewing the output.
This is where the models earn their keep with the least ceremony. It is also the type that improves a product without changing how you use it. A good invisible AI shows up as fewer errors and faster turnaround, and you may never know it is there.
The only thing to watch for here is the opposite of the type 2 problem: a vendor that describes a product as “AI-powered” when the model is doing something trivial in the background. Ask what the model is actually doing. If the answer is vague, the AI is just a label.
How to read an AI product
Understanding these categories makes evaluating a product easier.
- Does it work with your AI? Can your assistant read from it and act on it? This matters most, and it matters more every day.
- Is the AI a convenience or a cage? Handy interface for small in-product tasks is fine. Positioned as the primary way to get AI value out of the product is a warning.
- What is the AI doing in the background, specifically? Concrete answers are a good sign.
At Fenix we use all three. Type 3 does the heavy lifting on documents and dockets. Type 2 exists where it is convenient. But the design goal is type 1: your assistant, connected to your practice, doing the work across every tool you use. The product should serve the AI you already have, rather than compete with it.