Practical AI · 3-minute read
You open your laptop to finish a proposal. Someone recommends ChatGPT. Someone else says Claude. Your email has Gemini, and Microsoft would like a word about Copilot. Apparently your ordinary Tuesday now requires a seating chart.
The confusion is understandable. People use ‘AI’ to mean a company, an app, a model or a feature. Those are different things.
Separate the names from the jobs
The company builds or supplies the technology. The app is the product you open. A model is the trained system that processes inputs and generates outputs. A feature is a specific capability, such as summarizing an email thread.
OpenAI → ChatGPT: GPT families and versions underneath.
Anthropic → Claude: Families such as Haiku, Sonnet and Opus underneath.
Google → Gemini app: Gemini families and versions underneath.
Microsoft → Copilot: Can use models from multiple providers.
These are familiar examples, not the whole industry. One app can offer several models, choose one automatically or combine models with search and other tools. A model menu may also mix model choices with settings controlling how long a model works. More thinking time is not a warranty. [1–4]
Models differ in speed, cost and strengths. Some suit quick routine work; others are intended for more demanding reasoning. The best choice is the one that handles your task well enough to use after review. [6]
Built in and standalone can overlap
Standalone AI has its own app or website, such as ChatGPT or Claude. Built-in, sometimes called native or embedded, AI appears inside software you already use. Think Gemini in Gmail or Docs, or Copilot in Word or Excel. Gemini and Copilot also have their own chat experiences. [3,5]
These are ways to access AI, not permanent divisions between brands. OpenAI and Anthropic also supply models that developers can build into other products. [2,6] Built in does not necessarily mean the software vendor created the underlying model, or that processing stays on your device.
So how many are there? There is no single useful total without deciding whether you mean apps, model families, individual versions or features. Even the models available in one app depend on its current offering, your plan and settings. You do not need to collect them.
Make a map of your own work
Imagine Tumble Creek Marketing, a fictional agency. Its team wants help with proposals and already uses an office suite. Before buying another subscription, it lists the approved AI features already available and tries one fictional proposal task.
Help me map these tools: [names]. Separate company, app, model and built-in feature. For each, identify a possible use for [task], what access it would need and what we must verify. Mark unknown features, costs and protections ‘Check with provider.’ Do not assume our account includes them.
Have your account administrator verify the map against current vendor information. Check the actual work account, data access and permissions. A familiar logo does not make every feature appropriate for confidential material. Test whether the output treats different customers fairly before using it.
Your turn: Which tool you already use might deserve a closer look before you add another login?
Sources
