Chapter 16: The Expanding AI Toolbox and How to Choose
Start with the work you already do
People ask which AI is best. It is close to the wrong question, for two reasons.
The rankings shift with every release, so any answer has a shelf life of weeks. And the leading models have converged enough that for most everyday work, the difference between them is smaller than the difference between a well-built prompt and a lazy one. The skills in Part III will do more for your results than any switch between tools.
The better question is: where does the work already happen? That is what actually distinguishes the products now, and it is the theme of this chapter.
Google Gemini
Google’s assistant, available as a standalone app and, more consequentially, built into Google Workspace. Chapter 14 covers it in full: the plans, the configuration, the Workspace side panels, and the distinction between the consumer product and the work one. For the purposes of this chapter, the relevant fact is structural. Google folded Gemini into Workspace subscriptions rather than selling it as a separate AI fee, so an organization already paying for Workspace gets it at effectively no incremental cost, and a great many are not using it.
What to check before committing. The privacy posture on consumer plans deserves attention: Google treats consumer Gemini chats as consumer data, with a subset reviewed by humans and used to improve its models, and the principal control is turning off Gemini Apps Activity, which also stops chats being saved to your history. Paying for the consumer Pro tier buys better models, not different privacy terms.
That last sentence generalizes across the industry and is worth carrying with you: on consumer plans, paying more usually buys capability, not confidentiality. Different data terms generally come with business and enterprise agreements, not with a personal upgrade. Your privacy rules should reflect this difference.
Microsoft Copilot
Microsoft’s assistant, embedded across Windows, Edge, and the Microsoft 365 applications. Chapter 15 covers it in full, including the four different products sharing the name, what grounding in your organization’s data buys, and what it does inside Word and Excel specifically. What matters here is the cost model, which is the opposite of Google’s: the full business version is sold as a paid per-seat addition to a qualifying Microsoft 365 plan, so the real price is the two together.
A structural point worth understanding. Copilot has historically run on models built by OpenAI rather than by Microsoft. This is the clearest illustration of the chapter’s main idea: the assistant you are using and the model underneath it are increasingly separate things, and the company selling you the assistant may not be the company that built the intelligence.
The rest of the field
Search-first assistants. Products like Perplexity are built around answering with live sources rather than from training. Useful when currency and citation matter most. The verification rule still applies: a citation is a place to look, not a guarantee.
Open-weight models. Systems whose weights are published, runnable on your own hardware or a provider of your choosing. Nothing leaves your control, and the capability gap behind the frontier has narrowed considerably. Bonus Chapter C covers this properly for readers who need it.
Specialized tools. Transcription, image generation, video, music, presentation building, coding. Frequently better at their one thing than a general assistant, and one more subscription each.
Assistants inside software you already own. Your CRM, your accounting package, your practice management system. Often the highest-value AI in a professional’s life and the least discussed, because nobody writes headlines about a summarization button in a claims system.
The durable insight
Here is what to take from this chapter when every product name in it is obsolete.
The model is becoming a commodity. The integration is the product.
Frontier models are converging in capability, the leaders trade places routinely, and the same underlying model may power several competing assistants. What differs (durably) is where the assistant sits, what it can see, what it can act on, and what contract governs your data.
Which means the right way to choose is not to read benchmark comparisons. It is to ask where your work already lives.
Choosing, in four questions
1. Where does my work already happen? Google Workspace, Microsoft 365, or neither. If you are deep in one, its native assistant starts with an enormous advantage that no benchmark score overcomes.
2. What kind of work is it? Long documents and writing, or spreadsheets and meetings, or code, or research. The tools have genuine emphases even as they converge.
3. Who controls my data, under what terms? Consumer terms, business terms, or self-hosted. For anyone handling client information, this question outranks capability entirely.
4. What am I actually going to pay for? Most people need one general assistant plus whatever is already bundled with software they own. Two general subscriptions is usually one too many.
Using more than one
There is a real argument for keeping a free account on a second tool, and it is not variety.
When you re-ask a question cold, doing it in a different tool is stronger than doing it in a new conversation of the same one. Divergence between two independently built systems is a meaningful signal that the ground is thin.
Remember the limit: models trained on heavily overlapping text can repeat the same error. Disagreement between them signals uncertain ground, while agreement provides only weak corroboration and still requires verification.
Switching costs are low, and you should keep them that way
Almost everything you have built while reading this book (your CLEAR prompts, your verification habits, your judgment about what to delegate) is portable. That is deliberate, and it is the whole thesis of the book.
What is not portable: conversation history, configured workspaces, custom setups, and connected integrations. So keep your prompt library in a plain document outside any single vendor, and think twice before building a workflow that only functions inside one product. Treat portability as a design principle for your personal system.
Next comes the skill that makes this part durable: teaching yourself tools that do not yet exist.
Exercise 16.1: Audit where your work lives
List the ten applications you spend the most time in. Mark which already include an AI assistant you are paying for and not using.
Most people find at least one. That is free capability sitting idle.
Exercise 16.2: Run the divergence test
Take a factual question in your field with a specific answer, and ask it in two different tools, cold, in fresh conversations.
Where they diverge, verify at a source. Note which one was right. Do this three or four times and you will have a calibrated sense of both tools that no review can give you.
Exercise 16.3: Read your actual data terms
For whichever tool you use most, find the answer to three questions: Are my conversations used to train models? How long are they retained? What changes if I upgrade?
Write the answers down with today’s date. They will become part of your personal AI policy.
Learn More About Artificial Intelligence
Learn more about the CLEAR, TRUST and LEARN methods for Artificial Intelligence