Claude 3.5 and the competition among assistants

In 2024, Anthropic introduced Claude 3.5 and with it climbed another rung in a competition that was already very intense among artificial intelligence assistants. For anyone following the topic from the outside, the news reads like a sprint: every month a model appears that does something a little better than the last. But if you run a small business, the question that really matters isn’t who’s winning the race, but what this technology can do for your business today.

And there’s the point many companies overlook: there’s no “best assistant” in the abstract. There’s the right assistant for a concrete task, with your budget, your data, and your processes. That there are several strong options, and that they improve month after month, is actually good news for you: it means more capacity available, more price competition, and less risk of getting locked into a single tool. The challenge is no longer getting a powerful model; it’s using it with judgment.

Why the competition benefits you

When several providers push at the same time, the benefit trickles down to the end user. Capabilities that once seemed like science fiction are now within reach of a small business, and the costs of using these models have followed a downward trend. For a small business, that completely changes the calculation: automating customer responses, organizing scattered information, or writing first drafts no longer requires a big-company budget.

The important thing is not to marry a single option out of fashion. A good development is designed so the model is an interchangeable piece, not the heart of the system. That way, if a cheaper or more capable alternative appears tomorrow, you switch without redoing everything.

Choosing the right model for each task

The key is to match the tool to the job. You wouldn’t use a semi-truck to go get bread, or a motorcycle to move a house. It’s the same with AI assistants: some shine at writing and reasoning over long texts, others are ideal for quick, high-volume tasks where cost per query matters.

• Complex tasks requiring judgment: analysis of long documents, careful writing, or summaries that require understanding context. Here a more capable model is worth it, even if it costs a bit more per use.
• Volume and speed tasks: classifying emails, tagging messages, or answering frequent questions. Here a fast, economical model wins, because you’ll call it thousands of times.
• Sensitive tasks with your data: quotes, files, or histories. The priority isn’t the model, but how you protect the information and who has access.
• Tasks where an error is costly: anything that touches money or commitments to customers should go through human review before it goes out.

The goal isn’t to have the most advanced AI, but the one that solves your problem reliably and at a cost that makes sense.

From enthusiasm to decisions with trustworthy information

It’s easy to be dazzled by demonstrations. What’s hard, and what’s valuable, is turning that capability into measurable results within your operation. An assistant that writes beautifully is worthless if it doesn’t connect with your way of working, if it doesn’t respect your data, or if no one measures whether it actually saves time.

That’s why at Normandia Web we insist on custom software: instead of adapting your business to a generic tool, we build around your processes, preserve your data, your history, and your way of operating, and leave the AI model as a component that can be fine-tuned or replaced. That way the technology fits your way of working, and not you to its.

How to take your first step with AI

If you want to take advantage of this wave without tripping up, we recommend proceeding with order:

• Start focused and measurable. Choose a single repetitive, annoying task—for example, answering frequent questions—and set a clear goal for it: how much time it should save.
• Define what a good result is. Before automating, write down what a correct answer looks like. Without that, you won’t know if the AI is helping.
• Protect your data from day one. Be clear about what information goes into the assistant and who can see it.
• Keep a person in charge. For anything that touches customers or money, always have human review.
• Don’t anchor yourself to a single provider. Ask that the solution allow switching models without redoing everything.

The competition among assistants will keep accelerating, and that’s fine: it’s capacity coming within your reach. What will make the difference isn’t which model you choose today, but how well you put it to work on a real problem in your business. If you have that repetitive task in mind that you’d like to take off your plate, at Normandia Web we can discuss it with you and design that first experiment together.