Choosing AI tools

Chat apps such as Claude, ChatGPT and Gemini are where a person works with a model directly. A human starts every task and reads every answer, which makes them ideal for drafting, analysis and testing whether a task suits AI at all.

APIs let your own code send prompts and receive responses. Nothing happens unless a program calls it, but you control the input, output format, logging and cost. Coding agents such as Claude Code go further: Anthropic describes Claude Code as an agentic coding tool that reads your codebase, edits files and runs commands.

Automation platforms such as n8n, Zapier, Make and GoHighLevel workflows connect a trigger, like a new form submission, to a chain of actions. Most now include an AI step, so you can add a model without writing an application.

Ask who starts the work. If a person starts it each time and reads the result, a chat app is usually enough. If an event should start it automatically, such as a new lead or a new ticket, you need an automation platform or an API.

Then ask about volume, where the data lives and how custom the logic is. A few tasks a week with simple steps fit a no-code platform. Thousands of runs, complex branching or a custom user interface point toward code calling an API.

A sensible path for most business tasks: prove the prompt in a chat app, move it into an automation platform once the steps are stable, and move to custom code only when the platform becomes the bottleneck.

A closed model is one you can only use through the vendor's app or API. The vendor hosts it, updates it and sets the price. An open-weight model publishes its trained weights so you can download, run and fine-tune it yourself.

Licences vary by release, so read the model card. Google's Gemma 4 models, for example, are published under Apache 2.0, while other families use their own community licences with extra conditions. Open weights usually do not include the training data or code.

Running open weights gives you control over where data goes and can lower cost at high volume, but you take on hosting, scaling, security patches and upgrades. For most first projects, a hosted model through an API or platform is the faster and safer choice.

Match the model to the job. Classification and extraction usually work on a fast, low-cost model. Complex reasoning, long documents and agent work justify your provider's current flagship. Test both on real examples rather than assuming.

Before you paste client or customer data anywhere, check the tool's data and retention terms and your agreement with the client. Workplace and API plans often have different data handling from free consumer plans.