Welcome to the Pinnacle Perspective from Pinnacle AI. Here's the problem. Some customer interactions are simple. Read this, send that, move the work along. Others take a little more judgment. Maybe a customer gives half the details, changes the request, or asks something that needs a few steps before the answer makes sense. And as an owner, you don't want the same setup handling every job the same way. That's what this update is about. The AI agent action in workflows now gives you more choice. Before, this action was limited to GPT-based models. Now Anthropic and Google models are available too, alongside OpenAI. So when you're building a workflow, you can choose the provider, choose the model, and for supported models, choose how much reasoning it should apply. The model picker has been redesigned to make that choice easier. Models are grouped by provider. Some are marked with a thinking label, which means they support reasoning. You'll also see a short description, so you have some context before you pick one. For models with that thinking mark, you can set the effort level right there in the same drop-down. Low, medium, or high. That matters because not every task needs the same amount of thinking. If the workflow is doing a simple read or a basic operation, a lighter option may be enough. If accuracy is critical or the interaction has multiple steps, you can choose a model with more reasoning and set the effort higher. In day-to-day terms, this helps when work comes in while your team is busy. Maybe someone is at the front counter near closing, a lead is waiting for a reply, and the workflow needs to sort through the request before the next step happens. You can set the AI agent action to match the job instead of treating every interaction like it has the same level of complexity. To use it, you add the AI agent action inside the workflow builder. Then you open the model drop-down, browse by provider, look for the thinking mark if you need reasoning support, and choose the model that fits. From that same dropdown, you can pick low, medium, or high effort where it's available. After that, you write the instructions for what the agent should do, save the action, and your workflow is ready. And the owner is still in control here. You decide which model gets used, you decide how much reasoning to apply. You write the instructions. When the interaction reaches the point where a real person needs to make the call, your team takes it from there. That's the update. More model options for the AI agent action, better labels when you're choosing, and effort control for the jobs that need more careful reasoning. To learn more, visit getpinnacle.ai.