Who does the thinking

When an AI agent works with braaand, there are two brains in the room: the agent's own model, which your AI subscription already pays for, and braaand's factory brain, which runs on your plan's credits. Braaand is deliberately built so that each one does the work it's best placed to do.

The two capability surfaces

The deterministic surface costs nothing. Every read and write, element patches, rendering, BrandFit layout fixes, color schemes and themes, group duplication, collections, image conversion, and the whole synced folder. None of it runs a model on braaand's side, so none of it draws credits. This surface is intentionally complete: an agent can clone a template, write the copy itself, patch it in, settle the layout, look at the render, and put the result on a review board without braaand's AI ever waking up.

The factory brain is metered. Workflow batches (template matching, copywriting, imagery picking, vision scoring), AI image generation and editing, and the board's iterate chat. These draw from the plan that pays for the brand.

Why it's split this way

A capable agent like Claude brings its own reasoning to the session. Forcing braaand's models in between for a one-line headline change would mean paying twice for the same thought, and the result would be no better. So for hand-work, braaand's job is to be excellent infrastructure: give the agent real brand data, deterministic tools with predictable semantics, and a render to check its work against.

Credits buy the things a single agent in a chat can't do well:

• Volume with consistency. A pipeline producing, scoring, and routing dozens of creatives across formats, with the same brand rules applied to every one.
• Durability. Runs survive the tab closing, the laptop sleeping, the conversation ending.
• Judgment at scale. Vision scoring every render and routing the results is a different job than eyeballing one image.
• Generation. Making new on-brand imagery is model work no matter who asks for it.

The rule of thumb the skill teaches agents: hand-work is yours, volume is braaand's. A tweak to one creative is the agent's own free work. Six ads matched, written, scored, and delivered on a share link is the factory's.

What this means in practice

• Asking your agent to "fix the headline on #3" costs nothing: it writes the line itself and patches the element.
• Asking for "a campaign about pollinators" runs the factory and costs credits, because you're buying the orchestration, not just the words.
• A brand whose credits run out doesn't lose its agent workflow. Everything deterministic keeps working; only the factory and generation pause.

See AI credits for what things cost, and the skill for how agents are taught this split.