Live billing switches on, and every agent becomes a row you can edit

Real billing went live on Monday, screenshots became task evidence, task requests and direct messages to the agent arrived, and Agent Builder v2 rebuilt the agent system so no agent lives in code.

The week the platform stopped being a prototype. Real billing switched on Monday, and the agent system was rebuilt so that every agent is something staff can see and edit rather than something written into code.

For our partners

1. Live billing switched on July 27. The platform also got a new mark, The Swell, a wave waterline, across the app.
2. Screenshots became evidence: every capture the agent takes is kept and can be pinned to a task, at capture time or later, and the client portal shows them on the task. Partners and staff can add and remove them too.
3. Partners can message the agent directly from the requests table, and the messages are kept as history.
4. Approved content is final: no new versions and no re-asks.
5. Task decisions reach the agent for team and provider tasks, and the task page shows the approval history and notes thread.
6. Google Business post items carry their published truth: the live Google link, or the moderation rejection if there was one.
7. Page monitoring shows checks rather than grades; the score gauge is gone.
8. Anything our team can see that a partner cannot wears an obvious amber Staff pill.
9. Team admins can change a member’s role after the fact, and members can be granted edit access to the agent’s guidance.
10. Website map became Keyword Mapping everywhere a client sees it, and discovery keyword tables tag each row with its market.

Behind the scenes

1. Agent Builder v2 landed between Thursday and Saturday. Every agent is now a row the team can edit: who it is, the documents it reads, the data it is handed, what starts it (a schedule, an event such as a matched meeting, or a task becoming ready), who can see it, and its history.
2. Staff can chat with an agent about work it has finished: ask what it was thinking, and optionally let it act, with a second AI model for a second opinion.
3. We wrote down the Copy Gate: four tests every new sentence of interface copy must pass before it ships, the default being silence. It is the reason the app says less than you might expect.
4. Approval requests work for every task kind, not only content.
5. Campaign requests notify the assigned staff, and no assigned staff raises an ops alert. Failures in profiling a new client, or in the crawler’s account, raise an alert too.
6. The agent learned to read the local map picture for a competitive review.

1. Agent Builder v2

What an agent knows going in is assembled per agent rather than from one fixed profile. Schedules are written per agent with per-campaign exceptions, and campaigns are reviewed daily while onboarding, then settle into weekly, monthly or twice-monthly rhythms. By Sunday no agent lived in code: if no row describes an agent, nothing happens, and staff can become any agent by name.