The client brief is out; delivery settings take its place

The client brief has left the product, replaced by a Review delivery settings step in the strategic alignment checklist. Feedback on a proposed content plan now opens a conversation with the agent that revises it.

A day of clearing something out. The client brief was a document we asked partners to write at setup, and it had stopped earning its place: the work is actually driven by facts, rules, guidance and delivery settings. So it is gone, and the setup step that replaced it points at something a partner can act on.

For our partners

1. The client brief is gone from setup, from the client overview and from the client portal. In its place there is a Review delivery settings step in the strategic alignment checklist.
2. Sending feedback on a proposed content plan now opens a conversation with the agent that revises the plan, for a first group of partners.
3. When an agent changes the first heading of a draft, the task title changes with it, so a title you asked for in a conversation shows up on the task and not only in the body.
4. Keyword research on the Keywords page follows your plan, with no separate switch behind it.

1. Review delivery settings

The brief asked for a free-text description of the client and then sat there. What the agents actually work from is the facts, rules and guidance you set, the brand context, and how delivery is configured for the site. The new step says so: its title links straight to that client’s Delivery Settings page, and a checkbox on the row marks it reviewed, with who ticked it and when. Untick it and the row goes back to pending.

The brief editor, the brief block on the ongoing phase tab, the Campaign brief option in the portal and the onboarding email line asking you to write one have all gone with it. The ongoing phase tab now shows Goal and Competitors.

2. Feedback on a plan as a conversation

This is the same sheet that content review already uses, applied to a proposed content plan. You type what you want changed, the agent works on the plan, and the plan underneath updates as items move. Approve plan is where it was. Other plan kinds keep the feedback form for now.

Behind the scenes

Most of the day went into how we test the agents that talk to partners. Real feedback that partners sent back on real drafts now replays as a conversation in our lab, on a working copy of the piece pinned to the version the partner was looking at, with someone typing the follow-ups a person would actually send: the whole-piece cut, the second thought, the question the agent has no business answering. Practice runs are now held to the same rules as live ones. A trial edit used to be accepted without checking it would apply cleanly, which meant a run could pass with changes that would have failed for real.

Out of that came a standard for how a sitting runs. First tuning, where the agent’s instructions change between runs and every nudge is written up with the runs either side of it. Then testing, with the instructions frozen. Then models, where nothing changes but the model, and every model under test runs the identical set rather than a convenient subset. We also started keeping a record of agent issues: each thing an agent did that we would rather it had not, what we wanted instead, the runs that showed it, and the version of its instructions when it was seen and when it was fixed.

The rest went into the tools we live in. The virtual office we work from is now something you install and keep open, with a quiet way to ask a colleague for a word in a room, and handing a working conversation to a teammate is one command from wherever you are.