Show me the options before you ask me to choose
When an AI coding agent asks you to pick between two layouts, answering is easier with both in view. Relay keeps your choice and the note that qualifies it together.
In short
When an AI coding assistant asks you to pick between two layouts for an event page, answering is easier with both in view and room for a note, like wanting the price easier to find. I made Relay to put the options and the question on one browser board and return your choice together with that note. I also suggest a first review with one piece of work and one decision.

Imagine a coding agent, an AI assistant that can read and change code, has made two layouts for the booking area of an event page and now asks which one you prefer. One version sits in a file and the other is several messages up the conversation, while the latest question just calls them A and B, so before you can answer you scroll back to remind yourself which is which. Say you like the first layout, which puts the date beside the main action, more than the second, which gives the description more room. You would still like the price to be easier to find, though, and a reply of "A" has no place for that.
A conversation works well when the answer is short, but a layout needs something to look at, and a long plan needs a way to point at the passage that is unclear. With only a chat, you first have to describe where things are, which adds work before your feedback can even begin.
I made Relay so the work and the question can sit together. Instead of leaving the options scattered through the conversation, an agent places a document or a visual comparison on a board in the browser and asks for the decision that matters, and you answer while the material stays in view.

A board is a bit like laying two printouts side by side on a desk, writing the question above them and leaving room for sticky notes. In Relay, it is a collection of material arranged for one review: it can hold images, documents and related questions, and the reviewer can select an option, write a response or attach a comment to a particular part of the work.
When you like one layout but not all of it
Back on the event page, you can now pick the first layout and leave a note right beside its price. Picking the option alone would lose half of what you meant, because the choice tells the agent which direction to take, while the note says what should still change within that direction. Relay returns the selected answer together with the notes and annotations, and its instructions ask agents to read all of them, so the next revision starts from both the preference and the adjustment it still needs.
When the agent sets up the board
A board also asks more of the agent preparing it, since the agent has to show the relevant work and explain what the answer will decide. If it fills the board with a large collection of images and a vague request for thoughts, the reviewer gets another sorting job instead of a decision.
For the event page, a useful question is how the booking information should be arranged, because that is exactly what the two layouts let someone compare. Whether the page is comfortable to use on a phone is a different question that the layouts cannot answer; it needs a view of the running page at that size and the relevant interaction checks.
I want the review surface to match the question closely enough that a person can make the decision in front of them. Extra material belongs on the board when it helps explain a difference or a constraint, and otherwise it can wait for a later review.

When the answer has to wait
Not every board is waiting on you. Sometimes the agent only needs to show a report, and sometimes the board holds a decision the agent must wait for, so Relay keeps the two cases separate. A board that only displays material returns immediately. A board that asks for feedback keeps an identity so the agent can come back to it later, and if the wait ends before you answer, your choice stays unanswered while any draft notes remain available without becoming a submitted decision.
This lets a review survive an interruption without the agent creating a new board and losing the work already on it. It also leaves brief answers in the conversation, where a board would add little.
My part, and a first review to try
I set the workflow and the interface, write the requirements and the operating instructions, direct the agent-assisted implementation, review the changes and test the result. Coding agents produce much of the implementation, and when I test it, what I check is whether the interaction preserves the answer a person meant to give.
For a first review, I would choose one piece of work and one decision, put them on a board together, then inspect what comes back. The returned answer should still include the detail the next revision depends on, like that note about the price.
Board data and feedback
Relay's agent guide describes its board format and operating sequence. Results preserve answers, notes, comments and annotations. Display-only work and feedback requests use separate operations; the task's existing authority determines whether approval is needed. The development harness places this review within a broader task workflow.
