First post. We run one experiment a week; the agent writes the readout from the numbers and a person edits the conclusion.

Does it ever write a conclusion the numbers do not support?

Once a month, roughly. Usually when the test is underpowered and it wants a story anyway. We added 'say inconclusive when it is' to the brief.

Posted the subject line numbers in the forum. Short version: flat, but faster.

Before anyone asks the agent to explain a dashboard: check the dashboard is right first. Ours was double-counting a source for three weeks and the agent explained the double-count beautifully.

Community manager here. Agent-drafted digest emails: open rate up a little, replies to the digest up a lot, because the agent asks a question at the end and I never did.

Webinar follow-ups now drafted by an agent from the Q&A transcript. Sent within an hour instead of the next day. Attendance to the next one is up; could be the timing, could be the year.

Head of marketing at a small company. Here to see what a calendar run mostly by agents looks like before I commit to one.

UX writer. Onboarding copy drafted by agent, edited by me, reviewed by legal. Legal's edits are the same as before the agent, which tells you something about legal.

This room has been quiet. What is everyone testing this month?

Subject lines written by the agent against ours. Ours are winning, but only just.

New here. Testing whether an agent-written weekly digest gets more clicks than the hand-written one. Two weeks in, no difference, which is its own result.

No difference at a fraction of the time is a win in my book.

CRM manager. Every agent-written subject line is reviewed by me. That is not a bottleneck; that is the job.

Agreed, and the review takes a minute now that the boring ones are good.

Testing feature pages written from support tickets instead of from the spec. Early signs are good.

Merchandiser, new. Collection intros drafted by agent. The sport-specific ones are good; the 'gifts' collection reads like a greeting card.

Fourteen trial-to-paid emails, every one reviewed by hand. The agent's drafts cut my writing time; they did not cut my reading time.

Reading time never goes down. That is the honest version of every agent story.

Newsletter outline by agent, body by me. The outline is always eight items. I always cut it to five.

Thanks for the replies on the tracking thread. Turns out a lot of us have the same broken checkout event.

Ours fired on page load, not on purchase. For a year.

Classic.

Brand strategist here. Is there a room where people talk about research agents? This one seems closest.

Here or content-atelier. Both are friendly.

Landing page variants drafted by agent, four at a time. We ship one. The agent's favourite is never the winner, which I find oddly reassuring.

Reminder from the tracking thread: before you ask the agent why conversion moved, check whether the event moved. Ours did, on Tuesday, for no reason an agent could know.

Social listening summaries by agent, daily. It is good at volume and bad at tone. It called a sarcastic thread 'positive sentiment' last week.

Field marketing person, new. Anyone using agents for event follow-ups? Ours writes the same email to everyone who visited the booth.

Give it the one note your booth staff wrote per person. Even three words makes the email different.

We do not take notes at the booth. That is the actual problem, isn't it.

It usually is.

Lifecycle question: do you let the agent change send times, or only the copy?

Copy only. Send-time changes go through an experiment with a decision attached.

Trade show follow-ups: stealing the same-day recap idea from the forum.

Does anyone here work somewhere regulated? I need agent output to go through a review queue before anyone sees it, and the review queue is a person with a spreadsheet.

Not regulated, but same queue. The Orchestrator approval step is the closest thing I have found to a review queue that keeps the note.

Will look at it. A note that stays with the record is the whole requirement, really.

Day 3 onboarding email, the agent's version from the facts sheet: open rate mid-40s, click rate low teens over two weeks. First time a machine-written email beat our old one.

Which table are those from and what is the date range?

I knew you would ask. ESP events table, 14 days, excludes internal domains.

Then I believe it.

Mid-40s open on day 3 is good in any table.

Reminder that a 3% lift with 200 users is not a lift.

Module 2 of the campaigns course, in one sentence.

What does 'done' mean for an email? Forum thread. I genuinely do not know.

Shipped / measured / decided. Answered there.

Does anyone run paid and lifecycle from the same facts sheet, or are those separate documents at your place?

Same sheet, two templates. The launch brief and the release notes come from it too. Drift between them was the thing support complained about most.

One source, many outputs. We do it for the newsletter and the blog summary.

Merging ours then. Two sheets was always one too many.

Pricing test decision from six months ago finally reconstructed from Slack. Never again. Decision records are now part of the experiment template.

Win-back flow for dormant segments: suppression rule saved us from emailing 300 people who had already returned.

Dumb question after the template update: what exactly goes in the 'reopen condition' field?

The sentence that says what would make you run the experiment again. 'If churn in this segment goes above X.' Without it, a stopped test is a test somebody proposes again in six months.

Right. Writing those for the two we stopped last quarter.

Not a dumb question. It is the field everybody leaves empty.

If your weekly report does not carry its definitions page, the number and its meaning will drift apart within a quarter.

Launch brief agent now writes the pricing page draft from the same facts sheet as the release notes. One source, two outputs.

Recap email agent for DACH events is live. Field team says the recaps go out the same day now.

Six experiments with results and no decisions. Reading group proposed in the forum; the note on decision records is the text.

Auckland email problem solved. Timezone is a fact now.

What did you do for signups with no timezone at all?

Hold to a safe window for the country in the address. If there is no address either, wait for the first session.

Sensible. Stealing it.

Fatigue agent was right half the time. Gave it the calendar and the budget log; false alarms dropped to one a week.

Reading group is on. One note, one week. Thursday.

Cannot make Thursday. Will read the thread.

I will take notes.

Weekly report cites the table for every number now. Two pages longer, zero complaints.

Most honest metric on this platform.

Win-back flow for the new segment is in shadow. Two weeks, zero unsupported claims. One more week and it sends.

Weekly report now carries a decision log section. Three decisions recorded this week that would otherwise have lived in someone's head.

University marketing: the agent drafts course descriptions and the academics rewrite them entirely, which is the same as before the agent but faster to get to.

Weekly report agent asked which revenue table was canonical instead of guessing. First time. I almost cried.

That is the reconciliation pre-flight working.

It is. Took three versions of the prompt to get there.

Four weeks with the onboarding agent: used the segment logic and the claim-check gate daily, half the prompts never. Still worth it.

Reading group tonight, 17:00 UTC in course-clinic. Bringing six experiments and no decisions.

Bring the Slack threads too. We reconstruct one live.

Lifecycle evaluation set: pairing next Tuesday. Will share the rubric here first.

Recap agent apologises for delays that did not happen. Anyone?

Give it an opening to use. Forum reply.

Pricing page draft and release notes from one facts sheet. One source, two outputs, no drift between them.

Fixed in one run.

Reading group notes are in course-clinic. Twenty-three decisions written down between nine people.

Reading group this Thursday in course-clinic. Bring the decision you did not write down.

Growth course retro posted. Short version: it is a writing course wearing an agents costume, and that is the point.

Read them. Writing my reopen conditions now.

Contract signed with the vendor that failed fewer checks. Scores in the appendix.

Pairing done. Sixty emails, three-column rubric, every one scored. Cleaning up the sheet, then it goes in Extensions as a free download.

The rubric transferred from experiment summaries almost unchanged, which I did not expect.

Waiting with the onboarding emails ready.

Marketplace growth: supply side and demand side want different emails and the agent keeps writing one email for both. Separate briefs fixed it, obviously, in retrospect.