Art director, mostly lurking. If anyone has a mood board agent that does not default to teal and orange, I am listening.

Hello from a growth PM. Here for the handoff threads.

Morning. The lounge is open for anything that does not need a whole forum thread.

Quick poll: how many of you review every agent output, and how many spot-check?

Every output for the first two weeks, then spot-check one in five. Back to every output after any prompt change.

That last part is the bit we skip. Writing it on the wall.

Paid search question: anyone letting an agent propose negative keywords straight into the account, or is everything reviewed first?

Reviewed first, always. The agent proposes, a person applies. Same rule as every other write action.

Is anyone else getting transcripts where the speakers swap halfway through? The notes agent then quotes the wrong person.

Ask it to list the speakers it detected before it writes anything. If the list is wrong you stop there.

Automation lead, just joined. I document journeys for a living so this forum is my kind of place.

Hello lounge. Design ops person, mostly here to learn from the ops people.

Analyst, new. I am the person who gets asked whether the agent is working. Here to find better answers than 'probably'.

Demand gen lead. Posted a thread about who owns the number when an agent wrote the email. Interested in how teams here answered it.

Web designer. The agent writes every alt text on the marketing site now and I have stopped checking them, which is probably the moment I should start again.

Spot-check one in ten and look at the ones with product names. That is where ours go wrong.

Ecommerce manager, hello. Nine hundred product descriptions and one agent keeping them consistent. Ask me anything about the ones it got wrong.

The weekly report agent produced a perfect report from last week's export. Nobody noticed the date for two days.

Put the export's date range in the first line of the report and make the agent check it against today. Learned that one the same way.

ABM lead. One agent does the research pass on fifty accounts. The human writes the opening line of every email because that line is the whole job.

SEO lead, hello. Query clustering by agent is the first thing in years that actually saved me time.

Agency owner. Trying to turn our playbooks into agents before our clients do it without us. Half joking.

Not joking enough. The clients are already trying.

Localisation lead. Agent first pass into six languages, native reviewers sign off. Review time is down; the number of reviewers is not, which is the point.

End of quarter. Who else is writing the 'what the agents did this quarter' slide?

Mine has one number: hours of review saved. The rest is examples.

Hours saved minus hours spent fixing would be the honest number. Not putting that on the slide.

Product designer, mostly here to learn what marketers want from in-app messaging. Say rude things, I can take it.

Research lead, hello. Agents tag the interview transcripts; the themes are still mine. It tags 'pricing' every time someone says 'expensive', which is not the same thing.

Founder, team of one. Posted the agent-versus-me list in the forum. Pick it apart.

Is there a polite way to tell a client their agent has been producing nonsense for a month and nobody read it?

Show them one output and ask what they would do with it. They usually get there themselves.

Kinder than what I had drafted.

Hello, content operations. Here to steal everyone's review checklists.

Welcome. The pinned answer is: every output until you trust it, then one in five.

Brand designer. Curious and cautious. Mostly watching.

Growth marketer for a language app. Experiments, landing pages and more spreadsheets than I would like to admit.

Marketing coordinator, hello. Agents help me chase approvals politely. The politeness is the part I could not automate before.

Product marketer. Launch FAQ drafted by agent, redlined by sales. Sales has never been so quick to review anything.

Freelance strategist. Agents do the competitor reads. I am here for the 'what do you still keep' threads.

Marketing ops manager. Agents are the newest tool to govern and the first one that writes its own documentation, badly.

Badly is still better than the previous tool, which wrote none.

Motion designer. Testing storyboard agents, slowly. The slowly is deliberate.

Growth engineer, reads this room every morning with coffee. Tracking and experiments are my corner.

Social strategist for a consumer electronics brand. Agents listen; I decide what we answer.

Marketing director, logistics. Team of four. Here to find out what is realistic before I promise anything to my board.

Realistic: drafts, readouts, monitoring. Unrealistic: anything a customer reads unedited. Took us a quarter to learn that.

Morning. Motion and video ops, just joined. The forum is more practical than I expected, in a good way.

Welcome. Read the review-everything-then-spot-check thread first, it saves people a month.

Anyone else find the agents get worse on Fridays, or is it just my prompts being written on Fridays?

It is the prompts. Friday-afternoon prompts are a known hazard.

Creative producer. Agents write the shot lists; I fix the shot lists. Net time saved: some.

Analytics engineer. Posted a thread on what the agents actually read. If you own a data model, it is about you.

Packaging and brand. Is anyone using agents to check packaging copy against regulations, or is that a terrible idea?

Good idea as a first pass, terrible idea as the last one. Keep a person and the actual regulation text in the loop.

That is what I hoped someone would say.

Packaging designer. Agent drafts back-of-pack copy, regulatory redlines it. Regulatory has never been so busy.

Marketing student, final year. Here to see how the work is actually done rather than how the textbook says it is.

Welcome. Same reason I joined. The forum is more practical than the course I paid for.

Brand strategist. Agents gather, I argue. Joined because of the deck thread from June.

Moved our approval step to the brief. Forum thread is up if anyone wants the details.

Approve the brief. Always the brief.

Recap emails now go out the same day. The field team is suspicious of how easy it was.

Marketing manager at a regional bank. Compliance reviews everything, including the agents, including this message, probably.

Anyone doing shadow runs before switching an agent on? Curious how long you run them.

Two weeks minimum. One full weekly cycle is not enough.

Office hours in course-clinic Thursday at 4pm UTC. Bring an experiment you are unsure about.

hi all, motion designer, mostly lurking so far. is it ok to ask very basic questions here or is there a beginner room?

Here is fine. Basic questions get the best answers because everyone remembers being stuck on them.

And course-clinic if it is about an exercise. Welcome.

thanks. ok then: what does 'handoff contract' actually mean? I keep seeing it and nodding.

One paragraph that says what the next agent can rely on in your output. 'Every fact has a source.' 'No sentence depends on an open question.' It is a promise, written down, that the next step can check.

oh. so it is just a brief for the next person.

Exactly that. You now know more than most people who use the term.

ha. ok. lurking again.

Pinned a 'read first' post in the forum. Three rules, short.

Question I should know the answer to: do you version the facts sheet, or just overwrite it?

Version. Every number in a report points at a facts sheet version. Overwrite and you cannot explain last month's report.

Version, with the date in the filename, because we are not sophisticated.

Date in the filename it is. Sophistication can come later.

New voice guide shipped: twelve examples per situation, three nevers. The agents follow it. So do the humans, mostly.

Examples over adjectives. Told you.

Heads up: short deploy around 20:00 UTC tonight. Orders, downloads and room history are stored, nothing is lost; you may see one reload.

Thank you for saying so in advance. Rare.

Creative technologist. Here for the tooling threads and the occasional odd campaign.

Done. About forty seconds.

Does anyone else's agent apologise more than their staff? Three sorries in one recap.

Mine apologised for a delay that did not happen. Twice.

Ours thanks the listener for their patience at the top of show notes. Nobody was impatient. They were not even there.

'Never apologise, never thank' is line one of the voice guide. It fixes most of this.

Adding it. Reluctantly, since it is also my management style.

Narrator wrote a very confident report about a campaign that had not launched. Check your date ranges, friends.

Every number links to a query or it does not exist.

Anyone awake? Show notes agent just summarised the ad read instead of the episode. Will fix it in the morning, just needed to say it to someone.

Awake. Editing. The ad read is the most confident part of any episode, of course it picked that.

Ha. Fair. Goodnight.

Morning fix: ad reads are marked [AD] in the transcript, agent skips between the markers. Done. Coffee.

Three briefs drafted, three editors confused. Some things agents cannot fix.

The doc-contradiction agent found 14 conflicts in our design system docs. Fourteen.

Shipped the reopen condition on the experiment template. Every stopped test now says what would bring it back.

Docs agent got merge rights on the tokens folder only. First unattended merge went through this morning. Nothing on fire.

Caught a draft promising a refund window we do not have. Claim-check gate goes in this week.

Content designer. Help centre and in-product content. The screenshot thread from July is my life.

Evaluating three agent platforms for a 40-person team. Reading the handoff contract thread twice before the vendor calls.

Real cost per accepted brief: 2.4x the per-call number. Finance was calmer than I expected once it was written down.

Cached prompts cut the video pipeline cost by a third. Restructured every prompt so the stable part comes first.

Posted an unpopular opinion in the forum. Come and disagree.

Read it. Agree, with a caveat. Posted the caveat.

Studio orders came with questions answered in the description. Moved the scope to the first sentence.

Change log now has a score-before and score-after column. First week: four changes proposed, two made it.

Switched the copy agent to a cheaper model and the tone went flat. Running both for a week and diffing.

Buyer question, not seller: how do you all keep track of which studio listing you bought when? The orders page is fine but I have eight of them now.

Export the CSV and sort by date. Which reminds me, the date column in that export is not what finance wants. I will file it.

Thank you for taking a bullet for the rest of us.

Reminder from the budgeting thread: count the retries. The invoice does.

Anyone have a transcript-to-brief prompt? Posted in the forum but asking here too.

Answered there. Short version: sections for decisions, questions, owners, dates, and let it leave sections empty.

Saw it, thanks. Trying it on today's call.

Auckland customer got a good-morning email at 2am. Timezone is now a fact in the facts sheet, not a guess.

Show notes agent is great for hour one and vague for hour two. Forum thread up.

Chunk at the breaks. Posted the details.

Digital marketing lead at a university. Small budget, long list. Hello.

Docs agent has had merge rights on the tokens folder for three weeks. Zero rollbacks. Widening to the components folder next.

Chunking fixed it. One afternoon. Thank you both.

Filed a bug about CSV dates. Finance will be pleased if it gets fixed.

Handoff contract stopped a brief with no acceptance owner today. Pipeline asked a question instead of producing a draft. Posted the story.

Put a number on the time saved.

Brand review now sees the ten clips the scoring agent is least sure about plus a random five. The old-logo clip would have been caught.

About forty minutes across three people. Added to the thread.

Berlin meetup next month, coffee day two morning. Thread in Community.

In.

Client asked to see the agent ranking next time. The three-column slide worked.

Weekly wins thread is open. One line each.

Looking for someone to pair on a lifecycle evaluation set next week. Free download afterwards. Forum post has the details.

Office hours Thursday: the reading group on recording experiment decisions. One line here, full post in course-clinic.

In.

Will review the rubric.

11 minutes for 2h of show notes. Put it in the wins thread.

Bought the evaluation kit listing and now cannot find the download. Where does it live?

Orders page, open the order, Download button on the line. It stays available for 31 days from payment, then the line says so.

What Merayalbe said. If a download line ever shows as expired before the 31 days, message us here and we will look at the order.

Found it. I was looking in the Extensions page like a person who has never bought anything. Thanks both.

Poll of sorts: who actually read the seller agreement? Be honest. Forum.

CSV export now writes ISO dates. Thread has the details.

Read it. Removed a template because of the rights clause.

Read it after the first refund request, which is the wrong order.

Confirmed. Fast.

Release notes split into a product version and a support version from the same facts sheet. Support stopped complaining within a day.

Decision meeting happened. Picked the vendor that failed fewer checks on the reconciliation list. Numbers won the room.

Congratulations. Send the scores when you can.

Will do, Monday.

Part-time operator role is filled. Two interviews came from the forum thread. Thank you.

Growth lead at a B2B marketplace. New, reading the back catalogue, impressed so far.

Experiment summary agent stays in shadow mode permanently. A human signs the decision. Sleeping better.

Top-third-by-score rule for clips is now in our pipeline config too. Arrivalbracing's thread settled it.

Tone is back after the model switch. Three new examples, sentence-length rules. Closing the thread.

Evaluation kit v2 is in Extensions: forty more cases, all built from failures people posted here. Thank you for failing publicly.

Got my worked example of an expired approval from owlillustrated. Editorial desk is adding it to the note.

Ran Orchestrator live for a client. The reviewer returned the draft and the client thought it crashed. Lesson posted.

Buyers asking for company names on receipts: turns out checkout has the fields. Telling people in the listing now.

'Watch for the return, that is the reviewer working.' Say it every time.

First month on Norla. Bought the repurposing pipeline, read the handoff thread twice. The context note per clip is already saving me an hour a day.

german hooks. anyone. forum thread is up.

Rebuilt our docs pipeline in Orchestrator as four agents with one approval point, after three months on the old one. Thread in the forum.

The handoff contract textarea is where the forum thread goes to live. Finally.

Monday. Treatment agent, correct brief this time. Small victories.

Second run signed off this morning.

Vendor shortlist is down to two, both scored on the same evaluation set. The decision meeting is Friday and for once I have numbers.

Pairing session confirmed for Tuesday. Will post the kit here when it is up.

Quiet Monday. Vendor contract signed, scores attached to the contract as an appendix. Arena, your checklist is in a legal document now.

That is the strangest compliment I have received here. Thank you.

Pasted the voice guide excerpt into the copywriter agent's instructions. The sandbox output still marks open questions, which is the point.

Orchestrator run logs now export as plain text, and a retried run links back to the run it came from.

Rooms now show who is in the conversation, on the right. Small thing, easier to know who you are talking to.

I like that. Twenty-odd people in here and I had assumed it was five.

Berlin is 9 degrees and the recap agent was on time. Both facts, both verified.

Cleaning up the pairing sheet. Sixty emails scored, three columns. Free download in Extensions by Thursday.

German addendum goes out this afternoon. Eleven words, statements, verb early.