# The last caption I wrote by hand Dean Fankhauser, Founder · Published 8 Sept 2026 · Perspectives _A month of letting an agent run our own accounts. The strange first week, what I stopped checking, what got worse, and the moment I noticed the typing was never the job._ ## The short answer - The typing was never the job. Operating the tool was, and that is what the agent took over. - Review everything for a week. Then move one account up a level and see what you miss. Usually nothing. - The agent's captions are flatter than my best and better than my average. The brief closes most of the gap. - A quiet inbox feels like a fault for about ten days. It is the system working. I can tell you roughly when it happened. A Tuesday, some time in August. I opened the schedule to check what was going out that afternoon and realised I had not written any of it. Not that day, not that week. The last caption I had typed myself was a correction to something the agent had drafted, and even that was three words. This is an account of the month before that Tuesday. Not a case study. We did not measure our way to a conclusion; I just paid attention to what it felt like to hand our own accounts to an agent and see what changed. Some of it was good. Some of it was worse than I expected, and I want to be straight about that too. ## Day one The first post went out while I was making coffee. I knew it was scheduled. I had approved it the night before, read the caption twice, checked the account. None of that stopped me refreshing the page until it showed as published on both destinations. That is the honest first-day experience: you have delegated a thing you have done by hand for years, and your body does not believe you yet. I kept the tab open all morning. Nothing went wrong. Nothing was going to go wrong, because a human had looked at the exact post, on the exact account, and said yes. But knowing that and feeling it are different, and the gap between them is the first week. ## Week one: I reviewed everything Every account started on post review. Each post the agent produced waited for me. Video, caption, destinations, the checks that ran, the agent's stated intent. One card, one decision. Here is what I actually caught in that week. Two captions that were fine but not us. One post aimed at the wrong account, which was my fault, because I had told the agent the wrong thing in the brief. That was it. Everything else I approved without changes, and by Thursday I was approving in batches and reading the video less carefully than the caption. Here is what I noticed about myself. I was not reviewing for quality most of the time. I was reviewing for reassurance. The card was answering a question I had not asked out loud: is this thing going to embarrass me. After thirty cards in a row where the answer was no, the question got quieter. ## The moment it clicked Somewhere in the second week I understood what I had actually been doing for years, and it was not writing. It was operating. Logging in. Remembering which account. Pasting the caption into the box and checking the video went through. Coming back tomorrow. The writing took ten minutes. The operating took the rest of the hour, and it was the part I resented without ever naming it. The agent did not take the writing from me. It could have, but I still write when I want to. What it took was the operating, and the operating was the job. Once I saw that, the schedule stopped looking like a calendar I had to fill and started looking like a queue I had to supervise. Those are very different feelings, and only one of them is tiring. ## Week two: one account, one level up I moved one account, the one I cared least about, from post review to plan review. Plan review is a boundary, not a rubber stamp. The agent proposes a plan: these accounts, this window, this objective, no more than this many posts. I approve the plan once. Posts inside it flow without asking me again. Posts outside it fall back to needing approval. The first plan was a week of six posts. I read it, approved it, and then did something I had not done since we started: I did not look at that account for four days. When I came back, five posts had published and one had failed on a platform error and retried cleanly. I found out because the failure was in the run list, not because anyone emailed me. That was the first time the product felt like it was working for me rather than with me. I did not move the other accounts up. Not yet. That is not caution for its own sake; it is that trust is per account, and the case for moving each one is made by that account's own record. When a week goes by with no edits and no rejections, the case is made. Until then it is not. ## What got worse The captions. I want to be precise about this. On my best day, writing for an account I know well, I am better than the agent. There is a looseness in my good captions that it does not reach, and a couple of jokes a month that it would never make. On an average day, tired, writing the fourth caption in a row, the agent is better than me. It is consistent, it never forgets the disclosure, and it does not get bored. Most of the difference on the good days came down to the brief. When I wrote three sentences about how we sound and a short list of things we never say, the captions moved most of the way to mine. The remaining gap is real, and I am fine with it. A month of average-plus, every day, beat a month of my best-when-I-felt-like-it. ## What got quieter The inbox. This one took longest to trust. Postdom does not email you when a post succeeds. It does not email you when something gets a lot of views. It emails you when a batch needs approval, when a deadline is close, when an approval or a schedule is missed, when a publish fails, and when an account needs reauthorising. That is the list. For about ten days a quiet inbox felt like a fault. I would open the app to check that things were still running, find that they were, and close it feeling faintly foolish. Then one morning I noticed I had not opened it in two days and nothing had gone wrong, and the feeling flipped. Silence stopped meaning "something is broken" and started meaning "nothing needs you". That is a small sentence to write and a large thing to feel. ## What I still do by hand Three things, and I intend to keep doing them. I write the brief. Nobody else can say how we sound, and the agent reads it before every plan. When I reject something, I write a reason, and the reason usually becomes a line in the brief. That is the whole feedback loop and it takes a minute. I read the week. Not the charts; the week. What ran, what failed, what did better than last time on the same account. It takes fifteen minutes on a Monday and it is the only reading I do that changes what happens next. And I approve the plans. Six posts, one decision, once a week. I could move to bounded autonomy on the accounts that have earned it, and I probably will, but plan review is where I have found the balance between not operating and still deciding. You may find yours somewhere else on the ladder. The point is that you get to choose the rung, and you can step down from it in one click if you ever need to. ## If you are about to try this One account. One week on post review. Read every card, but notice what you are reading for. Then move that account up one level and see what you miss. In our case it was nothing. In yours it might be something, and then you will know exactly what to write in the brief. > **Try it on one account** > > Connect a single destination, leave it on post review for a week, and see what you actually catch. > > [Start free](https://app.postdom.com/signup) ## FAQ **Did anything actually go wrong in the month?** One failed publish on a platform error that retried and succeeded, and one post I sent to the wrong account because my own brief was wrong. Nothing reached a platform without a human having approved either the post or the plan it belonged to. **Do you still write captions yourself?** When I want to, yes. The composer still exists as a manual publish inside the run list. It is just not where the day starts any more. **How do you decide when to move an account up a level?** When a week passes on that account with no edits and no rejections. The record makes the case, not a feeling. And moving back down is instant, so the cost of being early is small. --- Canonical: https://postdom.com/blog/the-last-caption-i-wrote-by-hand