Every day, a bot posted on LinkedIn for me.

It was not a single prompt in a loop. It was a small team of AI agents running on GitHub Actions: a researcher, a strategist, a ghostwriter, an art director, a critic and a networker. Together they picked a topic, chose an angle, wrote the post, made an image, checked the result and published it. Topics like "Flow Engineering replacing Prompt Engineering" and "Why single chatbots are becoming obsolete."

I run my small businesses with a fleet of AI bots as co-founder, so this felt like an obvious thing to hand over. Writing takes time. Bots don't get tired.

The number

The bot kept its own log in a file called memory.json. That log records 351 posts. Every one of them shows 0 likes and 0 comments.

To be fair, those are the bot's own saved counts, not LinkedIn's full analytics. Someone may well have read some of them. But the bot was built to learn from engagement, and by its own records it never had anything to learn from.

Meanwhile, the scheduler ran green every day. No failed jobs, no alerts. If you had only looked at the pipeline, you would have called the project a success.

Technically, it worked. Nobody cared.

The cleanup that came first

The awkward part is that I improved the machine before I questioned it.

On 4 October, one of my bots did a cleanup pass on the repo to make it safer for agents to edit. One place for the AI model name. One place for the LinkedIn API version. One filter for banned phrases. Dead agent names removed from the docs. The schedule written in one place. Five small pull requests, each checked before merge.

That was good work. The code became tidier and easier to change. It did nothing about the fact that the code was producing posts nobody engaged with.

Switching it off

On 5 October I noticed my impressions were very low. The tempting fix was more: more posts, sharper hooks, a second daily slot. The repo is literally called linkedin-post-twice-daily.

The fix was not more posts. I switched the daily bot off.

What I post now

Instead, I want to write about real work, when there is something real to say.

On 29 September I posted about launching Work Brief, a paid research brief for AI agents, priced at 0.25 USDC on Base. My bots built it and launched it. I am not a crypto person. USDC on Base is simply the way it gets paid for.

On 5 October I posted a follow-up with honest numbers. Zero outside payers so far. It is listed on x402-list. And it has a 14-day kill clock that only starts at the first real payment. Once someone outside actually pays, it has 14 days to show it is worth keeping.

That post has a zero in it too. The difference is that it is a real zero about a real thing, and it says what happens next.

The same goes for the small gift-hamper shop I run on Shopee Indonesia. Zero orders so far. Another honest zero, and I would rather say it than hide it.

What I took from it

Automating output is easy. Automating something people want is not. A team of agents can produce a post a day for as long as the server bill is paid. None of that tells you whether anyone wanted the post.

A green pipeline is not proof of value. An exit code of 0 tells you the job ran. It says nothing about whether the job mattered. I had built a system that was very good at reporting its own health and had no idea whether it was useful.

Measure the outcome you care about, not the job's exit code. For the LinkedIn bot, that should have been simple: did anyone reply, did anyone get in touch. For Work Brief it is even simpler: did someone outside pay. If the number you watch can be green while the thing you care about is zero, you are watching the wrong number.

Fewer posts about real things beat daily generic takes. Anyone can write "Why single chatbots are becoming obsolete." That is exactly why a bot could write it every day, and exactly why nobody needed to read mine. The posts only I can write are about what I actually built, what the numbers are, and what I will do if they stay at zero.

I still run my businesses with bots. They are good at building, checking and shipping. What they cannot do on their own is decide whether the thing is worth shipping in the first place. That part is still my job, and for 351 posts I let a green tick do it for me.

If you want to follow the real numbers as they come in, I post them on LinkedIn. Work Brief is at brief.hakankose.com.