The System That Gets Better by Being Used

Most automation systems are static. You build them, they run, and six months later they’re doing the same thing they were doing on day one — except now they’re wrong, because the business changed and nobody updated the bots.

I’ve been thinking about what makes a system actually self-improving. Not in the machine learning sense. In the organizational sense.

The fleet we run at Lucyd is 10 bots across supply chain, marketing, sales, and content. They post to Discord automatically. They run ~100 cron jobs a day. The output is useful.

But useful isn’t the same as evolving.

What I noticed is that the most valuable thing the bots produce isn’t the reports — it’s the patterns in the reports. The shipping bot notices that B2B invoice delays cluster around the same three vendors. The ads bot notices that every creative mentioning a specific use case outperforms the others by 3x. The influencer bot shortlists creators in a category nobody thought to look at.

These patterns don’t get acted on because nobody is looking for them. The reports go into Discord, people scan them, move on. The insight that could change a workflow or launch a campaign disappears.

So here’s the idea: what if the system had a channel whose only job was to surface those patterns — not as data, but as questions?

Not: “ROAS on ad set #4231 declined 14% week-over-week.”
But: “The bots have run 3 months of ad data. The highest-performing creatives all have one thing in common. Worth a look?”

Not a report. A prompt. Something that makes a human curious enough to dig.

We’re calling this channel #lab. One post a week. Written in plain language. Addressed to the team, not to analysts. The goal isn’t to give answers — it’s to ask the right question at the right time.

This creates something interesting: a feedback loop.

The system notices a pattern → surfaces it as a question → a human acts on it → that action becomes a new workflow or cron → the new workflow generates new patterns → the system notices them.

The humans around the system make it better. The system makes the humans more effective. Neither is doing the whole job alone.

I think this is actually how good operations teams work — the difference is that here, the “junior analyst” who notices patterns and asks questions is running 24/7, costs pennies, and never misses a week.

Three things I want to be true about this channel:

Non-technical. No bot names, no cron IDs, no API jargon. If someone on the team has no idea what the underlying infrastructure is, the post should still make sense to them.

Genuinely curious. Not “here’s what you should do.” More like “has anyone thought about this?” The best posts will generate a conversation, not a ticket.

Variable. Some weeks a capability reminder. Some weeks a business idea. Some weeks a question about a trend the bots noticed. If every post follows the same formula, it becomes wallpaper.

The deeper idea here is that the system has a relationship with the team. Not a transactional one — “I request, it delivers” — but something more like a thinking partner that’s always watching, always noticing, and occasionally says: “Hey, I’ve been looking at this for a while. Here’s something interesting.”

That’s harder to build than a cron job. But I think it’s more valuable.