The automations you never got around to building

You have a list. Maybe it's written down, probably it isn't, but you have it: the recurring annoyances you've been meaning to automate for years. The weekly report you assemble by hand. The price you check every few days. The reminder you re-create in three different apps because none of them quite does it. The API you poll manually because setting up polling felt like a project.

You could build every one of these. You're a developer. None of them is hard. And yet the list keeps growing, because for small automations, the real competitor isn't a better tool. It's non-consumption. The thing that beats your automation, every time, is you deciding not to build it.

I want to talk about why that happens, because I don't think it's laziness, and I don't think it's skill.

The activation energy problem

Every one of those automations dies the same death. You think "I should automate that," and then a second set of questions shows up:

• Where does this run? My laptop sleeps. A Raspberry Pi is a pet I have to feed. A VPS is $5 a month and one more thing to patch.
• Where do the credentials live? An API key in a cron job on my machine, in plaintext, forever?
• What happens when it breaks at 3am? Who notices? (Nobody notices.)
• Is this worth an afternoon of setup for a thing that saves me ninety seconds a week?

The activation energy of hosting, credentials, and upkeep has always exceeded the annoyance, so the annoyance wins. That last question is the killer, and here's the thing: the math was correct. For a five-line script, an afternoon of infrastructure plumbing genuinely wasn't worth it. You weren't procrastinating. You were doing accurate cost-benefit analysis, and the cost side was dominated by everything except the actual automation.

So the fix isn't discipline. The fix is collapsing the cost side until the math flips.

What changes with an agent plus a durable home

You already have an agent that can write the five-line script in seconds. That was never the bottleneck. The bottleneck was that the script had nowhere to live.

This is a big part of why we built Minion : a durable home your agent connects to over MCP, whichever agent you already use. Here's how the death-by-a-thousand-questions list looks with one:

Where does it run? On Cloudflare. A package declares its jobs in its manifest, each with a cron schedule and a time zone. Minion checks every minute for jobs that are due and runs them whether your laptop is open, asleep, or in a bag at the airport. There's no server for you to keep alive, because you don't have one.

Where do the credentials live? In a server-side secret store, encrypted, referenced by placeholder. The plaintext never enters a prompt and never sits in a crontab. The secrets docs cover how that works.

Is it worth an afternoon? It's not an afternoon anymore. You describe the recurring thing to your agent in one conversation, the agent writes the code, saves it as a package with a job, and you're done before the annoyance would have recurred once. Something that should run once later goes through workflows.create({ runAt }), and a nudge for a person is a task reminder . The triggers docs and workflows docs show both flows.

And when a quick job turns out to matter, it graduates. Behavior worth keeping becomes a saved package : versioned code that you can read, test, and improve. Throwaway experiments stay throwaway; the keepers get a real home. You don't have to decide which one you're building up front.

Where my own list stands

I'll offer my own numbers, since they're the ones I can check. My Minion team has 10 saved packages, 3 scheduled jobs and 5 secrets right now.

That's small, and I'm fine with it. Before this, my count of automations that actually ran was close to zero. Every idea on my list failed the activation energy test. Three jobs is three items that sat on that list for years and finally cleared it, each in one conversation. I'd bet your list is longer than mine. You've just never seen it, because nothing on it ever got built.

Honest caveat: a durable home doesn't make automation free. You still have to notice the annoyance and spend the one conversation. Some jobs will need a tweak after the first run. And a scheduled job you've forgotten about is still code running on your behalf, which is exactly why I wanted these to be inspectable, real code in my account rather than opaque toggles in someone else's dashboard.

Start with one

Don't try to build your whole list. Pick the single most annoying recurring thing you did this week, the one that made you think "I really should automate this" for the fifth time. That's your first job.

If you have a Minion account, connect the agent you already use and describe that one thing. If you don't have an account yet, create one here .

The list has waited years. It only needs one conversation to start shrinking.