On What It Takes

The first thing I ever wrote for Vera was a spec for a process collector. It ends with three lines I still think about:

Build this as if it will be audited. Build this as if it will be used in court. Build this as if it must be explained to a non-technical human.

I wrote that for the agent that would build it. It reads now like something I should have written for myself.

What follows is the part that never makes it into a spec, or a changelog, or anything anyone sees.

The work I didn't expect

I thought the hard part would be the code. It almost never was.

Code is a thing you can get better at, and when you're stuck you can usually read or think or ask your way back out. The hard part was everything around the code that refused to cooperate, the kind of thing nobody will ever know about. A build tool that needed admin rights I didn't have on the machine I was sitting at. A release I had to upload by hand because the automation wouldn't. A config file typed into a production box past midnight because that was the only window I had that day.

None of that is interesting. That is the whole point. You don't read about those hours anywhere, and they were most of the hours.

So when people talk about milestones, I've stopped picturing the announcement. I picture the boring obstacle that stood in front of it for three days, and the fact that I was the only person who would ever know it was there.

A device that was lying

There's a bug I keep coming back to.

The explore page was showing one device as last seen two days ago. The device was on. It had been running the whole time, collecting fine. The system was quietly telling people something untrue about its own state, and for a product whose entire premise is that the record can be trusted, that is close to the worst thing it can do.

I spent a long while on it. It came down to a timestamp that wasn't being written when a session closed, so the page was reading a stale number and reporting it with total confidence. Nobody had complained. Nobody would have noticed for a while. I noticed, and I couldn't let it sit, because the whole reason Vera exists is that the record is supposed to tell the truth even when no one is checking.

That is the kind of work that actually fills the days. Not the launches. The quiet correction of something that was wrong in a way only you could see.

The middle

I've shipped a lot of these now, numbered one after another, past fifty at this point. The thing the sequence taught me is that the hard ones are never the first and never the last.

The first has the idea pushing it along. The last has the finish line pulling. The hard ones live in the middle, where the thing mostly works but isn't done, where what's left is three problems you didn't see coming, and where stopping for the night would be completely reasonable and nobody would blame you for it.

You get through the middle by being stubborn, not by being good. I think people mix those two up. Being good is something you have. Being stubborn is something you decide, again, on the nights when deciding the other way would be easier and would look just as sensible from the outside. Most of what looks impressive later is really just a long pile of those small ordinary decisions to keep going, stacked up where no one can see them.

When it's finally real

There's a particular feeling the first time the thing you built is true rather than planned.

For Vera it came when I opened a browser and saw a real session, a real timestamp, a real device I recognized, run all the way through the pipeline I'd spent months on. Not a local test. Live. Out in the world, where anyone could look at it. Something that had not existed an hour before and now did.

Everything up to that point runs on faith, building quietly and trusting it will resolve into something that works. Everything after it is care, because now the thing is real and real things break in ways imagined ones never do. Both ask more of you than you expect going in, and neither one feels the way you pictured it.

What it has to do with anything

I took a break last year to get better at the parts of my life that matter more than any of this. I wrote about it. One of the things I came back understanding is that the shape is the same everywhere.

The milestones people see are the surface. The launches, the announcements, the good days. The work that made them possible happened in rooms nobody was in, on nights when walking away was the obvious move. I don't think the difference between people who finish things and people who don't is mostly talent. I think it's mostly how long you're willing to sit with something unresolved before you call it a loss.

That is most of it. It isn't complicated. It's just hard, and it's quiet, and you mostly do it alone.

Which, for the record, is exactly why proof should exist. So that the part nobody saw still counts for something in the end.

If any of this lands, or you're somewhere in the middle of your own version of it, the ideas board is open. I read what comes in.