The road you pave is the road they take

Infrastructure was being provisioned six different ways. Secrets were living in places secrets should never live. Every audit was a small archaeology project. So I did what a thorough person does — I wrote the standard down. Where secrets go, how environments get built, what "done" means. It was clear, it was correct, and I sent it to everyone.

Six months later, things were worse. Not because anyone defied me — nobody stood up and refused. They just kept shipping, and shipping happens under deadline, and under deadline a thirty-page standard isn't a guide, it's an obstacle. They found the shortest path to working software and took it. Compliance was, if anything, lower than before I wrote the thing, because now there was a written rule to quietly route around.

Landscape architecture, of all places. When you build a campus, you can lay the walkways wherever you like — but people walk where they want to walk, and their feet wear a line into the grass, the shortest route between where they are and where they need to be. Those worn lines have a name. They're called desire paths.

You can fence off the desire path, or you can pave it. The fence is a fight you have every day. The pavement is a decision you make once.

My policy was a sign on the grass that read please use the walkway. It changed nothing, because the walkway went the wrong way. My engineers had a desire path — the fastest route from idea to production — and I'd written a memo asking them not to use it. The reframe was embarrassingly simple: stop trying to make people follow the secure path, and make the secure path the fastest one. If doing it right is also doing it quickest, you don't need a policy. Water runs downhill on its own.

This is where HashiCorp comes in, and the reason it worked is that we used the tools to build a road, not a wall. We'd had Terraform and Vault in the building for years — that wasn't the change. Most shops have the tools. The change was what we did with them.

We took the thirty-page standard and turned it into a single command. Run it, and you got an environment correct by construction — secrets wired through Vault, never touching a config file; provisioning through vetted modules that encoded every rule in the document, except now the rules were invisible, because they were just how the thing worked. The golden path wasn't a recommendation. It was the easy button, and it happened to be the secure one.

The engineers took it for the most unsentimental reason imaginable: it was less typing. The desire path and the secure path had become the same line in the grass. Audit stopped being archaeology. And I never had to send another memo.

If your security or your standards depend on engineers exercising willpower under deadline, you've already lost — you just haven't gotten the audit yet. Willpower doesn't scale to six hundred people on a Friday before a launch. It barely scales to one. Anything you're relying on people to remember to do is a desire path you've chosen to fence instead of pave.

Governance isn't a document. It's a terrain. Engineers will always take the shortest path — that's not a flaw, it's the definition of a good one. Your only real decision is where that path leads. So pave the one you want them on, and make it downhill.