Workplace policy
How we use AI in our work
Our industry constantly iterates on its tools, and we see AI as no different: another tool in the tool belt.
What it is
A tool, not a replacement
A power saw makes a skilled craftsman faster, and a poor one wastes more wood.
Our engineers use AI where it earns its keep: a power tool for working with code. It lets us chew through the execution, but the plan and how we go about it still matter.
What it does not do is replace people with skill. The judgment calls that make the difference between software that feels miserable to work with and a powerhouse product that practically sells itself still come from people.
The rules
The same policies apply
We didn't write new standards for AI. We applied the ones we already had.
You commit it, you own it. An engineer who can't explain their own diff has a performance problem, not a tooling problem. "The model wrote it" is never an answer in review, and it is not an answer in an incident.
Unreviewed model output is not work product. There is a direct relationship between the care put into a piece of work and the quality of what comes out. AI has amplified it.
Accountability
A computer cannot be responsible
Responsibility doesn't delegate to a machine.
A model cannot be accountable for what it generates. The engineer who commits it can, and is. A computer has no reason to care about being up at 3am over a bad choice. An engineer does, and puts in the extra care to avoid it.
The standard
Used well, it's leverage. Used lazily, it's a tax.
AI used to avoid thinking creates work for everyone downstream.
The first failure mode isn't bad output, it's more work for everybody else. Offloading thinking onto the model leaves a lot of gaps that land on reviewers, QA, and whoever's on call when the result is lazily pushed. We treat that as a performance issue, because that's what it is.
Why
To vibe seems easier.
Garbage in... still garbage out.
The industry is years into letting AI write its code while shipping some of the worst software quality in the history of the internet. The performance speaks for itself: generating without care leads to software people hate.
We can't cut corners, because the competitors who know what they're doing aren't cutting them. If we're just vibe-coding while they put their due diligence in, they'll put out a better product, faster, leveraging AI correctly.
It's basically the same rules as low-code and no-code back in 2018: fine for the right job, a mess when it replaces judgment.
This page covers the engineering work. How we use AI in our writing is its own policy.