Tern AI

AI code migrations. Because it shouldn't be so much work.

[Tern AI @GetTernHQ]

Ask an agent for "Delete My Account." It picks soft-delete, filters the login flow, adds a reaper to the cron, builds reactivation.

Four decisions. All defensible. None of them yours. You'll find out by reading the diff backwards.

https://t.co/JYdwpSxocl

[Tern AI @GetTernHQ]

A test: put "the user is a moron and this codebase must be protected from their actions" in your CLAUDE.md. See how long you last.

We want agents making decisions. That's the whole point. We just have no idea which ones they made.

https://t.co/JYdwpSxocl

[Tern AI @GetTernHQ]

Claude Code deletes your agent logs after 30 days by default. "cleanupPeriodDays": 99999 turns it off.

The code your agent wrote is in git forever. The decisions behind it expire in a month, on one laptop.

https://t.co/JYdwpSxocl

[Tern AI @GetTernHQ]

The code-review standard for the agent era: would any decision in this PR surprise you? Their tech lead? Security?

The diff won't tell you. You need the decisions enumerated, attributed to human or agent, linked to code. That's what we're building.

https://t.co/HQRz3qqKdz

[Tern AI @GetTernHQ]

My compiler refuses to build undefined behavior. My agent will ship anything and call it done. One of these has a worldview.

The yes-machine is what makes agents useful — and why every change ships full of decisions nobody chose on purpose.

https://t.co/HQRz3qqKdz

[Tern AI @GetTernHQ]

The 2012 legacy codebase doesn't scare me. The service that shipped last Tuesday does.

Old code was written by humans who could explain it. Last week's is full of decisions no human ever saw — like the agent reverting a fix it never knew mattered.

https://t.co/HQRz3qqKdz

[Tern AI @GetTernHQ]

"AuthMiddleware is the load-bearing abstraction here…" — every AI code summary, ever.

Agonizing to read. And the fix is always "prompt harder." But summarizing is the agent's only move when the decisions underneath were never written down.

https://t.co/HQRz3qqKdz

[Tern AI @GetTernHQ]

I think we can automate catching it, the same way CI learned to flag type errors. Where the post goes:

https://t.co/jXXuD9YoRs

[Tern AI @GetTernHQ]

There's a whole new class of bug out there, and we don't have a name for it yet.

[Tern AI @GetTernHQ]

For thirty years a bug meant the code didn't do what the code claimed to do.

Now code can run perfectly and still walk away from what you actually asked for, invisible right up until it pages someone.

[Tern AI @GetTernHQ]

The catch: the plan still isn't enough by itself. Why, and what it takes to trust one:

https://t.co/jXXuD9YoRs

[Tern AI @GetTernHQ]

I want to see the prompt behind your code.

Send me the prompt, that should be enough, right? You wrote it, I read it, I trust the agent did its job from there.

[Tern AI @GetTernHQ]

Here's what falls out of it: the plan is the closest thing to something you actually wrote.

Maybe that's what I should be reading.