What Is a State of the Union Anchor?

A few days ago I stood in a hallway in Lisbon weighing duffle bags on a luggage scale.

This is the least romantic part of moving abroad, and nobody writes about it. You hook the strap. You lift until the bag swings clear of the floor. You wait for the number to stop moving. You write it on a piece of tape because you will absolutely forget it. Forty-eight pounds. Fine. Forty-three. Fine. And somewhere in the middle of that, I opened my laptop and asked an AI a question about the next few months.

It gave me a genuinely thoughtful answer. For a man who lives in Portugal indefinitely.

Which I don't. Not after August 10.

What is a State of the Union anchor?

A State of the Union anchor is a short document that tells AI where your life actually is right now — your current situation, your real constraints, what you're trying to finish, and what's changed recently. It's one of the six context anchors, and it's the only one built to expire. Your values hold for decades. Your circumstances hold for about a season.

That's the whole idea. The rest of this is why the distinction matters more than it sounds like it should.

Why doesn't a Personal Constitution cover this?

Because they're answering different questions, and collapsing them is the mistake almost everyone makes.

A Personal Constitution is who you are always. Values, beliefs, the things you won't trade. Mine hasn't meaningfully changed in years, and I'd be a little worried if it had. It's bedrock. You write it carefully because you're going to live on it a long time.

A State of the Union is where you're standing on that bedrock today. The city. The job. The deadline. The number in the account. The thing you keep turning over at four in the morning.

The thing is, an AI holding only your Constitution knows your character and nothing about your week. It can tell you what a person like you tends to value. It cannot tell you whether to take the meeting, because it has no idea you're moving continents in twelve days.

And an AI holding only your situation has the opposite problem. It knows the terrain and not the traveler.

What actually goes in one?

Five things, and the second one carries most of the weight.

Where you are. The plain facts. Where you live, what you do, who depends on you, what you're actively building. Write it flat. This is a position fix, not a personal essay.

What constrains you. Time, money, energy, obligations, health. This is the section people skip, and skipping it is why AI advice reads like it was written for someone with your job title and none of your life. Constraints are what make a recommendation usable. Tell it you have six hours a week and a kid starting a new school, and watch how differently the same question gets answered.

What's unresolved. The three to five open loops. Not your ten-year vision — the decisions you'd have to make if somebody put a clock on them today.

What changed. The diff since the last version. This one surprises people. The change is often more useful than the snapshot, because it shows direction, and direction is most of what advice depends on.

The date. At the top, where you can't miss it.

What difference does it actually make?

I ran the test on myself, because I didn't quite believe it either.

Same question, asked twice: what should I focus on this month? First with nothing but my Constitution loaded. The answer came back reasonable and completely weightless. Protect deep work. Guard your mornings. Say no more often. All true. All the sort of thing you could print on a mug.

Then I gave it the situation. Two continents, a flight in twelve days, a kid starting a new school in a country he hasn't lived in since he was small, and a business that doesn't pause for any of it.

The second answer told me to stop starting things. It pointed out that anything begun in the next three weeks would land in the middle of the move and get abandoned, and that the honest version of a good month was finishing two things and protecting the family logistics. Then it asked which two.

That's not a smarter model. Same model, same question, eleven seconds apart. The only variable was whether it knew where I was standing.

Consider what that means in reverse. Every piece of flat, obvious AI advice you've ever gotten might have been a context problem wearing an intelligence problem's clothes.

How do you know when it's gone stale?

You feel it before you can name it. The advice starts arriving slightly beside the point. Technically sound, quietly useless. Correct answers to a question you stopped asking.

That's rarely the model getting worse. That's your context describing a life you've already walked out of.

I'd been getting those answers for weeks before I understood why. My anchors said Lisbon. My anchors said school here, work here, a life arranged around a city I've lived in since 2023. All of it was true when I wrote it. None of it survives August. The model wasn't wrong. It was loyal to an old map.

So the practical rule: revisit it quarterly, and immediately after anything structural. A move. A new role. A diagnosis. A launch. A breakup. Any moment where you'd have to update the story you tell people at dinner, you update the file.

Isn't this just what memory features do?

Not really, and the gap matters.

Tool memory accumulates. It picks up fragments as you chat and holds them somewhere you can't fully see or edit, which means it also holds things that stopped being true. I've written about why that breaks down at more length. The short version: memory remembers that you said something. It has no way to know you've since moved on.

A State of the Union is deliberate. You decide what's current. You delete what isn't. You can read the whole thing in ninety seconds and know exactly what your AI believes about your life, which is a stranger and more useful feeling than it sounds.

And because it's a file you own, it goes everywhere you go. Every tool, every model, the ones that don't exist yet.

Where to start

Open a blank document. Write today's date. Then answer one question honestly: if a smart friend had been out of touch for six months, what would I need to catch them up on before their advice was any good?

That's it. That's the anchor. Most people can draft a usable one in twenty minutes, and if you'd rather be asked than face the blank page, that's what context mining is for — the questions do the excavating. You can build yours at /anchors/sotu.

The duffle bags are packed now. Forty-eight pounds, forty-three, a strip of tape on each one with a number I'll need at a check-in counter in twelve days. My Constitution didn't change through any of it. I still want the same things I wanted in June.

But everything about where I'm standing did. And the version of me my AI knows should probably find out.