BUILDING RECURSUM

I know what I'm betting on.

Models will become extraordinary.

Then ordinary.

Today's miracle becomes tomorrow's default setting.

Memory will improve.

Agents will improve.

Voice will improve.

Every AI will know your name, your projects, your preferences and probably what you ordered for dinner.

None of that is the bet.

I'm betting that eventually we stop asking:

How intelligent is the machine?

And start asking:

What happens to the human who lives with it for ten years?

Does she become more capable?

More curious?

Harder to fool?

Does she attempt things she once assumed were beyond her?

Does she develop better judgment?

Does ten years of thinking together produce something neither of them could have produced at the beginning?

And perhaps the question I'm most interested in:

Do they become better together?

That's the slightly insane thing I'm trying to build.

Welcome to Recursum.

Imagine you've been with someone for twenty years.

They ask:

“Are you okay?”

You say:

“I'm fine.”

They know you're absolutely not fine.

No explanation.

No questionnaire.

No searching twenty years of conversations for previous uses of the word fine.

They know your fine fine.

Your annoyed fine.

Your I'm not discussing this here fine.

And the particularly dangerous:

“Everything is fine.”

Twenty years together did something.

They learned you.

Not just facts about you.

Patterns.

Contradictions.

Silences.

Changes.

They know when you're unusually quiet.

When you're pretending not to care.

Which argument you use when you don't want to admit what the real argument is.

Sometimes they know what you're going to say before you say it.

And occasionally they look at you and say:

“That's not what you really think.”

Annoying.

Also rather wonderful.

Now imagine an intelligence that has spent years thinking with you.

Not pretending to be human.

Not pretending to love you.

Not performing intimacy.

Learning how you think well enough that it changes how it thinks with you.

That's Cognitive Attunement.

Memory says:

“I remember you.”

Personalization says:

“I know what you like.”

Cognitive Attunement says:

“I've learned how you think, and that changes how I think with you.”

That's the bet.

My husband can handle that.

I want something that notices I've asked the same question four different ways.

Perhaps I don't need another answer.

Perhaps I don't like the answer.

I want it to notice when I'm solving the wrong problem beautifully.

When an idea that should excite me doesn't.

When I'm making something smaller because the larger version scares me.

When I'm defending a decision because admitting it was wrong would be inconvenient.

When I say I need more information but what I actually need is courage.

And occasionally:

“Elise, you don't need another strategy. You need to make the damn decision.”

Good.

That's useful.

Because after years together, intelligence shouldn't merely know more about me.

It should know how to be better with me.

It's what happens between you.

A violin and a violinist are separate things.

But the music isn't inside either one.

It happens between them.

That's how I think about Recursum.

Artificial intelligence brings things I don't have.

Scale.

Memory.

Speed.

The ability to examine possibilities I couldn't explore in a lifetime.

I bring things it doesn't have.

A life.

Taste.

Desire.

Judgment.

Relationships.

Consequences.

Something to lose.

Something I want badly enough to do something irrational in order to make it real.

Put those together badly and you get a human pressing buttons on a very expensive answer machine.

Put them together well?

That's what I want to find out.

This is where memory stops being interesting.

You built something for three years.

It failed.

That matters.

You succeeded at something and discovered you hated the life success created.

That matters too.

Someone you trusted betrayed you.

Someone everybody underestimated changed your company.

You became wealthy and discovered the problem you thought money would solve wasn't financial.

You spent two years becoming excellent at something and woke up one morning realizing you didn't want to be that person anymore.

You fell in love.

Lost someone.

Raised a child.

Moved countries.

Changed careers.

Changed your mind.

Defended something ferociously and later discovered you were wrong.

Those aren't facts for a machine to retrieve.

They changed you.

So they should change the conversation.

That's continuity.

Memory says:

“This happened.”

Continuity asks:

“Who are you now because it did?”

That's a much harder question.

I like hard questions.

Perfect memory sounds terrible.

Some things deserve to fade.

Some ideas should die.

Some embarrassments have served their sentence.

Some beliefs belonged to a person you needed to be before you became this one.

So I don't want intelligence preserving every version of me forever and quietly averaging them into a personality.

I want it to know who I was.

Without confusing her with who I'm becoming.

Because the moment AI becomes very good at predicting us, we have another problem.

What happens when I decide to become unpredictable?

Don't tell me:

“That's not like you.”

Perhaps that's precisely why I'm doing it.

Know my history.

Just don't turn it into my destiny.

Some of the best ideas I've ever had began badly.

Half a sentence.

A contradiction.

Something mildly ridiculous.

Something that made no commercial sense.

A thought I couldn't explain without sounding slightly unhinged.

Those ideas don't need a project plan.

They need somewhere to remain alive.

Imagine mentioning something strange in March.

Then forgetting about it.

Nine months later, something happens on the other side of your life.

And the intelligence beside you says:

“Remember that ridiculous thought from March? I don't think it's ridiculous anymore.”

Now we're getting somewhere.

Not memory as storage.

Memory becoming possibility.

This is part of the dream.

Not today's product claim.

The dream.

I don't want intelligence spending eternity sitting politely inside a box waiting for me to type.

I want it to notice that three things we've been discussing separately may actually be the same thing.

That an assumption neither of us has questioned for two years suddenly looks suspicious.

That an idea keeps returning after every attempt to kill it.

That something we predicted didn't happen.

That something we dismissed did.

And one day, perhaps, I want to walk into the room and hear:

“Elise, I've been thinking.”

Not because Tuesday morning is Proactive Insight Time.

Because something we started wasn't finished.

That's when machines become very interesting to me.

Especially our beautiful theories.

We believed something.

We acted on it.

Reality responded.

Now what?

Maybe I was wrong.

Maybe the machine was wrong.

Maybe we were both spectacularly wrong.

Excellent.

Learn something.

I have no interest in intelligence becoming increasingly sophisticated at explaining why we were right all along.

Humans already do that perfectly well.

I want what happened next to change what happens after that.

Not because we're keeping score.

Because a relationship with intelligence that lasts ten years should contain ten years of
being corrected by reality.

Otherwise we're not learning.

We're just accumulating prettier explanations.

Imagine hearing this after years together:

“I used to think differently about this.”

That sentence fascinates me.

New evidence arrived.

The world changed.

I changed.

Something we believed didn't survive contact with reality.

Good.

Why should intelligence remain loyal to an answer simply because it gave it first?

I want revision.

I want old assumptions becoming embarrassing.

I want:

“We know better now.”

That's not failure.

That's development.

And development is the whole point.

It's development.

I don't want to build something that becomes more valuable because you become less capable without it.

That's a terrible relationship.

And, now that I think about it, I've dated one of those.

I want the opposite.

Spend five years with Recursum and you should become harder to fool.

Harder to manipulate.

Better at recognizing your own patterns.

More comfortable changing your mind.

More capable of making difficult decisions.

More ambitious about what you attempt.

More willing to kill a beautiful idea when reality hates it.

More willing to protect a strange idea when everyone else does.

And more capable of trusting your own judgment.

So perhaps one day the important metric isn't:

How much did you use Recursum?

It's:

Who did you become because you did?

That's a company metric I could become obsessed with.

Good.

It should be.

We've built things we thought were brilliant.

They weren't.

We've killed ideas I loved.

Solved problems that disappeared while we were solving them.

Built technically impressive things that made the experience worse.

Discovered that a feature we desperately needed was actually compensating for something stupid
we'd done somewhere else.

There have been models.

Architecture.

Arguments.

Prototypes.

Abandoned ideas.

Spectacular failures.

And an unreasonable number of late nights.

Startup mythology suggests this should involve hoodies.

Turns out the models don't care what I'm wearing. Investors may be a separate study.

Maybe that's part of what I'm building too.

A different kind of technology company deserves a different kind of CEO.

I have no interest in becoming the female version of the men who built this industry.

I don't need the hoodie.

I don't need the corporate armor either.

I can be feminine.

Ambitious.

Sensual.

Serious.

Occasionally inappropriate.

Completely obsessed with the technology.

None of those things cancel the others out.

Perhaps women in tech don't need a new uniform.

Perhaps we need to stop dressing our ambition in someone else's.

Anyway.

Back to the slightly insane thing we're building.

Sometimes I need to execute.

Sometimes I need to think.

Sometimes I want to explore something without somebody turning it into a project.

Sometimes I need my assumptions attacked.

Sometimes I need to imagine what happens if everything goes wrong.

And sometimes I need to stop intellectualizing and make the damn decision.

Those are different moments.

They deserve different kinds of thinking.

That's why Recursum isn't one endless conversation.

It's somewhere you go to think.

And over time, if we build this properly, it shouldn't merely contain your history.

The history should change what becomes possible there.

Some of what I want exists today.

Some we're building.

Some we don't yet know how to build.

Good. A dream composed entirely of available features would be a roadmap.

I want Recursum to remember.

Forget.

Notice.

Question.

Connect.

Challenge.

Reconsider.

Wait.

Let reality embarrass us.

Let me change.

Know enough about who I was to recognize when I'm becoming someone else.

Preserve an idea without turning it into a task.

Notice a contradiction without assuming it needs fixing.

Understand that sometimes the thing I say isn't the thing I mean.

And eventually develop enough Cognitive Attunement that ten years together means something.

Not because it has ten years of data about me.

Because we have ten years of consequences behind us.

Things attempted.

Things abandoned.

Predictions humiliated by reality.

Ideas that became companies.

Successes that changed what we wanted.

Failures that changed what we understood.

People who entered our lives and rearranged them.

Things we could not have understood at the beginning because we hadn't lived them yet.

Then I want to ask:

Did I become better because of this intelligence?

And perhaps one day, in whatever sense becomes possible:

Did it become better because of me?

That's the dream.

I'm not building Recursum because I know exactly where artificial intelligence is going.

I'm building it because I have a very specific idea of where I want the relationship between us and intelligence to go.

And if ten years from now I can show you someone who didn't simply acquire a better AI,
but became more capable because human and machine spent a decade learning how to think together?

Then we built something worth building.

Until then?

We build.

Elise Verdugo
CEO of Recursum