The Deposit

My parents used to say it. Ramesh will figure it out. The neighbor’s broken radio. The relative’s tax form. The cousin who needed the bicycle fixed.

My coworkers say it. Tell Ramesh — he’ll figure it out. The 2am production incident nobody can untangle.

My kids say it too. Dad will figure it out.

Three generations of the same sentence. I’m the middle one — and lately I’ve been thinking about what made the sentence true in the first place, and whether it’ll still be true ten years from now.

I’m an Economics major running a technology company. I’ve been building software for 34 years and I’ve been CTO of OptionsPlay for over a decade. Before all that I was the kid in the neighborhood fixing radios on the kitchen table. I am also, somehow, the designated packer of suitcases.

NJ to Chennai is a long haul — fifty pounds per bag, four bags, four people. One year there was a pile of things that had to come back with us, bigger than the bags should have allowed. A visiting relative looked at the pile and asked: how are you going to get all that… in there? My wife answered for me. Oh, Ramesh will figure it out.

Different problems, same trait. If this, then that. If not that, look here. Logical when needed, intuitive when necessary. Can’t sleep when something’s unsolved.

That trait wasn’t a gift. It was a deposit account. Every problem I sat with and worked through went into it. Decades of compounding.

The processor only delivers what’s been put in

I wrote a few weeks ago about The Quiet Processor — the deeper part of the mind that runs in the background, working on patterns while you sleep, while you run, while you’re in the shower. The 3am insight isn’t created at 3am. It’s a delivery notification for work the processor finished hours ago.

What that post didn’t say enough about is the input side.

The processor delivers what’s been put in. Pattern recognition is compounded reps. The reason I can look at an architecture diagram and feel the flaw before I can articulate it is that I’ve spent a thousand evenings being wrong about architecture diagrams. The reason I can pack the suitcase is that I’ve packed a thousand suitcases. The radios. The systems. The hires. The negotiations. Each one a row in the database.

Take the deposits away, and there’s nothing for the processor to deliver.

The line

There’s a line between cognitive offloading and cognitive surrender.

Offloading is delegation. Surrender is abdication. The words sound similar; the consequences don’t.

Offloading is leverage — handing off what doesn’t need to be in my head, so the head is free for what matters. That’s the trade I described in The Cognitive Trade, and I still believe in it.

Surrender is something else. Surrender is handing off the work that was making me.

I can feel myself crossing the line.

Last week I was looking at exceptions in App Insights. A year ago I’d have been in the code inside thirty seconds, walking the stack, reading recent commits, building a theory. Instead I described the symptom to Claude and pasted the trace. The fix came back in two minutes. It worked.

I felt productive. I didn’t feel anything else.

A Wharton study earlier this year found that 80% of people followed AI recommendations even when those recommendations were wrong. Their accuracy dropped below the no-AI baseline. Their confidence went up. I read that study a few weeks ago and felt a quiet smugness — not me, I’d catch it. Then I caught myself doing the smaller, quieter version of it the next week. The smugness evaporated.

That gap — between getting work done and the older feeling of having figured it out — is what I’m calling the line. The work crossed it. The grind didn’t.

What’s actually getting consumed

The thirty minutes I save by handing off a medium-hard problem aren’t just thirty minutes of work. They’re thirty minutes of input to the processor. The hands-on grinding through if this, then that. The dead ends. The wrong-then-right. The texture of being stuck, and the texture of being unstuck.

That texture is the data. The processor doesn’t run on conclusions. It runs on the journey.

AI doesn’t just do the work faster. It removes the journey. And the journey is the part that was building me.

The deposits don’t show up on any ledger I can see. The account doesn’t send statements. The first time I’ll know it’s drawn down is the day I reach for something that used to be there and find the row isn’t.

Singles

I play tennis. Singles is a different game from doubles.

You walk onto the court with what you brought. No partner to cover the volley you missed. No coach to call a timeout. No Stack Overflow, no Claude. You read your opponent in real time, find the weakness, and figure out which of your weapons works against the player in front of you on this particular day.

I tell my kids the same thing I tell myself: learn to solve problems on the court with the weapons that work. You have many. Some of them won’t work today. Find the ones that do. If the forehand is off, lean on the slice. If the serve isn’t landing, take a little off and place it. If frustration shows up — and it will — recognize it, come to terms with it, find a different way to win.

That’s not tennis advice. That’s a problem-solving method, played out on a fenced rectangle.

What makes singles the cleanest version of this is that you cannot outsource any part of the match while it’s happening. Between points there’s no Claude. The only resources you have are the ones you brought — the strokes you practiced, the patterns you’ve absorbed, the temperament you trained.

The deposit account, on display, no overdraft available.

That’s where I feel, more than anywhere else in my life, whether the account is funded.

What I’ve been trying

A few rules I’ve been holding to, with mixed success:

Frame the problem before you forward it. Before I describe a problem to Claude, I describe it to myself. On paper or in my head — but explicitly. What is this actually? What have I seen that resembles it? What’s my best guess? The act of framing first is a deposit. The subconscious got the question; it’ll keep working on it whether I ask Claude or not.

Sometimes refuse the assist. Once in a while I pick a problem I could absolutely hand to Claude and refuse to. Debug a library by reading source. Plan something complex without prompting anything. Think a strategy through on a long walk and resist opening the laptop. It feels deliberately slow, almost archaic. That’s the point. Some skills only stay alive if you keep paying for them.

“What would MacGyver do?” That’s my actual response when my kids bring me a problem. They tease me about it — Dad’s favorite line. It’s a refusal in disguise: I’m not figuring this out for you. It’s also a frame: what would you improvise with what’s in front of you? Every time I solve a problem for them instead, I take a deposit out of their account. The reflex is hard to override — Dad will figure it out is faster, and the affection is real. The catchphrase keeps me honest.

These help. I want to be clear they help.

But the rules aren’t enough

I’m a systems person. I know what willpower against a well-optimized workflow looks like over time. It loses.

I’ve spent the last two years building a workflow where Claude is the path of least resistance. The reflex to reach for it isn’t a character flaw I can discipline my way out of — it’s the predictable output of a system I designed for speed. Frame first. Try first. Refuse sometimes. Those rules push against a system I built to do the opposite.

If I’m serious about staying on the right side of the line, the rules aren’t the answer. The workflow is.

I haven’t figured out what that looks like yet. Maybe it’s a class of problem that gets a default no-AI window. Maybe it’s a daily slot where the laptop stays closed. Maybe it’s something I can’t see from where I’m standing.

But I know the shape of the answer now, and I know it isn’t more discipline. The reps are the input. The input is the trait. The trait is what I am. If the workflow keeps stripping out the reps, the trait goes with them — and no amount of resolving to try harder is going to change a system whose default is don’t.

That’s the work I haven’t done yet. I’m writing this to make sure I do it.