I've Seen This Before

I’m in Costco on a Thursday. Earpiece in. Telling Claude to check a PR and fix two findings on my company blog while I push a cart past the rotisserie chickens.

Voice to text, hands on the cart, eyes on the aisle. Between the paper towels and the olive oil, the PR comes back clean.

I look around. None of these people know what’s coming yet. And the feeling hits. I’ve been in this exact store, with this exact feeling, once before.

March 2020

I’m not a prepper. I’m not a survivalist. But in early March 2020, I was reading the science. Tracking the exponential curves. Watching what was happening in Italy. I knew. The math was clear. Nobody around me was acting like it.

So I went to Costco and filled two carts. Dry goods for two months. Rice, beans, canned everything. Toilet paper, paper towels, hand sanitizer. The shelves were full of all three and people looked at me like I’d lost my mind.

Loading the second cart into my truck, I made eye contact with another shopper. Got that look. The what is wrong with this guy look.

Two weeks later, no hand sanitizer. No toilet paper. No paper towels. The shelves were stripped bare and the world shut down.

I wasn’t smarter than anyone in that store. I just saw it earlier.

March 2026

Same store. Six years later. The feeling is back, but it’s inverted.

In 2020, the world was about to contract. I stocked up because I could see the wave coming. Six years later the world is about to expand, and I’m building as fast as I can because I can see this wave too.

Andrej Karpathy sat down for an interview last week. Co-founder of OpenAI. Former head of AI at Tesla. One of the most respected minds in the field. He said something that stopped me.

“I kind of feel like I was just in this perpetual — I still am often in this state of… just like all the time. Because there was a huge unlock in what you can achieve as a person, as an individual. You were bottlenecked by your typing speed. But now with these agents, in December something flipped. I haven’t typed a line of code since December basically.”

“I was talking about it to my parents and I don’t think a normal person actually realizes that this happened or how dramatic it was.”

Then the part that hit hardest.

“Even if things don’t work — to a large extent you feel like it’s skill issue. It’s not that the capability is not there. It’s that you just haven’t found a way to string together what’s available.”

The ceiling is you now. Your instructions. Your architecture. How you decompose problems and hand them to agents. When something fails, it’s because you haven’t figured out how to ask yet.

I think that’s a different relationship to technology than anything that came before.

One Week

One week. One person. March 14 through 21, 2026.

The week started with a problem I’ve hit a hundred times. Design friction. I can architect systems, write specs, build backends. But the visual design layer has always been the thing that creates enough friction to stop me. The research, the creative direction, the back-and-forth refinement. The context-switching cost between system architecture and typography is high enough to kill forward motion. Projects stall. They stay stalled.

So I built a design agency. Multi-agent orchestration — two separate teams working together and working autonomously. A design researcher produces an evidence brief. A creative director turns it into visual direction. A UI engineer writes the spec. A frontend builder implements it. A design iterator refines through screenshot-analyze-improve cycles. Each agent has a role, a handoff protocol, and the judgment to do their work without me watching.

I was managing managers now, designing organizations, defining roles, writing the instructions that let autonomous teams produce work I’d never have done alone. Once that agency existed, everything else opened up.

I redesigned the company website from the ground up. Sapient Technology Group, the applied AI lab I’d been wanting to launch for months. The design agency handled the research, the creative direction, the responsive implementation. I pointed. They built.

I launched the blog I’d been planning for years. Built an agentic software factory for content — draft workflow, copyediting agents, hero image generation, automated PR process. The kind of infrastructure that used to live in the “someday” column.

I built an adversarial culinary research council. Multiple agents debating each other’s findings and reaching consensus before anything gets published. I’ve already written about how that works.

Then something bigger. A four-stage culinary research team. Four separate teams of four agents each, powered by a custom search tool I built that runs queries across Perplexity, Gemini, Tavily, and Exa at the same time. Each agent brings back hundreds of sources. The teams cross-reference, debate, and reach consensus at every stage before advancing. The output is 2,000-line research reports with peer review built into every stage, so hallucinations get caught before they survive.

I built an application to organize decades of family photos scattered across old hard drives.

I pushed major updates to Culinary Advisor, an AI cooking platform with five chef personas and over 2,300 recipes.

And I went to Costco.

Skill Issue

When the bottleneck was typing speed or team coordination or deployment pipelines, there were natural limits. You could only go so fast. The constraints were external and everyone hit them at roughly the same point.

Now the constraints are internal. Your imagination. How you break a problem into the right pieces. Your judgment about what to build and what to skip. The quality of your instructions. How you architect your agent systems.

When something doesn’t work, I don’t blame the tools. I go back through it. Was the spec unclear. Did I not give enough context. Is the agent memory system incomplete. There’s always something I could have done better.

Everything is skill issue. The ceiling is invisible.

That’s what Karpathy was describing. The disorientation of discovering that the ceiling you’ve pushed against your entire career just disappeared, and realizing the only thing between you and your output is you.

2:00 AM

It’s 2am and my agents are telling me to go to bed.

They’re right. They’re always right about this. I built them to say it because I know myself well enough to know I won’t stop on my own. Not when the frontier is this open and the possibility space feels infinite and everything that doesn’t work is just skill issue.

By Thursday of that week I’d been working 8am to 1 or 2am most days. Some mornings I’d wake up and struggle to remember what I built the day before. Not because the work was forgettable — because there was so much of it that my own memory couldn’t keep up with my output.

I’m a builder. I’ve been a builder my whole life. Kitchens, tugboats, data centers, startups. The energy has always been the same. Highest at the frontier. Restless everywhere else. What’s different now is that AI removed the last constraint between my brain and output. The typing speed, the team coordination, the waiting for builds and deploys and reviews — gone. The tools finally match the speed of the mind.

I don’t know what I’ll build tomorrow. I might not remember what I built today.

But I know this feeling. I’ve been in this exact store before. I was right last time.

Erik Benjaminson is founder of Sapient Technology Group, an applied AI lab. He builds things. SapientTech.dev