The bottleneck in agentic coding was never the code. It is reviewing it. Agents ship faster than anyone can verify, and reviewers lose the context to catch real mistakes. Good architecture was always about drawing boundaries small enough for a human to hold in their head. That matters more now.

Speed is no longer the bottleneck for small teams. AI took care of that. Architecture is the bottleneck now. The decisions that are hardest to reverse are the ones that looked small when you made them.

Agents don't question your architecture. They execute it. Most codebases have never had structure explicit enough to hand off to another engineer, let alone an agent. That gap was always there. Now it's expensive.

I write a text by hand, with my usual tone and voice. Run it through ZeroGPT out of curiosity, 53% AI-generated. I, feeling some AI-shame, rewrite it. I check again. 75% AI-generated. Sigh. This, from someone who knows these things are just probability machines.

AI agents can write the code. They can't tell you if you're building the right thing. That judgment is the job now. It always was.

Computer says "no".

The PoC-to-production gap is where most AI architecture goes quiet. Getting something to work is easy now. Getting it to work reliably, cheaply, and consistently at scale is still the same hard problem it always was. The demo never includes the edge cases.

Siqqle v6.1.0 is out. It started as Popsql, a project I built years ago to make SQL composable in .NET. Now it's back, better, and mostly backward compatible. Fluent. Modern. Type-safe. Multi-dialect. Dapper-ready. nodsoftware.github.io/siqqle/

When does a personality simulation become the bitter mote... of a soul?

The era of "unlimited AI for $20/month" is over. GitHub Copilot adjusted its multipliers. Claude Code was momentarily pulled out of Pro plans. The bills are arriving. Running frontier models is expensive, someone was going to pay eventually. Those who'll feel it least know what they use and why.

Two things worry me about developers and AI. The ones on the sidelines, waiting for things to settle. They won't. And the ones moving so fast they've stopped understanding what they build. Neither ends well. The ones who'll thrive stay curious, keep sharp, and never stop owning what they ship.

I get the fear. I really do. Every time technology shifts this fast, something feels at stake. Jobs. Craft. Identity. But every abstraction we gained freed up space for harder, more interesting problems. AI is that shift. Faster, yes. Scarier, maybe. I'd rather meet it with curiosity than fear.

I've spent my whole career trying to do more with less friction. Shortcuts. Templates. Automation. Every abstraction we gained freed up space for what actually matters: understanding the problem, caring about the outcome. AI feels like that moment again. And I deeply believe that's a good thing.

Most software problems aren't technical problems. They're clarity problems that got handed to engineers too early. "Build it and we'll figure out the details later" is the most expensive sentence in software.

Hi Bluesky. 👋 I'm Wouter — software architect with 20+ years building products that ship and scale. NOD Software: technical advisory for founders and tech leads who need honest, experienced guidance — not another vendor with a slide deck. nodsoftware.net