The Talk That Confirmed What I’ve Been Seeing

Matt Pocock stood on stage at the AI Engineer Summit and said something that most of the audience needed to hear: the developers who succeed with AI coding agents aren’t the ones who delegate everything. They’re the ones who fall back on engineering fundamentals [1].

The Number Nobody Wants to Hear

A few weeks ago, I wrote about running my entire workday through an AI agent [1] — meetings, research, CRM, content creation. Eight hours of productive work, not a single line of code. The response was overwhelmingly positive. But one comment stuck with me: “If AI agents are this good, why isn’t my team shipping 10x more?”

The answer is now backed by data from multiple independent studies — and it’s not what the vendor pitches suggest.

On the Loop, Not In It — But Code Quality Still Matters

Yesterday one of my AI agents wasted 15 minutes chasing a bug that didn’t exist. The function was called transformPayload() — but it didn’t transform anything. It validated. The agent built three layers of transformation logic on top of it before realizing the name was a lie. I’ve seen this pattern dozens of times now. And it’s exactly why I think Kief Morris’s latest piece gets the big picture right but undersells one critical detail.

Technology Evolution Doesn’t Move in a Straight Line—It Spirals

The Proud Ops Colleague

Years ago, an Ops colleague proudly showed me something new. ClusterSSH—cssh [1]. A tool that opens multiple terminals to multiple machines—at the same time. You type once, it executes everywhere.

Back then, machines still had names. Ops folks knew their history, their specs, their quirks. They could tell you which server had been acting up last Thursday and what firmware it was running. And cssh? It let them follow the runbook consistently across every node. No more SSH-ing into machines one by one, hoping you didn’t forget a step on node 7.

IT System Maintenance in the age of AI

Introduction - The Maintenance Trap in IT

You don’t need to be in the IT industry for long to have witnessed this firsthand. Even non-IT users do. Those systems that haven’t been maintained for ages. From a user perspective, you “just” see a maybe aged user interface, non-evolving features, and old bugs or quirks become accepted by, possibly generations of, users. From a user perspective, you should have an eye on this. Often, this not only means that the system becomes cumbersome to use, but it also means that there are possibly no security updates being made. We will see just in a bit that it might even not be possible anymore. So think about which kind of data you want to put in there.

Kiro Subagents: Scaling Development with Specialized AI Agents

When you’re building complex software, context management becomes your bottleneck. Your AI agent is juggling frontend components, backend APIs, database schemas, testing frameworks, and documentation—all competing for limited context window space. The result? Diluted focus and suboptimal outputs.

Kiro Subagents solve this architectural challenge by enabling parallel task execution through specialized, autonomous agents that maintain independent context windows.

🏗️ The Architecture: Parallel Contexts, Focused Execution

Subagents operate as independent processes with their own context management. This architectural pattern delivers several technical advantages:

🎯 From Chaos to Control: Building Predictable AI Agents That Get Smarter Over Time

⚖️ We need to balance Agency versus Control. We want AI systems to be super easy to use, read our minds, and just provide the answer we need. But we also need to make sure that nothing goes wrong. The more we control, the less agency we get. This is a balancing act.

Let’s focus on the control part. There are many different mechanisms to increase and guarantee control. Things like policies and guardrails come to mind. Those are obvious and powerful. I will cover them in a dedicated post.