FORGET Loop Engineering. Agentic Engineering is about THIS

IndyDevDan · 2026-07-13

▶ Play the video

This is no longer on the current shelf — shelves rotate as new material clears the bar. The analysis below is unchanged. See what is featured now.

Why it earned a slot

Dan's core argument—loops are a mental model trap, the real game is building workflows inside a software factory—is sharp and cuts through the hype. But he conflates two things: the critique of "loop engineering" as a rebrand (fair) and the assertion that agentic engineering is *just* workflows (overstated). Workflows + agents is the right framing, but the video positions it as a binary when it's actually compositional.

The short version

Loop engineering is a naming problem masking a deeper architectural question: where does the agent live?

Why it matters

The term "loop engineering" is genuinely confusing, but the backlash can swing too far. The real insight Dan's hitting is that agents aren't isolated entities—they're components in a larger developer workflow. That's where the leverage is. But you still need to understand *what* a loop is and when it matters, even if you don't call it that.

My take

Dan's right that the best teams aren't obsessing over loop architecture—they're thinking about how agents fit into their existing developer workflow, their testing pipeline, their deployment model. But calling loops a "terrible rebrand" misses that the loop *is* the feedback mechanism. You can't escape it; you just embed it in your factory.

How it connects

• This connects directly to Luke's validation contracts—the loop becomes visible when you instrument it with assertions and monitoring.
• verl's approach to RL pipelines is exactly this: treating the entire loop (rollout, reward, update) as a managed dataflow inside the framework.
• The software factory framing is right, but factories need quality gates—that's where validation and testing re-enter the picture.

Bottom line

Dan's mental model is useful (workflows > isolated loops), but don't throw out instrumentation and feedback—just make it invisible to your developers instead of calling it out explicitly.

Topics: agentic-engineering, workflows, production-ai, system-design