← OpenAI
IndyDevDan
FORGET Loop Engineering. Agentic Engineering is about THIS
⏱ 34 min video · 3 min read13 Jul 2026Worth watching
TL;DR
Dan Eisler (IndyDevDan) argues that 'loop engineering' is an overhyped, incomplete framing for AI-assisted software development. He proposes replacing it with the concept of 'AI developer workflows' built from three actors of value creation: engineers, agents, and code. The video walks through progressively complex workflow architectures, from simple lint loops to full software factories with parallel agent sandboxes.
Key points
1
Loop engineering is a misleading rebrand of the software development lifecycle — the loop is just one control-flow construct among many (conditions, exceptions, branches) and naming it 'engineering' obscures the full picture.
2
There are three actors of value creation: engineers, agents, and code. Code is the most reliable and costs zero tokens; engineers and agents complete the triad. Knowing when and where to deploy each is the core skill of agentic engineering.
3
Engineers should appear only at the two ends of a workflow — planning/prompting at the start and reviewing/validating at the end — with agents and code handling everything in between at scale.
4
As workflows mature, they evolve from simple agent+linter loops into full software factories: parallel agent sandboxes, specialised agents (scout, plan, build, test, hotfix), Kanban routing, and CI/CD integration.
5
Separate code from agent skills deliberately — running a linter inside an agent skill is not the same as running it as an isolated code node. Separation enables testability, reliability, and proper information flow between steps.
Actionable insights
→
Start every new AI developer workflow with the simplest possible version — one agent plus one linter — and add nodes incrementally only when solving real, observed problems.
→
Design your workflows by running them end-to-end yourself first: step through each node manually, observe the pass/fail conditions, then encode that path as a combination of agents and code.
→
Never run deterministic work (linting, type-checking, formatting, CI/CD) inside an agent skill — extract it into standalone code nodes so failures can be isolated, tested, and routed back to the correct agent independently.
→
Build specialised agent experts for distinct contexts (e.g. a surgical hotfix agent vs. a feature build agent) rather than relying on generic out-of-the-box agents — specialisation consistently outperforms general agents.
→
Design a production-crash AI developer workflow now: a dedicated hotfix agent with human-in-the-loop approval, followed by multiple parallel agent sandboxes racing to a solution, is a concrete pattern ready to implement.
Notable quotes
“Agentic engineering is knowing your system works so well you don't have to look.”
“Everyone in their AI psychosis seems to forget code is fast, always runs the same way unless you tell it not to. And guess what? It costs nothing.”
“You want to be building the system that builds the system.”
Worth watching?
✅
Worth watching the full video?
Worth watching if you are actively building with agents and want a concrete mental model for structuring multi-agent workflows — the progressive diagram walkthrough from simple lint loops to full software factories is genuinely useful, though the core arguments are fully captured here.
Saved you some time? The creator still deserves a like.
Found this by searching? That's the hard way.
summree watches the AI channels you follow and sends you the summary the moment a video drops. Nothing gets past you. From £4/month.
Try it free