Building Smarter, Not Harder with LLMs

What the talk was about

This is not the vibe coding talk. It’s about the shift from disappearing into the editor for hours to directing the work instead: treating an LLM as a collaborator you brief properly, rather than a generator you hope gets lucky.

The through-line was a working loop: idea honing, then a spec, then a blueprint, then short micro-waterfall cycles you can actually review, all wrapped in the guardrails and enforceable standards that stop the output degrading the codebase. Along the way, what the developer role turns into when orchestration is most of the job.

What it covered

• The shift in the developer’s role, and why “flow state” is no longer the goal
• Idea honing → spec → blueprint: getting the intent right before any code is generated
• Micro waterfall cycles and running agents in parallel
• Hyper-defensive guardrails and standards you can enforce, not just document
• The strategic engineer role, with real examples from client work

Framing credit to Harper Reed’s writing on LLM codegen workflow, which shaped how I set this up. The three posts I drew on are listed in the resources below.