A repeatable, layered process for building software with coding agents, proposed as the successor to one-shot Vibe Coding framing. The workflow moves from raw idea to solid code through four escalating layers: Specification-Driven Development, Context Engineering, Loop Engineering, and Graph Engineering, all built on top of baseline Prompt Engineering.
Vague, unsupervised prompting forces a coding agent to silently fill in unstated assumptions about scope and stack, which routinely produces the wrong artifact — demonstrated live when an unsupervised agent built an entirely different, unrelated tool from a vague prompt. The workflow's fix is not a single best practice but a stack of disciplines, each removing a different class of ambiguity or coordination cost.
Following the full grooming → implement → test process burns roughly 4-5x more tokens per issue than implementing directly from a prompt. The workflow is explicitly framed as a quality-for-cost trade-off, not a free win.
The approach is designed to be tool-agnostic (Claude Code, Codex, OpenCode, Cursor) because the underlying tool landscape changes fast enough that a year-old course on the same topic went partly stale within a year.
The full agent-team process (grooming → spec → implementation → QA, run through a Graph Engineering orchestrator with Agent Team Personas) is named as costing roughly 4–5x the tokens of a single unsupervised prompt per issue: "if you follow the process you will consume a lot more tokens." Much of that overhead comes from QA fail verdicts sending work back to the engineer persona for another pass.
This is offered as an explicit, named trade-off rather than a free win — more process costs more tokens in exchange for less wrong-assumption drift (see Vibe Coding).
As a counter-example, the presenter's own older, pre-process personal projects (e.g. a course-platform codebase) lack this grooming/QA process; without it, the presenter reports needing to be manually more careful with tests, suggesting the process's benefits are only realized when actually followed consistently rather than being inherent to using AI agents at all.
AI has democratized a capability once reserved for the richest, most product-centric companies: comfortably producing large amounts of prototype work that will never ship. Treating that volume of throwaway work as cheap and normal, rather than precious and to-be-minimized, is framed as a hallmark of genuinely AI-native work, enabled by the compressed prototyping cycle in Context Engineering and Specification-Driven Development.