April 15, 2026
Rubber ducks vs. robot coders
My AI-Assisted Workflow
AI work hack or homework hell? Internet splits over “think first, code later”
TLDR: A tech lead’s plan-first, AI-second method says the real magic is writing a clear plan before coding, then letting AI stress-test it. Commenters split: some applaud the structure, others call it influencer fluff, while jokers say the new workflow is just Googling via Gemini.
A tech lead just dropped a “think on paper, then code” guide to using AI: write a brain-dump, let an assistant grill you into a proper Product Requirements Document (a PRD), then slice the work into tiny, demo-ready tasks—some the bot can do solo, some needing a human. The goal: keep AI fast without turning your app into spaghetti. The internet? On fire.
The hottest take came from a grumpy realist who called it “influencer vibes,” insisting nothing’s changed since 2022 except the hype. Others cheered the structure, with one fan swearing by a “/rubberduck” chat ritual to force clarity before typing a single line. Meanwhile, the jokesters rolled in: one dev deadpanned that their modern workflow is just “Google → skim Gemini → avoid Stack Overflow.” Ouch.
Underneath the memes, a real split emerged. Process lovers say AI thrives when you spell out the plan; speed demons say this is just extra homework for a robot that was supposed to do the boring parts already. The quiet consensus? AI can crank out code, but it won’t fix fuzzy thinking. Whether that’s wisdom or busywork depends on how allergic you are to planning—and how much you trust your rubber duck.
Key Points
- •The author proposes an AI-assisted workflow that prioritizes written reasoning before any coding begins.
- •Step 1 is a private, free-form plan detailing the problem, solution ideas, constraints, and uncertainties.
- •Step 2 converts the plan into a structured PRD using the write-a-prd tool, which interrogates the author and codebase to surface assumptions and decisions.
- •The resulting PRD includes user stories, implementation decisions, module design, testing choices, and out-of-scope items, with user stories as the foundation.
- •Step 3 uses prd-to-issues to create vertical-slice issues classified as AFK (AI-executable) or HITL (human-in-the-loop), preferring AFK when possible.