{"slug": "use-prototypes-to-plan-production-software-with-ai", "title": "Use Prototypes to Plan Production Software with AI", "summary": "A developer presented a planning workflow for building software with generative AI, combining agent skills from Matt Pocock and Ryan Singer to generate a PRD and implementation plan. The developer found that production-grade software requires three to eight hours of planning, while prototype-grade software only needs fifteen minutes, and advocated for using AI to build quick prototypes to inform production plans.", "body_md": "Eight hours of planning, or fifteen minutes? Depends what you're building: Production-grade or prototype-grade.\n\nIn June, I gave a talk\n\nat [Agentic Builders Collective Singapore](https://www.agenticbuilders.sg/)\n\nabout a planning workflow for building software with generative AI.\n\nIn a nutshell, it combines the grill-with-docs agent skill by [Matt Pocock](https://www.linkedin.com/in/mapocock/)\n\nwith the shaping agent skills from [Ryan Singer](https://www.linkedin.com/in/feltpresence/),\n\nand runs them in a highly customised way.\n\nThe result is a PRD plus an implementation plan.\n\nFeed those into an AI harness (OpenCode, Pi, Codex, Claude Code, …) and you get production-grade software.\n\nThe people who tried it came back with the same verdict:\n\nThe planning took too long.\n\nThree hours at best.\n\nOne person reported over eight.\n\nThe obvious response was to optimise.\n\nFewer questions.\n\nTighter interviews.\n\nSame rigour.\n\nI attempted to tune the planning skill, while still aiming for production-grade specs.\n\nIt didn't yield much improvement.\n\nAfter several iterations of failing to bring the time down, I finally understood why.\n\nThe hours spent planning **were not** a waste after all.\n\nIf you are genuinely planning production-grade software\n\n(the kind a team will maintain for months or years),\n\nthen an intense interrogation of the requirements **is** the correct amount of work.\n\nWhen humans build over months, mistakes in the plan get discovered and fixed along the way.\n\nWhen an AI harness builds the entire thing in mere minutes, those opportunities simply don't surface.\n\nThe plan is the *only* place that errors get caught.\n\nOf course it takes a while… it needs to!\n\nMy aha moment was the realisation that the problem was never about efficiency.\n\nThe problem was a mismatch of grades.\n\nHere is the distinction I was missing:\n\nProduction-grade software deserves **production-grade planning**.\n\nThree to eight hours of it, if that's what it takes.\n\nPrototype-grade software deserves **prototype-grade planning**.\n\nFifteen minutes of it is more than enough.\n\nMost of us don't actually need production-grade software right off the bat.\n\nWhat we need, at that point, is to **know what to build**.\n\nThat is a different problem, which needs a completely different tool.\n\nThe right tool is an old one: Prototyping.\n\nPrototyping fell out of fashion because it used to be expensive:\n\nEven throwaway software took days or weeks.\n\nIf you were going to spend that much, you may as well plan carefully and build it properly once.\n\nGenerative AI changed the economics. *Radically*.\n\nA prototype now takes under an hour to plan and build, for a single developer paired with an AI harness.\n\nWhen the experiment is that close to free, it makes *much less sense* to skip the prototype.\n\nIt is now a more valuable use of your time to build a prototype,\n\nand use that to make a *more informed* plan for the software that you intend to ship to users.\n\nIf you are familiar with agile methodologies, think of it this way:\n\nA prototype is a spike.\n\nExcept instead of spiking one risky feature, you spike the entire system.\n\nYou learn the true shape and specs of the thing you wanted to build,\n\nnot what you assumed it would be when you wrote the plan.\n\nOrder of operations:\n\n`REQS.md`\n\nfile (requirements).`prototype-plan`\n\nagent skill`PRD.md`\n\n(product requirements document) and `IMPL.md`\n\n(implementation plan).The key point here is that you *intentionally* spend very little time planning up front.\n\nThe idea is to build a working prototype ASAP.\n\nThis **is not** something you will ship to your users.\n\nThe prototype *reveals* to you what should go into the plan for the production-grade version of your software.\n\nUsing a prototype, you do this hands-on with a real, working application.\n\nWithout a prototype you would have had to imagine the intended application, which is a lot harder.\n\nThis entire process is now made easy and repeatable, using a few agent skills.\n\nFirst, use the `prototype-plan`\n\nagent skill\n\nto plan a prototype-grade version of your software.\n\n(Next, build a working prototype using `PRD.md`\n\nand `IMPL.md`\n\n, and learn from it.)\n\nFinally, use the [ build-plan-product agent skill](https://github.com/bguiz/build-agent-skills/tree/main/skills/build-1-plan-product),\n\n`build-plan-specs`\n\nagent skillAn honest caveat, before you take this too far:\n\nA prototype will **not** surface your scaling limits, your security model, your compliance obligations,\n\nor the migration path from the system it replaces.\n\nThose problems don't show up in prototypes, and no fifteen-minute interview will find them.\n\nThat is why the two grades are **distinct**.\n\nThe prototype tells you *what* to build.\n\nNow planning can start from evidence instead of imagination or guesswork.\n\nAs you've already cut yourself on them, you know where the sharp edges are.\n\nPrototype to learn.\n\nThen plan and build for production.\n\nIn that order.\n\nThis post was originally published at:\n\n[blog.bguiz.com/2026/prototypes-plans-production-software-ai](https://blog.bguiz.com/2026/prototypes-plans-production-software-ai/)", "url": "https://wpnews.pro/news/use-prototypes-to-plan-production-software-with-ai", "canonical_source": "https://dev.to/bguiz/use-prototypes-to-plan-production-software-with-ai-45b5", "published_at": "2026-08-27 02:03:15+00:00", "updated_at": "2026-08-27 02:17:55.481449+00:00", "lang": "en", "topics": ["generative-ai", "ai-tools", "developer-tools", "ai-agents"], "entities": ["Matt Pocock", "Ryan Singer", "Agentic Builders Collective Singapore", "OpenCode", "Pi", "Codex", "Claude Code"], "alternates": {"html": "https://wpnews.pro/news/use-prototypes-to-plan-production-software-with-ai", "markdown": "https://wpnews.pro/news/use-prototypes-to-plan-production-software-with-ai.md", "text": "https://wpnews.pro/news/use-prototypes-to-plan-production-software-with-ai.txt", "jsonld": "https://wpnews.pro/news/use-prototypes-to-plan-production-software-with-ai.jsonld"}}