One person prompts. One person reviews. Everyone else is a spectator. The creative partnership that produces excellent work — the back-and-forth, the challenge, the "yes, and" — gets flattened into a single person's conversation with an LLM.
Strategic intent gets lost in translation. Someone defines what matters. Someone else interprets that into direction. A third person translates both into prompts. Multiple layers of telephone game.
People who span strategy, creative, and technology are rare. They burn out. They don't scale. But everyone's building tools for them instead of for the team that actually does the work.
The best work happens in a room. Not a meeting room. A creative room where three perspectives sharpen each other.
The plan file is
the room. The code
is downstream.
Kettel doesn't replace the partnership. It scales it. The plan file is where the conversation lives. The agent is how all three perspectives get synthesized into executable intent. Craft doesn't scale through faster execution. It scales through better thinking.
Research, customer context, creative references, technical spikes. Atomic entries linked to each other and to plans.
Structured intent, user flows, creative direction, technical approach, implementation steps. Perspective-attributed. Version-controlled.
Architecture decisions, standards, conventions. Promoted from plans. Persists across features and sprints.
Owns intent, scope, user flows, and acceptance criteria. Reviews output against stated goals. Redirects the agent when output diverges from intent.
Owns experience direction, interaction patterns, and creative references. Reviews craft. Redirects the agent on experience and interaction quality.
Owns technical approach, architecture, and implementation constraints. Reviews code quality, security, and structure. Can intervene at the code level.
Team gathers knowledge in the Notebook. Upload anything — PDFs, URLs, Figma exports — auto-converted to structured markdown. The thinking before the plan.
Strategy writes Intent. Creative writes Direction. Technology writes Approach. Section locking. Inline annotations. Every perspective has a voice.
Agent reads the full plan and linked notebook entries. Generates implementation steps that respect constraints from all three perspectives. The "yes, and" at scale.
Self-contained plan handed to any coding agent. Claude Code, Cursor, Windsurf — the plan is the interface contract. No execution lock-in.
Category forming. Aviator coined "multiplayer AI coding." YC S26 RFS names the "deciding what to build" gap.
Spec-driven validated. GitHub Spec Kit, Amazon Kiro, Thoughtworks Radar converging on plans-before-code.
Incumbents locked in. GitHub defaults to issues. Cursor to IDE. Augment to engineer-only. None plan-native.
The Figma pattern. The shared artifact is the viral loop. Every person who gets a plan link becomes a user.
"Having 8 agents write code all day... it generates more code in a day than one person can review in a week."Hacker News — 2026
"An effective SDD tool would have to provide a very good spec review experience."Martin Fowler — 2026
"What is missing yet is how to share all that context. I don't think any one current solution will be the winner."HN — Team collaboration with coding agents
"The intent behind decisions gets lost. You're left with a codebase nobody fully understands anymore."Red Hat Developer — The Uncomfortable Truth
Plan file, notebook, section locking, annotations, agent discovery and plan modes, GitHub integration, ingestion pipeline.
SSO, audit trails, access controls, hosted version, compliance, advanced analytics.
The shared artifact is the viral loop. Every person who receives a link to a plan and can comment without installing anything becomes a user.
AI amplifies the partnership, not eliminates it. An agent that reads three aligned perspectives produces better work than an agent that reads one person's prompt. The more powerful the agent, the more the plan matters.
Spec-driven development is validated by GitHub, Amazon, and Thoughtworks — but nobody owns the team layer. Every implementation is single-author, engineer-only.
The upstream gap is recognized. YC Spring 2026 calls out "deciding what to build" as the next frontier after "building what's been decided."
Incumbents are locked into their paradigms. GitHub builds issue-tracker-with-AI. Cursor builds IDE-with-governance. Augment builds engineer-only orchestration. None are plan-native.
Craft is the durable advantage. Anyone can ship code fast. Teams that ship the right code — shaped by strategy, elevated by creative thinking, built with technical rigor — win.