project

The Studio

A spatial thinking canvas where arranging goals visually triggers autonomous code execution — the first product UI for an AI factory system.

%% OUTLINE — Not finished. This is a skeleton for a writing session. %%

Core Insight (one sentence)

The interesting part of building a product UI for autonomous AI factories isn’t the dashboard — it’s discovering that spatial manipulation is a better interface for AI coordination than text.

Why It Matters / Who Cares

Anyone building with autonomous coding agents faces the same problem: how do you steer a system that ships 170 commits in a day? Chat interfaces collapse under that volume. Dashboards show you what happened but don’t help you think about what should happen next. The Studio is an attempt at a third option: a spatial canvas where your arrangement of goals IS the instruction set.

Skeleton

  1. The problem with steering autonomous agents — When your factory ships hundreds of commits in a day, the bottleneck isn’t execution. It’s deciding what to execute. Chat/CLI breaks down. Dashboards are read-only. You need something that helps you think, not just observe.

  2. Spatial manipulation as thinking — Kirsh & Maglio (1994) on epistemic actions: we move things around not just to achieve goals but to understand them. The Studio takes this literally — dragging a goal near another creates a relationship. Grouping goals into Arcs creates strategic threads. The canvas has labeled semantic zones (unlike Miro/FigJam), making AI interpretation reliable.

  3. What we actually built this week — Shaping sessions (bounded thinking episodes → AI proposes batch GitHub mutations), Portfolio/Arcs with multi-repo aggregation, ConversationBar hub, AI Experience Layer with proactive behavioral moments, Voice UX. The rename from “Shaping” to “The Studio” reflects the shift from a tool to a workspace.

  4. The factory feedback loop — Goals enter the factory via “Mark Ready” labels. Factory execution produces artifacts. Those artifacts feed back into The Studio as context for the next shaping session. The spatial canvas becomes the control surface for a system that runs continuously.

  5. What surprised us — [Fill in: what worked that wasn’t expected? What didn’t work? Where did the spatial metaphor break down?]

Evidence / Examples Available

  • Hundreds of commits across Studio and factory repos in multi-day sprints
  • Arcs feature: multi-repo portfolio intelligence across several autonomous factories
  • Proactive thought system: behavioral moments + cross-context synthesis
  • Shaping sprint: StorageAdapter abstraction, triage with AI whispers + enrichment popovers, Board View with arc lanes + dependency overlays, digest overhaul (comprehension-focused synthesis with temporal deltas), terrain visual upgrade, insights synthesis module
  • Maturity milestone: Auth + Postgres = crossed from internal prototype to deployable multi-user product. The narrative arc for the writeup: “tool built to manage AI agents → became a product itself”

Successor Chapters

The Studio’s evolution past the original spatial-canvas thesis got too distinct to live in this writeup:

  • writeups/plan-board.md — Apr 20–22: the shaping surface became the dashboard the IC runs on. Capital gates encoded into the meeting protocol.
  • writeups/audit-skill.md — Apr 27–28: the protocol that enforces the Tier 2 capital gate. “Rules, not rulers” at the experiment layer.