Skip to main content

Lab Notebook · Own Product Line

GISGen — the gap between generating code and generating a working system

A comprehensive AI-powered development platform with agent orchestration and a CLI, built to push past 'generate code' toward 'generate working systems' — and finding that the real bottleneck is orchestration, not a better prompt.

JR
Jon RossFounder, vocabotics — 15 years building safety-critical systemsLab report · dated 15 September 2025
Verified by a human. Drafted with AI, verified by a human. Jon Ross, 15 Sept 2025
Living document. Reviewed 15 Sept 2025
Entry date
15 September 2025
Category
Tooling
Access
🔓 Public
ResearchPartly proven

Generating code was never the hard part. Sequencing, wiring, and verifying it together as a working system was — and that's the gap GISGen went after.

Node.js + Python (FastAPI)
orchestration stack
Agent orchestration + CLI
core capability shipped

Honest evaluation

Partly proven

Agent orchestration and a CLI genuinely worked; the larger claim — generating a complete working system, not just code — stayed mostly documentation at the point of this audit.

What would prove or disprove it further

What would move this to proven: ship the full-stack integration surface as working code rather than documentation, and demonstrate an end-to-end run where a natural-language request becomes a sequenced, wired, verified multi-piece system without manual intervention. What would move it toward disproven: a later attempt at the same integration surface, using the orchestration insight this project produced, that still failed to close the gap.

The evidence — full reasoning behind the verdict

Verdict: partly-proven. GISGen set out to solve a step beyond code generation: not just producing correct code, but producing a working system — pieces correctly sequenced, wired together, and verifiable. The report is candid that this larger goal was only partly reached. Agent orchestration and a CLI were real, working pieces at the time of the audit; the rest of the intended full-stack integration surface existed mostly as documentation describing what it would do, not code that did it.

The evidence for what did work is concrete (a running Node.js/Python orchestration stack and CLI); the evidence against full completion is the report's own admission that the documentation-to-code ratio was itself a signal the orchestration problem hadn't been fully worked out yet. That's a more useful kind of honesty than claiming a finished platform — the report treats the insight it produced (orchestration, not prompting, is the real bottleneck) as the actual deliverable, which is a fair description of what was actually proven.

Generating a function is easy for an AI. Generating a working system — one where every generated piece is correctly sequenced, wired to the others, and verifiable — is a different problem. GISGen went after that gap directly.

What it was

GISGen — a comprehensive AI-powered development platform combining agent orchestration, a CLI, and full-stack integration, aimed at the step beyond code generation: generating systems that actually run together.

What we built

Research

A platform built on Node.js and Python (FastAPI) on the backend, React and TypeScript on the frontend, containerised with Docker, with Alembic handling database migrations. Agent orchestration and a CLI were the core working pieces at the point of this audit, with extensive documentation alongside them describing the intended full-stack integration surface.

What we learned — including the honest negative

The real lesson was conceptual, and it reframed how later platform work in the portfolio got approached: the gap between "generates code" and "generates a working system" is an orchestration problem — how the generated pieces get sequenced, wired to each other, and verified together — not a prompting problem. A better prompt makes a better individual piece; it doesn't make the pieces cohere.

The honest edge: in-development, not a finished product. The value of this entry is the insight it produced — orchestration, not generation, is the real bottleneck — more than the platform itself at this stage.

Where it went / status

In development at the time of the audit, with the orchestration-over- generation framing carried forward into later platform work. It sits alongside a similarly enterprise-shaped orchestration platform built the same window, testing the same idea from a different architectural angle.

What is still open — kept visible

The honest edges, next to the wins. This is what turns 🔬 into 🟢 — honestly.

  • In-development, not a finished product — the documentation-to-code ratio at this stage was itself a signal the orchestration problem was still being worked out.

Sources

  1. vocabotics project audit — GISGen (AI dev platform, agent orchestration), Sep 2025vocabotics internal project history · as of September 2025

    We use cookies.