Skip to main content

AgentLabUI / Gofannon

Classification Open-source agent tooling / workflow-to-application system

Status Historical / open-source

Materials AgentLabUI, Gofannon, LLM workflows, agent tooling, process descriptions, decision points, generated user interfaces, AI Alliance work

Problem​

AgentLabUI started as a practical tool for building agent-backed workflows. It later became Gofannon.

The core idea was simple: describe a process, mark the points where an LLM should help decide or generate, then turn that workflow into something a person could actually use.

That last part mattered.

Build​

Agent demos often stop at the interesting middle: a model call, a tool call, a chain of steps, a generated answer. Real users need more than that. They need an interface. They need a way to provide inputs, review outputs, understand what happened, and use the system without caring about the agent framework underneath.

Gofannon grew out of that gap. It was about making agent workflows usable, not just technically possible.

  1. Startprocess description
  2. Structureworkflow steps
  3. DecisionLLM decision points
  4. Contexttools / resources
  5. Surfacegenerated interface
  6. Useuser review
interface is part of the systemworkflow firstnot just a tool call

What it did​

  • Took described processes and turned them into agent-backed workflows.
  • Supported LLM decision points inside those workflows.
  • Helped generate lightweight user interfaces around the process.
  • Let non-agent-developers interact with the resulting system.
  • Connected the build layer to the user-facing layer.

What worked​

  • It made agent workflows more concrete.
  • It turned abstract process descriptions into usable application surfaces.
  • It helped separate practitioner tooling from end-user interaction.
  • It pointed toward a useful truth: the interface is part of the agent system.

What was hard​

  • Agent workflows are easy to demo and harder to stabilize.
  • Generated interfaces still need product judgment.
  • Process descriptions need structure before they can become reliable systems.
  • The useful part is often not the model call. It is the boundary around the model call.

Lesson​

The workflow is the product.

An agent is only useful when the surrounding system makes it usable: inputs, outputs, interface, review, constraints, and ownership.

Why it matters now​

This artifact is one of the clearest predecessors to aboriginal armadillo's current work.

The same pattern shows up in bespoke client agents:

  • start with a real workflow
  • identify where AI can help
  • define what humans still own
  • build the agent around available tools and data
  • give people an interface they can actually use
  • keep the system bounded enough to trust

Gofannon was not just about agents. It was about making agent workflows usable by people who did not want to think about agent infrastructure.

That is still the work.

Evidence​