The Triton method

The Triton is the AI-era turn of the same idea. It extends Desirée Sy's dual-track agile from Autodesk (2007), which put design a sprint ahead of development. It keeps design research and design as two separate tracks, called tines, and adds development as the third. Drawn out, it looks like a trident.

The three tines

1. Development leads. It ships first and produces the artifact.
2. UI design iterates on what was built.
3. Research evaluates what was designed and built, and feeds findings back.

What's different

• The flow is reversed. Most triple-track models run research, then design, then development. Here development starts each cycle, because with AI the build is often the cheapest way to make something to react to.
• Research is evaluative. It works on a shipped product, not only before building.
• Design and research are separate tines.
• It fits startup reality. One person can fill several roles.

The 2012 DRF diagram already had three tracks, with a discovery track that validated the previous sprint's release. What changed in 2026 is which track goes first.

In practice

The Triton was the working method on PayPaw from March 2026. It changed shape three times: first three rails split between a person and two AI agents, then six generator-and-evaluator loops, and by June one person and one agent, with research folded into the build loop as an evaluator.

Open questions

• Is the Triton a replacement for the discovery-led loops, or one mode of the DRF? Generative research still matters when there's nothing to ship yet. PayPaw's history leans toward "one mode".
• Name: "Triton" collides with NVIDIA's inference server. Say "the Triton method".