Dual-Track Agile

Dual-track agile, popularized by Marty Cagan and Jeff Patton building on Desiree Sy's early work at Alias, separates the activity of figuring out what to build from the activity of building it — without separating the people. The discovery track runs experiments, interviews, prototypes, and analysis to reduce risk on candidate ideas. The delivery track turns validated ideas into shippable, production-quality software. Both run continuously and in parallel, week over week.

The connecting mechanism is a single prioritized backlog. Discovery's job is to feed that backlog with items that have already cleared the risks that matter — value (will customers want it), usability, feasibility, and business viability. Delivery pulls from the top of that backlog. The team is the same group of people wearing different hats at different moments, not two staffed units.

The single biggest failure mode is treating the tracks as two teams: a "discovery team" of PMs, designers, and researchers that throws validated specs over a wall to a "delivery team" of engineers. This rebuilds the exact waterfall handoff dual-track was meant to dissolve. Engineers lose context on why something is being built, designers lose feedback on what is actually feasible, and the backlog fills with detailed solutions nobody on the delivery side helped shape.

Tracks are concurrent streams of work, not org units. Engineers participate in discovery — pairing on feasibility spikes, reviewing prototypes, flagging technical risk early. Designers and PMs stay engaged through delivery. The point of running them in parallel is that the same brains carry context both directions, so a discovery finding can reroute delivery the same week it lands.

Discovery starvation is the usual symptom: under delivery pressure, the discovery track quietly empties and the team reverts to building whatever the loudest stakeholder requested. The fix is to protect discovery as ongoing capacity, not a phase — a standing share of each cycle rather than a project that "finishes."

The opposite failure is discovery running far ahead of delivery, producing a stockpile of validated ideas that age out before they ship. A healthy cadence keeps the two tracks roughly in step: discovery stays one or two delivery cycles ahead, not ten. Watch for a second anti-pattern — "validation theater," where discovery produces decks and opinions instead of evidence. Discovery items should enter the backlog with the specific risk they retired named explicitly (e.g. "7 of 8 target users completed the task unaided"), or they have not actually been discovered.

Dual-track only works when discovery findings and delivery work share one source of truth — when the customer interview that validated an idea, the revenue cohort that requested it, and the sprint card that builds it are joined, not scattered across a research repo, a feedback inbox, and a separate tracker. When they live apart, context is lost in the handoff between tracks and the team relapses into building unvalidated work.

A product operating system built on a shared spine keeps these joined: an insight captured in discovery stays linked to the opportunity, the customers behind it, and the delivery task it eventually becomes. On AIOProductOS the Insights feed, the boards, and Account 360 read the same record, so a delivery card shows the evidence that justified it and a discovery finding is traceable to the work it triggered. One backlog, backed by one set of facts, which is the condition dual-track quietly assumes and most stacks quietly break.