A post-sprint workflow for teams that learn as they go. Five principles for the age of human-agent collaboration.
Sprints assume teams share a timezone. Kanban assumes humans watch a board. Tickets assume work has a beginning and an end. None of this holds when AI agents work alongside humans and understanding evolves continuously.
Intent Flow replaces five foundational concepts of sprint and kanban-based work:
These five changes are connected by a single mechanism: the learning loop.
Intent Flow is built on a single observation: real work is a cycle of intention, action, discovery, and refinement.
You start with an intent, a statement of what should happen and why. Work begins. During work, you discover things you didn't expect. These are findings. Findings refine the intent, with evidence linked. The cycle repeats, each iteration deepening understanding, until the intent is clear enough to act on with confidence.
This isn't a software concept. It's how all purposeful work under uncertainty operates. To see it clearly, consider an example everyone understands:
Each principle addresses a specific failure mode of sprint and kanban-based workflows.
The primary object is the intent: a statement of what should happen and why, at a specific level of abstraction.
A ticket is a flat record. The "why" is buried in a description someone wrote three weeks ago. The technical approach is in a comment. The deployment plan is in someone's head.
Intent Flow separates these into a hierarchy. Each level evolves independently, but changes propagate awareness (not automatic updates) to adjacent levels.
An intent is not an edit. It's an evolution. Every change is an append, not an overwrite. The previous statement is preserved. The reason for the change is captured. The evidence that informed it is linked. Six months from now, "why did we add the cache layer?" has a traceable answer.
Work is a stream: an append-only temporal log of what actually happened. Status is derived, not declared.
Kanban gives you columns: Backlog → In Progress → Review → Done. "In Progress" means a human moved a card there, not that work is actually happening.
In Intent Flow, humans and agents write to the same stream. The agent that implemented the rate limiter at 2am and the engineer who reviewed it at 9am are both authors in the same log.
Multiple actors attend to the same work on different schedules, with different capabilities. Work flows continuously.
"Assigned to Alice" fails when Alice is asleep, overloaded, or on vacation. Intent Flow replaces assignment with attention: scheduled relationships between actors and work.
When Alice goes offline at 6pm, the agent continues working. When it needs human judgment, it writes a finding and marks it for review. When Alice comes online at 9am, she sees what happened overnight and what decisions are pending. No handoff. No reassignment.
Investigation is just work. Its output is findings: structured, linkable, first-class objects that feed the learning loop.
A "spike" is an admission that your workflow tool can't handle work whose output is understanding rather than deliverables. The spike produces a finding, the finding informs a decision, the decision shapes implementation. But in traditional tools, these are three disconnected artifacts linked by hope and convention.
In Intent Flow, findings are structured output. They can propose intent changes with linked evidence. They flow across work boundaries. Knowledge moves through the same system as code. Nothing is trapped in a wiki page that no one will find.
Every mutation has an immutable timestamp. The system doesn't ask "what state is this in?" It asks "what happened, and when?"
"Show me this work as it existed on March 15th" returns the intent, the work stream, the files, the findings, all at that moment. Not a changelog you reconstruct mentally. The actual state of understanding at that point in time.
The learning loop isn't a software concept. The same cycle of intent, work, finding, and refinement plays out wherever the path isn't fully known in advance. Which is nearly always.
Intent Flow is a methodology, not a product. Start with these five changes: