Scrum was a social technology. It lived in a document addressed to human beings; it was read before the work rather than during it, and it depended on people to carry it. That was the right design for its constraint: when writing code was the expensive step, the useful thing a methodology could do was organize human attention around it. The agentic development frameworks now emerging alongside it are a different kind of object. They are artifacts read at runtime by the thing doing the work. Methodology has become source code, and that change explains how quickly these frameworks have spread.
What ships today is a file in a repository. GitHub's Spec Kit organizes the cycle into a constitution, specification, plan, tasks, and implementation. BMAD ships a roster of scoped personas, each producing a versioned artifact and handing off to the next. AWS has proposed AI-DLC, which replaces the two-week sprint with a bolt measured in hours or days, the epic with a unit of work, and backlog grooming with a ritual it calls Mob Elaboration. These are not descriptions of how a team ought to behave. They are context loaded into the model's window, sitting on the critical path of every action an agent takes. A rule forbidding an implementation agent from altering a schema defined upstream is not an aspiration a developer might honor. It is a constraint on the next token.
Everything downstream follows. A methodology that executes has to be installed, which is why these arrive through package managers and live in repositories, and why an available measure of their traction is a GitHub star count. Spec Kit sits above 130,000 stars, within striking distance of Claude Code itself; OpenSpec, GSD, and BMAD each sit between roughly 50,000 and 67,000. Those numbers deserve discounting, since stars never decay when a tool is abandoned, but with no certification body and no survey instrument, a vanity metric is carrying weight it was never built to bear.
The more consequential inheritance is version control. Process can now be forked, diffed, pinned to a version, and rolled back when a change makes the agents worse. Methodology iterates at software speed, which is a mixed blessing: Thoughtworks placed spec-driven development in the Assess ring of its most recent Technology Radar rather than Adopt, noting that the workflows are elaborate and opinionated, that some generate specification files too long to review, and that the field may be relearning the bitter lesson that handcrafting detailed rules for a model does not scale.
The role names mislead. BMAD's analyst, architect, scrum master, and developer borrow Scrum's vocabulary while doing unrelated work. Scrum roles assigned accountability to people. These personas are context-scoping and permissioning devices, each a bounded window with a defined deliverable and restricted authority. The thing reads like an org chart and functions as access control.
Which brings the commercial question into focus. Spec Kit, OpenSpec, BMAD, and GSD are free, and mostly MIT-licensed. A methodology that could be taught and certified has been joined by one that can simply be cloned, and the value has migrated underneath it, into the harness, the evaluation suite, and the platform that decides how much change reaches production without a human reading it (https://alten.capital/blog/the-agentic-margin). The frameworks are not solving the binding constraint. Review is. Telemetry across millions of pull requests shows machine-generated changes waiting several times longer for first review, arriving substantially larger, and being accepted at roughly a third the rate of human-written ones. Review capacity is the one input AI did not multiply, and the frameworks front-load specification and alignment precisely because that is the only lever available (https://alten.capital/blog/the-ai-agent-orchestrator).
For services businesses the implication arrives as an operational fact rather than a forecast. Each of these methods reduces the batch size. A unit of work that used to occupy a quarter now occupies a bolt, and an engagement measured in hours cannot be priced the way one measured in months was (https://alten.capital/blog/from-time-to-outcomes). The methodology is free, the batch is smaller, and the scarce input is the judgment applied at the review gate. Scrum's insight was that the process should be built around the constraint. That insight holds. The constraint moved.
Alten Capital invests in exceptional management teams to accelerate high-growth technology services businesses.