The Agentic Agile Manifesto

The values held. The constraint moved.

The Agile Manifesto (2001) replaced heavyweight waterfall with something that worked, and it has proved its value for twenty-five years. Its values were calibrated to the constraint of the day, human coding speed: working software over documentation, responding to change over following a plan. On that constraint, they held, and they hold still.

AI moves that constraint. Mid-to-senior-level code now arrives in seconds, not days. AI is text-centric, stateless, runs without ego or fatigue. With the coding bottleneck gone, the values still hold, but the ceremony layer built on top of them, sprint planning, daily standups, retrospectives, backlog refinement, was tuned to human-paced coding. That layer is what needs clarifying and extending, not the values beneath it.

The efficiency problem is no longer "how fast can engineers write code." It is: how precisely can teams specify what to build, and how fast can they validate what AI delivers.

The four values

We are uncovering better ways of building software by doing it with AI systems as members of the engineering team. Through this practice, we have come to value:

The team now includes the Silicon Software Engineer, the AI agent. What moves the work is the people and the agents, and how they interact, not the process framework or the tool of the month.

The source of truth is the primary knowledge of what the product must do, captured in text so an agent can read it: the customer's requirements, the architecture decisions, the transcripts of internal and client calls. It is where the truth lives, and a specification can be derived from it at any time. It matters for the silicon engineers who read it, and for the humans who use AI to work faster. Only the essential source, not exhaustive documentation.

The client collaborates by seeing and approving working software at every stage: concepts and prototypes at the start, the finished product at the end. Not a document negotiated into a contract. AI makes this cheap, a working prototype in a day instead of a slide deck, so intent is validated empirically and early.

Responding to change still wins over following a plan, but it has to be disciplined. A silicon engineer builds exactly what a change specifies; loose change management means it builds the wrong thing. Change is welcomed, and managed with a rigor that plan-driven teams never needed.

Agile's original four values still hold. These restate them for a team where an agent, not only a human, writes the code: the same left-over-right, with the silicon teammate written in.

Sharper, or in transition

Those four values come from re-reading Agile's twelve principles through one question: when an AI agent writes the code, which principles get sharper, and which are mid-transition? Almost none simply die. What dies is the ceremony layer built on top of them.

Sharper with AI

Eight principles need no revision; several matter more than before.

• Deliver value early and continuously; welcome change even late (1, 2). Agents make delivery faster and re-planning cheaper, so both get easier to honor.
• Ship working software frequently (3). Becomes the Stint, a two-to-three-day cycle.
• "Business people and developers work together daily" (4). Closer than ever, not further. The Pit Wall role puts the engineer directly against the business, turning intent into an executable spec. AI moves the developer toward the business, not away from it.
• Working software is the measure of progress (7). The only honest measure when an agent can produce plausible code that does the wrong thing.
• Technical excellence, and simplicity as maximizing the work not done (9, 10). An agent amplifies good and bad design alike, and makes the work you choose not to build almost free, so cutting scope is now the cheapest lever you have.
• "The best work emerges from self-organizing teams" (11). Sharper than ever. RACE Programming splits one large team into two small ones, a two-person Pit Wall and a three-person Pit Crew, and a handful of people self-organize far more readily than a big group, aiming together at the finished result. The agent does not self-organize; it executes inside the guardrails they set.

In transition

Four principles still hold, but are moving toward a new meaning.

• "Build projects around motivated individuals, and trust them" (5). Trust becomes something earned through verification rather than granted by default, and the individual you direct now includes the agent.
• "Sustainable pace" (8). The agent never tires; the limit moves to human attention. The person to protect from burnout is the one validating the work, not the one producing it.
• Reflect and adjust at regular intervals (12). Survives, and so does its ceremony: the Retrospective is the one Scrum ceremony RACE keeps as is. Inspect and adapt is the mechanism these adaptations came from in the first place. What is still settling is the interval, as reflection also runs continuously inside the cycle.
• "The most efficient way to convey information is face-to-face conversation" (6). Worth watching. As people lean on AI to analyze and decide, real-time face-to-face loses some of its edge, and you increasingly need time with your assistant before you can answer well. Where synchronous conversation settles is still unfolding; for the agents themselves, written, machine-readable text is already the primary medium.

None of the twelve principles retire. Scrum was one way to operationalize them for all-human teams; what Agentic Agile drops is the ceremony layer built on top: planning poker, standups as status, story-point estimation. The principles underneath stay.

From writer of code to orchestrator of value

The central shift Agentic Agile names: the human engineer is moving from a writer of code to an orchestrator of valuable tasks. This is not a demotion. It is a concentration of leverage. One engineer now contributes more per hour than the engineer who types code faster: the one who specifies what AI cannot guess, verifies what AI cannot judge, and validates what the client cannot express on their own.

AI fluency, the ability to delegate work precisely to AI systems, becomes the primary professional competency. Teams that develop it compound their advantage. Teams that treat AI as a faster code editor do not.

From principles to practice

Agentic Agile is the bridge between the AI-First Manifesto (values for AI-native engineering) and RACE Programming (the prescriptive framework): the value layer that turns the revisited principles into a way of working.

If your team uses Scrum, the From Scrum transition guide maps every Scrum artifact, role, and ceremony to its RACE Programming equivalent, with a 90-day transition plan and the metrics to validate it worked.