Skip to content

Implementation Sprint

The implementation sprint consumes the impact-analyzed specifications produced by the Spec Sprint and runs them through the per-change SDLC flow. Each change moves left-to-right through the agent fleet under structural validation. The Code Ontology stays current as a side effect of every merge.

The Per-Change SDLC Flow

A specification enters the flow from the left, one of many flowing through in parallel.

StageWhat runsWhat it produces
1. Spec pulled from backlogThe product owner (via the implementation sprint backlog in Jira or equivalent)An impact-analyzed spec, validated in the prior spec sprint, ready for the implementation team
2. Impact Analysis AgentThe Impact Analysis Agent traverses the four-layer graph for the specA pre-implementation impact report: which functional outcomes are touched, which design components, which architecture services, which code modules and database tables
3. Coding AgentA coding agent generates code under the impact report’s structural planA draft implementation, scoped to the impact report’s surface
4. Developer review (Zone 3) or autonomous execution (Zone 4)At Zone 3, the developer reviews and refines. At Zone 4, the agent runs the loop and the developer reviews the agent’s audit trail.Code ready for PR
5. PR Validation AgentThe PR Validation Agent checks the merge against all four ontologiesPass or fail. A failing merge is blocked until the underlying structural violation is fixed.
6. Code merge + KG SyncThe merge lands. The KG Sync Agent updates the Code Ontology.Updated graph. The next spec sprint runs against the current graph.

The implementation sprint does not run impact analysis on the spec it receives, because it has already been done in the spec sprint. When the implementation sprint’s Impact Analysis Agent (running for verification) detects a spec missing context that only the custodians can supply, the spec is pushed back to the spec sprint backlog. The implementation team does not block on it. They pull the next spec from the backlog.

The Three Sources of Truth in Operation

The flow uses all three sources of truth (covered in Three Sources of Truth on the Process landing) together.

StepSource consultedQuestion answered
Spec pulledSpecificationWhat is the change supposed to do?
Spec correlated to ticketTicket systemWhich workstream is this part of? Which team owns it? When is it scheduled?
Impact analysisKnowledge graphWhat will this change touch across all four layers?
Code generated and validatedAll threeImplementation against the spec, validated against the graph, tracked in the ticket
Post-merge syncKnowledge graphThe graph now reflects the change. Future impact analyses operate against the new state.

After the change is merged, the same Impact Analysis Agent can rerun the analysis against the new state of the knowledge graph and compare the post-implementation reality against the pre-implementation specification. The three sources of truth, used together, are what make that comparison meaningful.

Zone 4: Agents Run the Loop

At Zone 3, the developer is in the per-change loop, reviewing the coding agent’s output and applying implementation judgment. At Zone 4, the agents run the loop and the developer moves upstream.

The diagram shows the operating picture at Zone 4. Three custodial layers stack vertically. At the top, the four ontology custodians (PO, Architect, Designer, Engineering Team) keep the four-layer knowledge graph current. In the middle, the per-change SDLC flow runs through a six-stage agent fleet: Specification (PO extends the Functional Ontology), Impact Analysis Agent, Coding Agent (autonomous), Test Generation Agent, PR Validation Agent, and Code merge with KG Sync. At the bottom, a new custodial layer appears: the Engineering Team as custodian of the agent fleet, stewarding the four agents from Impact Analysis through PR Validation.

The Engineering Team’s day-to-day in this mode includes approving Promotion Agreements that move agents to higher autonomy levels, reviewing the agent audit trail on a defined cadence, refining agent prompts when failure patterns emerge, catching the edge cases where the impact report missed something, and calibrating the policies that govern agent behavior. Progressive Autonomy covers the agent-side discipline. Engineering Team’s Custodianship of the Agent Fleet covers the human-side discipline that pairs with it.

What Changes for the Implementation Team

The shift from Zone 3 to Zone 4 is gradual. The agent fleet does not become autonomous overnight. Each agent earns higher autonomy through demonstrated evidence, governed by Progressive Autonomy and Promotion Agreements.

DimensionZone 3Zone 4
Developer’s per-change roleReviews the coding agent’s output, applies implementation judgment, refinesReviews the agent’s audit trail, catches edge cases the impact report missed
Developer’s upstream roleLight: occasional agent prompt refinementSubstantial: custodian of the agent fleet, approves Promotion Agreements, reviews audit trail on cadence
What runs the per-change loopDeveloper + coding agent togetherThe agent fleet, with developer custodial oversight
What blocks at mergePR Validation Agent against the graphSame; the gate does not change
What the engineering team optimizes forImplementation judgment, code quality, edge case handlingAgent reliability, prompt quality, autonomy threshold calibration

The progression is not a single transition. Each agent in the fleet earns higher autonomy on its own evidence. A team can be at Zone 4 on Impact Analysis (the agent runs autonomously, dev reviews audit trail quarterly) while still at Zone 3 on the Coding Agent (dev still in the per-change loop). The team’s overall zone is the floor across the fleet.

What the Implementation Sprint Does Not Do

The implementation sprint does not author specifications. That is the spec sprint’s work. Trying to author specs inside the implementation sprint is the Zone 2 ceiling that the spec sprint discipline is designed to remove.

The implementation sprint does not make architecture decisions. Architecture decisions belong in the Architecture Ontology, maintained by the architect during the spec sprint. The implementation sprint executes against the validated specs.

The implementation sprint does not govern the knowledge graph. Graph governance is the enablement team’s responsibility. The implementation sprint surfaces findings through PR validation failures and the merge events that trigger KG sync; the enablement team acts on the patterns.

How Accion Labs operationalizes the implementation sprint

The Breeze.AI platform runs the Impact Analysis, Coding, Test Generation, PR Validation, and KG Sync agents that operate the per-change SDLC flow. The engagement model staffs the implementation team and the Zone 4 custodial role evolution.


Spec Sprint is the cadence that feeds the implementation sprint. Team covers the operating model that staffs both sprints. The Agents describes each agent in the per-change SDLC flow in detail.