Durable context
Product intent, architecture, conventions, current state, operational knowledge, decisions and evidence persist beyond any individual worker. Context follows scope and retains provenance.
Foundation
A system designed for interchangeable workers
Humans and agents use the same working substrate and meet the same project-defined standard. The system preserves continuity when either enters or leaves the workflow.
Interchangeable workers
A developer, local agent or remote agent can enter, delegate, review, leave or resume. Handoffs carry state and artifacts rather than relying on the original conversation.
One quality bar
Human and agent contributions meet the same repository contract. Quality depends on the change and its risk, not on the identity of the contributor.
Variable control
Users choose their level of abstraction and involvement. The system manages coordination complexity while exposing decisions, risk, evidence and required action.
Stable primitives rather than one mandatory pipeline
Intent
Plan
Execute
Verify
Integrate
Learn
A small change may collapse stages into one interaction. A cross-system migration may use nested intents, parallel execution, staged verification and production observation. Reusable workflow patterns emerge from composing the primitives to fit the work.
Connected reality
The project does not live in isolation
Context and control form a hierarchy across organisation, domain, system, repository, intent and task. The authoritative source can vary; provenance cannot.
Federated context
The SDLC connects the systems needed to understand and change software while preserving identity, version, authority and derivation.
- Code and repositories
- Jira, Linear, Confluence and company knowledge
- Related services and deployment tooling
- Grafana, OpenTelemetry and incident systems
Production feedback
Runtime evidence connects desired outcomes with real behaviour. Telemetry can diagnose failures, verify changes, measure persistent goals and trigger governed remediation.
Goal or SLO→Intent→Change→Observe→Remediate
Control and governance
Autonomy is explicit and evidence-informed
Known rules become deterministic gates. Judgment and autonomy scale with maturity and risk under a governance function independent of the delivery workflow.
Deterministic gates first
Permissions, tests, policy checks, required evidence, deployment controls and SLO thresholds enforce known rules consistently.
Verification fits risk
Acceptance may use deterministic checks, independent agents, humans or several reviewers according to the maturity and consequence of the work.
Multiple identity models
Agents may inherit user access, receive task-scoped delegation or act through persistent roles. Every action remains attributable and policy-governed.
Governed autonomy
A higher-level human or agent function promotes or reduces autonomy using evidence. A delivery workflow cannot expand its own authority.
Shared tools
Developers and agents use inspectable primitives such as Markdown, CLIs, reproducible environments, machine-readable contracts and observable state.
Rich artifacts
Chat supports direction and clarification. People interact through the form best suited to the work: code, text, diagrams, dashboards, audio, video or simulations.
Open composition
The system provides opinionated contracts while allowing higher-order workflows to be assembled from understandable, observable parts.