AI-SDLC Target Workflow v4.0

Revised from cross-functional All Hands review · Iterative — not sequential · Compliant with Team Structure Principles v1.0 · Mitratech Legal Solutions · 2026

Agent-driven
Human owns outcome
Quality gate
Team ritual
Click any box to see discipline responsibilities
◈
Shared Context Layer — referenced by every phase ↗
A persistent layer above all phases. Every agent draws from and writes back to this layer. Spans source code, knowledge base, ADRs, Jira, Confluence, ProductBoard, customer signals, and org docs. Click to explore.
Knowledge base + ADRs Codebase graph ProductBoard Jira + Confluence Customer signals Agent-maintained context doc
↻ This is iterative, not sequential. Phases cycle and overlap — output from any phase can feed back into an earlier one. Teams will adopt gradually; ceremonies familiar to your team provide stability during transition. Story points required for capitalization.
Phase 1 — Discovery & Spec
Discovery
PM sets intent + testable acceptance criteria + edge cases. ProductBoard one-pager updated.
PM × Design Collab
PM spec + design prototype developed together. No tool-specific constraints — approach adapts per product team. Shared context — no manual handoffs.
Agent: Functional Spec Draft
Agent drafts functional spec, acceptance criteria, and edge cases drawing from the shared context layer.
PM Reviews & Owns Functional Spec
PM owns functional spec. Engineering owns technical spec separately. Dual pathway begins.
Phase 2 — Technical Spec
Inputs
Functional spec · Design artifacts · Context layer · ADRs
Agent: Tech Spec Draft + TDD
Orchestrated AI generates technical spec. Test cases defined upfront (TDD) before implementation begins.
Dev Reviews & Owns Tech Spec
Dev owns technical spec, trade-off decisions, and test strategy review. ADRs updated.
Spec Gate
Functional spec + Technical spec + Design artifacts all present before implementation begins.
Agent: Story Draft + Pointing
Agent generates Epics + Stories draft and suggests story point estimates based on prior sprint patterns and velocity. Human review required before Jira creation.
Human: Review Story Draft
Team reviews agent-generated stories for correct sizing, technical detail, and acceptance criteria before Jira creation.
Story Gate → Jira + ProductBoard
Approved stories with points created in Jira. ProductBoard synced. Points set for capitalization tracking.
Phase 3 — Development
Dev: Intent & Dispatch
Dev lead sets task intent. Approves agent dispatch. TDD test cases confirmed upfront.
Agent: Orchestrated Loop
Orchestrator routes to scoped coders. Govern → Validate → Execute cycle. Context layer manages scope and prevents redundant token usage.
CI Gate
Automated enforcement of security, architecture, and coding standards. Coverage threshold enforced.
Dev Review & Owns
Dev reviews all agent output. Named on commit — accountable. No unreviewed output in production.
DoD Gate
Coverage + human review complete. Auto deploy on pass: continuous delivery, no manual steps. Merged to main → production.
Phase 4 — Release
UAT
User acceptance testing. Feature flags for large releases. On-prem products require additional gate.
Company Readiness
GTM, Support, CS, and Enablement confirm readiness. Training materials, comms, and documentation prepared before release.
Release Sign-off
PM + Dev confirm release readiness. Release notes generated. Stakeholders notified. On-prem vs. cloud paths diverge here.
Adoption
Monitor feature adoption post-release. Agent tracks usage signals. Feedback loops back to Discovery for next cycle.
Phase 5 — Maintenance & Operations
Agent: Bug Pickup
Agent monitors incoming bug reports and security signals. Automatically picks up and triages new issues against the context layer.
Agent: Root Cause + Fix
Agent identifies root cause using context layer. Devises a solution and generates a PR with a fix and test coverage.
Human: Review & Approve PR
Dev reviews the agent-generated PR. Confirms fix is correct and tests are adequate. Signs off — name on commit.
DoD Gate + Deploy
CI gate enforced. On pass: automated deployment. Release notes generated from PR and context. Audit trail complete.
Agent: Monitor + Learn
Agent monitors post-fix production signals. Confirms issue resolved. Updates context layer with learnings. Feeds patterns back to CI gate.
Continuous Loop — Ceremonies & Learning
Ceremonies
Teams continue Agile ceremonies: Refinement, Planning, Daily Stand-up, Sprint Review, and Retrospective.
Standup Pre-work
Agent summary + blockers ready before every standup. Program Manager surfaces cross-team blockers.
Retro + Learning
Human judgment on blameless learning. Agent drafts retro notes. Cross-team patterns shared. Pilot teams lead.
10–15% Automation
Fixed sprint allocation — protected capacity. Non-negotiable. QA automation transformation tracked here.
DORA + AI Adoption Metrics
Quarterly maturity self-assessment. Program dashboard. Engineering + AI diagonal check.

Every box is clickable — see responsibilities for each discipline: PM · Designer · Development · Program Manager · AI

Mitratech Legal Solutions · AI-SDLC Workflow v4.0 · 2026