--- id: roadmap title: Roadmap description: Assess current code quality and delivery processes, close critical gaps, then pilot, expand and sustain AI adoption. hide_table_of_contents: true --- import ArchifyDiagram from '@site/src/components/ArchifyDiagram';
Understand today’s code and how work really reaches production before introducing AI into the workflow. Fix what would make a pilot unsafe or impossible to evaluate—not every item of technical debt. **Recommended sequence · Target operating model, not an approved schedule or deployed governance.** ## 1. Understand code quality and delivery Inspect representative repositories across important services and change types: tests and CI reliability, maintainability, technical debt and security findings. Trace recent changes through the actual SDLC with the people doing the work—from request to production and feedback. Check ownership, review and release practices, handoffs, rework and bottlenecks; do not assume the documented process is the real one. **Output:** baseline findings, a process map and prioritized gaps. Use these to identify a low-risk, repeatable workflow with observable acceptance—not simply the easiest repository. ## 2. Fix critical readiness gaps Resolve gaps that threaten safe execution or a trustworthy comparison: for example, exposed secrets, unclear approval authority or unreliable checks needed to accept the pilot’s changes. Establish accountable people, data boundaries and recovery for the selected scope. Broader refactoring and non-critical debt can run alongside the pilot with a separate improvement backlog. **Output:** a readiness decision and bounded pilot charter. Proceed only when the critical gaps for that workflow are closed; otherwise narrow the scope or hold. ## 3. Run a bounded pilot Try a small set of reviewed skills in the selected workflow, with human review and no new privileged autonomous actions. Compare equivalent work with the baseline: accepted outcomes, escaped defects, full cost and verification effort. [Skills Measurement](./measurement.mdx) explains what to collect, calculate and report; agree the comparison rules before starting. **Output:** a pilot report and a continue, adjust or stop decision. Missing evidence is a reason to investigate, not proof of improvement. ## 4. Expand only on evidence Extend to another workflow or team only when quality, usefulness, cost and support capacity justify it. Prove reliable skill distribution, revocation and recovery before expanding authority. Introduce autonomous SDLC execution progressively under human direction; new departmental workflows need their own data boundaries and acceptance evidence. **Output:** an evidenced expansion decision with verified controls for the next scope—not organization-wide rollout by default. ## 5. Sustain and improve Review outcomes, drift, incidents and maintenance effort. Retire ineffective skills, revisit process bottlenecks and repeat assessment when repositories, clients, data or authority change. Work toward [Target](./target-operating-model.mdx): an autonomous agent executes and governs the repeating SDLC while humans direct, guide and review. **Output:** maintained skills, a refreshed improvement backlog and evidence for the next cycle. Keep human takeover available; narrow or stop workflows when evidence no longer supports them.
Supporting gate evidence and hold conditions These are recommended implementation gates, not assigned human owners, approved thresholds or installed integrations. Agree acceptance criteria for each selected workflow before evaluating it. ### Assessment → readiness Retain the sampled repository revisions, test/CI results, debt and security findings, and traced changes behind the process map. Record coverage and unknowns. Distinguish critical pilot blockers from parallel improvements; hold selection if the sample cannot explain how the candidate workflow is delivered and accepted. ### Readiness → pilot Inventory actual client hosts/builds, provider/data terms, authority and event availability. Confirm accountable people and independent reviewers, forbidden actions, data/retention decisions, recovery and a reproducible baseline. Hold if accountability, lawful collection or critical acceptance checks remain unresolved. ### Pilot → expansion Use a small registry, immutable skill contracts, qualified client adapters, evaluations and coaching. Verify the measurement calculations with synthetic fixtures before interpreting real outcomes. Require mandatory safety checks, independent task acceptance, reconciled source coverage and user feedback. Hold or narrow scope if quality or coverage is inadequate; usage alone does not justify expansion. ### Expansion → wider authority Prove canary updates, effective-version readback, version locking, exception expiry, stale/offline handling, rollback and revocation. Hold if effective versions or controls cannot be verified or the catalog cannot be supported. For autonomous execution, begin with observation and evidence assembly, then proposing and routing work, before granting scoped execution authority. Recommended controls include exact-artifact approvals, idempotency, duplicate/outage recovery, separation of duties, audit and kill-switch drills. Prove human takeover; never infer permission from missing evidence. The agent ultimately executes and governs the cycle, not only routes tasks. ### Sustained operation Operations, Marketing, Finance & Accounting and HR & Administration workflows each need a workflow owner, outcome, data boundary, capacity and domain-specific review; Product and Technology continue quality improvement. Keep sensitive decisions human. Use accepted outcomes, guardrails, total cost and experience to continue, narrow, stop or roll back. Changed client, provider, data or authority boundaries reopen the relevant gates.
**Next:** [Return to the strategy destination →](./target-operating-model.mdx)