Solutions
Explore workflow patterns before choosing technology.
These are proposed solution patterns to assess against your workflow and risk. Portfolio notes describe product directions and their current evidence limits.
- Solution patterns
- 6
- Levels of AI autonomy
- 3
- Evidence status
- Proposed
- Delivery responsibilities
- Scoped
Solution patterns
Levels of AI autonomy
Evidence status
Delivery responsibilities
The autonomy ladder
Pick the lowest rung that solves the problem.
Higher autonomy can introduce additional failure modes. Assess each workflow and only consider more autonomy when the acceptance test and the evidence trail can carry it.
AI-assisted
The human does the job; the model removes the blank page and the lookup. Usefulness, risk and review cost still require evaluation.
AI-enabled
The system drafts, routes, classifies or scores, and a person approves. Authority is bounded and agreed decisions are attributable.
Agent-run
The agent executes a defined job autonomously within explicit limits, with monitoring, evidence and a rollback path. Reserved for work where the acceptance test is objective.
Patterns
Six proposed patterns to assess.
Agent-run operations
A defined job an agent is allowed to run end to end — with explicit authority limits, an evidence trail and a human review point. Pennifi and Hyrolo provide product-direction context, not proof of operational delivery.
AI document and evidence work
Drafting, parsing, extraction and matching applied to the step that carries the cost — policy sets, invoices, receipts, assessments — instead of a chatbot bolted to a sidebar.
Compliance platforms
Control models, assessment flows, evidence lockers and remediation tracking, designed against a real framework such as NIST SP 800-171 rather than a spreadsheet with statuses.
Multi-party coordination
Provider, partner or supplier graphs where verification state, routing and handoff across organisations are first-class data — a proposed pattern, not a commissioned client example.
Data foundations
The domain model, access control and integration layer everything else stands on. Row-level security in the database, not permission checks scattered through the UI.
Operator systems
CRM, delivery and back-office systems modelled on how the business actually runs, so the shadow spreadsheets disappear instead of multiplying.
Governance
Agent authority is an architecture problem, not a prompt problem.
For a proposed agent design, agree: what it may act on, what it must escalate, what it records, how it is evaluated, and how a bad run is detected and undone. If those five answers don't exist, the workflow stays on a lower rung.
Fit check
Bring one workflow and we'll place it on the ladder.
Discuss whether deterministic automation, assistance or bounded agents may fit. Assessment and any written recommendations require agreed scope.