AI Tooling Map

This page consolidates the AI usage scattered across the role pages into one map: which tool, which role, what for. Treat tool names as current examples - the patterns are what matter and outlast any specific product.

Tools by role

Role	Primary AI tools	Used for
Stakeholders (Product Owner)	Claude/ChatGPT, Productboard/Aha! AI, analytics copilots, Power BI Copilot, meeting assistants	Backlog drafting, feedback clustering, outcome analysis, report digest, demo prep
Project Manager	Jira AI / ADO Copilot, Claude, meeting assistants, predictive analytics	Status drafting, risk detection, forecasting, action tracking
Business Analyst	Claude/ChatGPT, Mermaid AI, Confluence/Notion AI	Story & AC drafting, edge-case discovery, diagrams, spec summaries
Architect / Dev Lead	Claude, Copilot/Claude Code, cloud AI advisors, Mermaid	Design options, ADRs, IaC scaffolding, threat modelling, diagrams
Developer	Copilot / Cursor / Claude Code, Claude/ChatGPT, Snyk in-IDE	Pair programming, refactoring, unit-test gen, code explanation
Cloud Engineer (DevOps)	Copilot/Claude Code, cloud AI advisors (AWS Q/Well-Architected, Azure Advisor, GCP Gemini Cloud Assist), Checkov/tfsec/Trivy, AIOps copilots (Datadog/Grafana/PagerDuty AI)	IaC authoring (Terraform/Bicep/CFN/Ansible/Helm), config translation, policy & security scanning, deploy automation, anomaly detection & incident summaries
Quality Engineer (QC/QA/Test)	Copilot review / CodeRabbit, SonarQube, Snyk, GH Advanced Security, Playwright+AI, Postman AI, Applitools	AI code review, SAST triage, dependency & standards checks, test-case gen, automation, defect triage, visual/a11y
Data Engineer (data projects)	Databricks Assistant, dbt + AI, Snowflake Copilot, BigQuery/Synapse/Redshift AI, Claude/Claude Code	Research & solution exploration, ELT/ETL & SQL generation, debugging complex pipeline errors, data-quality tests, documentation

The Data Engineer row applies only on data-centric projects; the other six roles are present on every project. See the Data Engineer role for detail.

Where AI touches the flow

flowchart TB
    subgraph Direction
        A1["ST/PO: backlog drafting & report digest<br/>PM: status & forecasting"]
    end
    subgraph Definition
        A2["BA: stories, ACs, edge cases<br/>SA: design options, ADRs, IaC"]
    end
    subgraph Build
        A3["DEV: pair programming, unit tests"]
    end
    subgraph Verify
        A4["Quality Eng.: AI review + SAST<br/>+ test gen + automation + triage"]
    end
    subgraph Run
        A5["Cloud Eng.: AI-assisted IaC<br/>+ deploy + monitor + AIOps"]
    end
    Direction --> Definition --> Build --> Verify --> Run --> Direction

    classDef ai fill:#eef6ff,stroke:#3b82f6,color:#1e3a8a;
    class A1,A2,A3,A4,A5 ai;

Categories of AI assistance

Across all roles, the help falls into five repeatable patterns:

1. Draft generation - stories, code, tests, designs, reports. AI produces a first version; a human edits and owns it.
2. Synthesis & summarization - turning long input (transcripts, logs, diffs, feedback) into something digestible.
3. Augmented review - a tireless first-pass reviewer for code, security, requirements, and tests.
4. Discovery - surfacing edge cases, risks, options, and root causes a human might miss.
5. Acceleration of toil - boilerplate, scaffolding, status reports, changelogs.

The constant rule

AI drafts and assists. A human reviews, decides, and owns the outcome.

No AI output ships, merges, or is sent to a stakeholder without a named human accountable for it. The guardrails that enforce this are in Integration & Governance.

Tool selection principles

• Prefer tools that integrate where work already happens (IDE, PR, board) over context-switching to a separate app.
• Keep a human-in-the-loop checkpoint for anything that affects security, money, scope, or external communication.
• Favor portability. Patterns over products - when a better model or tool appears, the workflow shouldn’t have to change.
• Respect data boundaries. Use enterprise/team tiers that don’t train on our code or client data; never paste secrets or client IP into consumer tools.