Skip to content

Workflows

Turning AI capabilities into repeatable business processes.

Ten structured workflows combine the catalog's specialized agents into complete, end-to-end processes — from framing an AI product and setting an AI strategy to testing and running a project post-mortem. Each one defines what happens, in what order, and what must be validated before the next step can start.

Each workflow is a dated, versioned file, and that same file is what the Agentic Runtime executes — so the procedure on paper and the work actually performed cannot drift apart. These processes are written in BPMN, the standard notation companies already use to draw a process — an onboarding, a claims file — here applied to work carried out by agents.

How a workflow is defined

Each workflow file carries a YAML identity card giving the workflow's number, what it is for and the agents it calls, plus an ASCII BPMN flow diagram, per-agent step sheets, and a final deliverable checklist. Core agents run the backbone, the steps every run of that workflow goes through; optional agents attach on context (GDPR, SAFe, business case, organizational change). In high-stakes runs, an optional counter-review gate can be inserted before the final report: a separate agent (the AI Methodology Auditor), outside the chain it reviews, checks how the work was done — it challenges methodological rigor, it never redoes the domain work. Each identity card also carries a recommended model for the run; the operator keeps the final say on which model actually executes the chain.

The ten workflows

ID Workflow Domain Agents Duration Status
WF-001 AI Product Scoping Agile & Product 4–10 45–90 min ✅ Live — 3 / 3
WF-002 SAFe Agile Delivery Agile & Product 6–8 60–120 min ✅ Live — 5 / 5
WF-003 AI Application Launch Dev & Engineering 7–12 90–180 min ✅ Live — 7 / 7
WF-004 AI Consulting Engagement Management & Consulting 6–9 60–90 min ✅ Live — 6 / 6
WF-005 Strategic Intelligence & Growth Management & Consulting 3–6 30–60 min ✅ Live — 3 / 3
WF-006 Pre-sales / Commercial Proposal Management & Consulting 6–9 75–120 min ✅ Live — 6 / 6
WF-007 Mission Onboarding — Day 1–5 Management & Consulting 4–6 45–75 min ✅ Live — 4 / 4
WF-008 AI Act / GDPR Compliance Audit Compliance & Governance 7–11 90–150 min ⟲ Returned
WF-009 IT / AI Recruitment HR & Talent 4–7 60–90 min ✅ Live — 6 / 6
WF-010 Project Post-mortem Management & Consulting 4–8 45–75 min ✅ Live — 4 / 4

All ten processes have been run for real, from beginning to end — real executions on the live model, each one recorded and dated, not a demo assembled for this page. Nine went the whole way and passed every check. The tenth — the AI Act / GDPR compliance audit — got six steps in and was then stopped by its own review step, which sent it back for rework with five documented reservations rather than let the report go out. It is listed here on the same footing as the nine: a chain that can stop itself is the harder thing to show.

The four terms this page uses

  • Workflow — a chain of agents that carries a request through to a finished deliverable. This page also calls one a process; they are the same thing.
  • WF-001 … WF-010 — the reference number of each process. The same number is used on every page of this site, so any claim can be traced back to one run.
  • Catalog version — which dated edition of the agent library a run used. WF-001 ran on v4.0.0, WF-002 and WF-003 on v3.27.0, and the seven others — WF-008 included — on v4.1.0. Naming it is what keeps a result checkable months later: the agents keep evolving, while each run stays pinned to the exact contents it had on the day.
  • Returned — a run that a review step stopped before its deliverable. WF-008 is the one; six of its seven steps had passed.

WF-001 — AI Product Scoping

Agile & Product · 4–10 agents · 45–90 min

  • Problem — A client brief lands, or a product idea needs shaping, but there is no prioritized backlog and no acceptance criteria to build against.
  • Capability — Chains four core agents — Business Analyst → UX Designer → Product Owner (Scrum) → QA Agile — turning a raw brief into a MoSCoW-prioritized initial backlog with Gherkin acceptance criteria.
  • Proof — ✅ Proven live, end to end: 3 / 3 steps passed on catalog v4.0.0. See the run →

WF-002 — SAFe Agile Delivery

Agile & Product · 6–8 agents · 60–120 min

  • Problem — A PI Planning is starting, or an ART sprint kicks off, and the team needs aligned PI objectives, a prioritized program backlog, and reporting the executive committee will trust.
  • Capability — Chains six core agents — Product Manager SAFe → Release Train Engineer → Product Owner SAFe → Scrum Master → QA Agile → AI Project Manager — producing WSJF-prioritized PI objectives, a sprint plan, and an executive-committee dashboard.
  • Proof — ✅ Proven live, end to end: 5 / 5 steps passed on catalog v3.27.0. See the run →

WF-003 — AI Application Launch

Dev & Engineering · 7–12 agents · 90–180 min

  • Problem — A business case is validated and the green light is given, but going from idea to a deployed, secured AI application spans architecture, code, CI/CD, and a security audit.
  • Capability — The densest pipeline in the catalog: Financial Analyst → Prompt Engineer → AI Architect → AI Python Developer → QA Agile → DevOps / Cloud → AI Security — delivering a deployed app, an operational CI/CD pipeline, and a passed OWASP LLM security audit.
  • Proof — ✅ Proven live, end to end: 7 / 7 steps passed on catalog v3.27.0. See the run →

WF-004 — AI Consulting Engagement

Management & Consulting · 6–9 agents · 60–90 min

  • Problem — An AI consulting engagement is signed, and the client expects a maturity diagnostic, a costed roadmap, a training plan, and an executive-ready deliverable — not just an opinion.
  • Capability — Chains six core agents — AI Consultant → Financial Analyst → CDO / Chief AI Officer → Change Manager → AI Trainer → AI Content Writer — producing a maturity audit, a 12–24 month roadmap with AI OKRs, an ADKAR adoption plan, and an executive summary.
  • Proof — ✅ Proven live, end to end: 6 / 6 steps passed on catalog v4.1.0. See the run →

WF-005 — Strategic Intelligence & Growth

Management & Consulting · 3–6 agents · 30–60 min

  • Problem — A weekly cadence or a detected market signal needs to become a qualified synthesis and publishable thought-leadership content, fast and at controlled cost.
  • Capability — Chains three core agents — Strategic Intelligence → AI Growth / Marketing → AI Content Writer — turning raw signals into a qualified intelligence radar, an SEO content plan, and posts ready to publish.
  • Proof — ✅ Proven live, end to end: 3 / 3 steps passed on catalog v4.1.0. See the run →

WF-006 — Pre-sales / Commercial Proposal

Management & Consulting · 6–9 agents · 75–120 min

  • Problem — An RFP arrives, and winning it means a GO/NO-GO decision plus a full technical-commercial proposal: scope, architecture, schedule, person-day costing, price, and prospect ROI.
  • Capability — Chains six core agents — AI Consultant → Business Analyst → AI Architect → AI Project Manager → Financial Analyst → AI Content Writer — producing a qualification verdict, a target architecture, a costed schedule, and the final commercial proposal.
  • Proof — ✅ Proven live, end to end: 6 / 6 steps passed on catalog v4.1.0. See the run →

WF-007 — Mission Onboarding — Day 1–5

Management & Consulting · 4–6 agents · 45–75 min

  • Problem — A new engagement starts, and the first days set the tone: a kickoff plan, a Day-1 kit, a mapped client context, and the right relationships must exist by Day 5.
  • Capability — Chains four core agents — AI Project Manager → Business Analyst → Change Manager → AI Content Writer — producing a kickoff plan with a provisional RACI, a client context sheet, a stakeholder engagement plan, and a Day-1 report.
  • Proof — ✅ Proven live, end to end: 4 / 4 steps passed on catalog v4.1.0. See the run →

WF-008 — AI Act / GDPR Compliance Audit

Compliance & Governance · 7–11 agents · 90–150 min

  • Problem — An AI system must be audited under regulatory pressure — a CNIL / AI Office inspection or M&A due diligence — where a mis-qualified risk tier exposes the organization to penalties.
  • Capability — Chains seven core agents — AI Legal → AI Architect → AI Security → Data Engineer → CDO → Change Manager → AI Content Writer — producing an obligations matrix (AI Act + GDPR + NIS2), a risk mapping, a prioritized remediation plan, and target governance. An optional AI Methodology Auditor gate can challenge the audit's rigor before finalization.
  • Proof — ⟲ Returned for rework, live, on catalog v4.1.0. The six pre-report steps passed; the counter-review gate (AI Methodology Auditor) then returned the audit — five documented reservations, a bias log, ISTQB exit criteria — so the report step (STEP-07) was withheld by design. A gate that can say no. See the run →

WF-009 — IT / AI Recruitment

HR & Talent · 4–7 agents · 60–90 min

  • Problem — An IT / AI hiring need is identified, and filling it means a precise profile, a defensible technical assessment, an anti-fraud sourcing pass, and a publishable offer.
  • Capability — Chains four core agents — Business Analyst → AI Consultant → AI HR → AI Content Writer — producing a need sheet with must/nice criteria, a technical assessment grid, a scored shortlist with deepfake/CV fraud checks, and a publishable job ad.
  • Proof — ✅ Proven live, end to end: 6 / 6 steps passed on catalog v4.1.0, on a synthetic, provided candidate pool — not autonomous sourcing (the pool is fictional by design, GDPR-safe for a public trace). The very same shortlist gate halted a no-candidate variant of the run, proving it discriminates rather than rubber-stamps. See the runs →

WF-010 — Project Post-mortem

Management & Consulting · 4–8 agents · 45–75 min

  • Problem — A project closes or a major incident hits, and the lessons risk evaporating without a structured root-cause analysis and a shared improvement plan.
  • Capability — Chains four core agents — AI Project Manager → QA Agile → Change Manager → AI Content Writer — producing a timeline with a 5-Whys analysis, a quality and technical-debt review, a team review, and a lessons-learned report. An optional AI Methodology Auditor gate can challenge the root-cause reasoning before the improvement plan is set.
  • Proof — ✅ Proven live, end to end: 4 / 4 steps passed on catalog v4.1.0 — the last of the ten to be run live. See the run →

Workflows orchestrate the catalog's 38 agents and run on the Agentic Runtime. Certifications named in each workflow describe a simulated persona's knowledge frame, not credentials held by the author, the software, or the model — see the catalog disclaimer.