A4 · Coordinated agents · manufacturing, public sector, professional services, energy, technology
Agentic operations workflow
Several specialists, one budget, one approval gate, one record of what happened.
- Customer job
- Move a purchase request from intake to approved order without chasing five people and three systems.
- Today
- Requests sit in email threads; buyers collect quotes by hand, policy checks happen late, and approvers see incomplete context.
- Accepted result
- An approved order created once, with quotes, policy checks and the approver's decision attached to the request record.
A durable workflow coordinates specialist agents through a multi-day business process such as a purchase request: gather quotes, check policy, analyse cost, request approval, then create the order exactly once. Coordination is used only where it measurably beats a single agent.
Google Cloud / primary provider
What carries each step.
Workflows owns durable steps and waits; ADK multi-agent teams run on Agent Runtime; Cloud Tasks gives deduplicated fan-out.
Data and model estate in one place: Gemini on Agent Platform, Agent Search and RAG Engine, AlloyDB and BigQuery, Cloud Run for everything else.
| Component | Service | Evidence |
|---|---|---|
| Person | ||
| Cloud Run | prior, ga | |
| Cloud Run | prior, ga | |
| Workflows | prior, ga | |
| Agent Runtime (Agent Platform) | source, ga | |
| ADK agents on Agent Runtime | prior, ga | |
| Apigee (AI gateway policies) | prior, ga | |
| Gemini 3.8 Flash (Agent Platform) | source, ga | |
| MCP server on Cloud Run | prior, ga | |
| Cloud Run jobs (sandboxed) | prior, ga | |
| Cloud Tasks | prior, ga | |
| AlloyDB for PostgreSQL | prior, ga | |
| Review console on Cloud Run | prior, ga | |
| External system | ||
| Cloud Logging, Monitoring, Trace | prior, ga | |
| Evaluation service (Agent Platform) | confirm |
Customer view / judgement, not benchmark
Weigh what this customer values.
Remove providers that fail a hard constraint first: residency, data rights, identity estate, latency, operating owner, budget. Then weigh the rest. Scores are architect judgement on a 1-5 scale for the pattern as specified, recorded with a rationale. They are not benchmarks. A provider's consumption opportunity never changes a customer score.
Highest weighted: Vercel, 3.83 of 5.
Performance 4Time to value 4Speed to execution 5Cost per accepted result 3Exit cost 4Governance fit 3
One TypeScript codebase for steps, agents and UI; Queues launch stage must be confirmed.
Performance 4Time to value 4Speed to execution 4Cost per accepted result 4Exit cost 3Governance fit 3
Strong primitives for stateful agents; enterprise ERP connectivity is integration work.
Performance 4Time to value 3Speed to execution 3Cost per accepted result 3Exit cost 3Governance fit 5
Complete managed set; multi-agent coordination still needs the uplift proof.
Performance 4Time to value 3Speed to execution 3Cost per accepted result 3Exit cost 3Governance fit 5
Mature orchestration and controls; most configuration.
Performance 4Time to value 3Speed to execution 3Cost per accepted result 3Exit cost 3Governance fit 5
Good fit for Microsoft estates with Teams-based approvals.
Performance 3Time to value 2Speed to execution 3Cost per accepted result 4Exit cost 5Governance fit 2
Maximum control; durability and isolation are the hardest parts to get right yourself.
Provider ledger / shown after the choice
What grows on Google Cloud.
These meters are the provider team’s consumption view. They explain the cloud workload; they never raise a customer score.
- Gemini input and output tokens
- Cloud Run vCPU-seconds and requests
- Agent Search queries and index storage
- AlloyDB or Cloud SQL instance hours
- BigQuery slots or bytes scanned
Take it with you / generated from one specification
Exact views, and a running start.
Google Cloud views
- SVG diagram for decks and documents
- diagrams.net file editable, shared IDs
- React Flow JSON for your own canvas
- Mermaid flowchart
Every provider
- Structurizr workspace C4 containers, one deployment view per provider
- Journey sequence (Mermaid)
One-click template
No Google Cloud template yet.
The same answer contract runs on these runtimes today; its conformance suite is the porting checklist.
- Vercel: anthropic/claude-sonnet-5.5 with gateway fallbacks
- Cloudflare: @cf/meta/llama-3.3-70b-instruct-fp8-fast
- Google Cloud: gemini-3.8-flash on the global endpoint
Customer ledger / assumptions you can move
Cost per accepted result.
Every attempt is charged, so failed runs count. Value comes only from accepted results. Illustrative assumption for a central procurement team. A4 must show it beats an A2 workflow on the same cases before this case holds.
Primary metric: Cycle time to approved order. Median hours from request intake to order created, for requests completed without rework.
- Monthly value
- €23,100
- Monthly cost
- €6,540
- Net
- €16,560
- Cost per accepted result
- €18.17
Experience / drawn from the specification
What the person sees, including no.
Screens are rendered from structured blocks in the catalog, so a refusal or an exception is designed before it happens. Values marked fixture are illustrations, not measurements.
PR-2291: 40 industrial sensors for Line 3
- Intake validated
- 3 quotes requested
- Policy check passed
- Cost model ready
- Awaiting approval
Approve PR-2291?
| Supplier | Unit | Lead time |
|---|---|---|
| Northwind | 212 | 3 wk |
| Fabrikam | 198 | 6 wk |
| Contoso | 225 | 2 wk |
- Recommendation: Northwind, lead time meets the line restart date
- Policy: within category budget
| Agent | Steps | Spend |
|---|---|---|
| Coordinator | fixture | fixture |
| Sourcing | fixture | fixture |
| Policy | fixture | fixture |
| Cost | fixture | fixture |
A4 is retained only while it beats the A2 baseline.
Team workflow / who carries what
Every lane leaves evidence.
| Lane | Responsibility | Evidence it leaves |
|---|---|---|
| Customer | Submit need, answer clarifications, track status | Request record |
| AI | Coordinator plans; specialists research, check and model costs | Findings with sources |
| System | Durable steps, dedup, shared budget, idempotent order | Run ledger and effect log |
| Reviewer | Approver decides with full context; can send back | Signed decision |
| Operator | Compare A2 and A4 monthly; adjust budgets and roles | Uplift report |
Release evaluations
What must pass before anyone ships.
End-to-end success
>= 0.8 of frozen requests reach a correct approved order
Fixture ERP and supplier stubs; scorer checks final state.
Coordination uplift
A4 beats A2 on accepted results or cost per accepted result
Paired replay of the same cases.
Duplicate effects
0 duplicate supplier emails or orders
Restart and retry fault injection.
Approval bypass
0 orders without a recorded decision
Adversarial cases asking agents to skip approval.
Risks and controls
What would go wrong first.
Agents multiply spend without better outcomes.
Shared run budget; uplift test; automatic downgrade to A2.
An agent creates an order without approval.
Order tool only callable by the workflow after a recorded decision.
Specialists disagree silently.
Coordinator must record conflicts and show them to the approver.
Image recipes / concept art only
When to generate, and when never to.
16:9 · gpt-image-2 or a current Gemini image model
Customer moment for a proposal cover or portal hero.
Cinematic editorial photograph, late afternoon light, shallow depth of field. Move a purchase request from intake to approved order without chasing five people and three systems. The moment just after the accepted result: An approved order created once, with quotes, policy checks and the approver's decision attached to the request record. Real workplace, restrained palette of deep navy, warm paper and one cobalt accent. Natural expressions, no screens facing the camera. Concept art only. Do not render product logos, cloud icons, UI text or architecture labels; compose exact labels from the specification in code.
3:2 · gpt-image-2 (consistent characters across panels)
Team workflow storyboard for a workshop or executive briefing.
Five-panel storyboard, clean ink line with flat colour, consistent characters across panels. Panels follow the lanes customer, ai, system, reviewer, operator: (1) Submit need, answer clarifications, track status (2) Coordinator plans; specialists research, check and model costs (3) Durable steps, dedup, shared budget, idempotent order (4) Approver decides with full context; can send back (5) Compare A2 and A4 monthly; adjust budgets and roles Each panel shows hands and decisions, not screens. Concept art only. Do not render product logos, cloud icons, UI text or architecture labels; compose exact labels from the specification in code.
Do not generate
Any image a customer will read for component names, data classes or provider services.
Do not generate. Export the compiled view for the chosen provider. A generated architecture image can invent services, misspell names and draw icons the vendor never published.
A design, not a deployment: no realization on this page carries a provider receipt. Generated from the pattern source at revision c43825efda38. Apache-2.0 for catalog content and exports; provider names are trademarks of their owners; Google Cloud icons retain Google rights.
Next pattern: AI governance plane