AG ARCH

USE CASES

One control plane for many kinds of agentic work.

The important distinction is not only which industry an AI agent works in. It is what kind of autonomous job is being delegated, which capabilities are orchestrated, what authority applies, and what must become true before the work can be called complete.

AG Arch orchestrates and governs work executed across agents, models, and tools under one execution contract and one outcome standard — from goal through action to evidence, verdict, and verified outcome.

Canonical path

  1. Goal / mandate
  2. AG Arch orchestrate + govern
  3. AI agent / model / tool
  4. Real-world action
  5. Observation + evidence
  6. AG Arch verdict
  7. Verified outcome

Five kinds of agentic work

Organized by the kind of autonomous job — not by industry — with AI agents, models, and tools explicit in every scenario.

Make Changes

Typical AI capabilities

  • Coding agent
  • Infrastructure agent
  • Campaign / content agent
  • Operational automation agent

Example work

  • Modify software or infrastructure
  • Update configuration or records
  • Change campaign or operational state

AG Arch role

  • Ground the work in current truth
  • Select and route the right capability
  • Constrain allowed actions and keep authority attached
  • Observe the real state change, collect evidence, and verify the intended effect
  1. Goal
  2. AG Arch control
  3. AI agent / tool
  4. Real action
  5. Evidence
  6. Verdict
  7. Verified outcome

Current truth · Authority · Constraints · Routing · Recovery

Decide / Recommend

Typical AI capabilities

  • Analysis / reasoning agent
  • Specialist / domain agent
  • Evidence synthesis agent

Example work

  • Legal or case assessment
  • Procurement recommendation
  • Finance / compliance analysis
  • Research recommendation

AG Arch role

  • Preserve source truth and evidence
  • Separate capability from authority
  • Constrain the recommendation mandate
  • Escalate where human judgement, risk acceptance, or irreversible commitment is required
  1. Goal
  2. AG Arch control
  3. Reasoning / specialist agents
  4. Synthesis + evidence
  5. Verdict / escalation
  6. Verified outcome

Current truth · Authority · Constraints · Routing · Recovery

Run Processes

Typical AI capabilities

  • Planner agent
  • Specialist agents
  • Tool-using operational agents
  • Verifier agent

Example work

  • Onboarding
  • Support resolution
  • Claims or finance operations
  • Supplier / back-office workflows

AG Arch role

  • Orchestrate handoffs among multiple agents and systems
  • Preserve one governed execution contract across handoffs
  • Keep goal, authority, evidence requirements, and completion criteria stable while executors change
  1. Goal
  2. AG Arch control
  3. Planner agent
  4. Specialist agents + tools
  5. Verifier agent
  6. Evidence → verdict
  7. Verified outcome

Current truth · Authority · Constraints · Routing · Recovery

Monitor & Respond

Typical AI capabilities

  • Monitoring / detection agent
  • Reasoning agent
  • Bounded response / remediation agent

Example work

  • Infrastructure response
  • Security operations
  • Compliance monitoring
  • Marketing / operations response

AG Arch role

  • Define when the agent may act autonomously
  • Bound response authority and observe effects
  • Escalate when the required response exceeds delegated authority
  • Prove whether the response restored the intended state
  1. Signal / goal
  2. AG Arch control
  3. Detection + reasoning agents
  4. Bounded response
  5. Evidence
  6. Verdict
  7. Verified state

Current truth · Authority · Constraints · Routing · Recovery

Investigate & Learn

Typical AI capabilities

  • Research agent
  • Diagnostic agent
  • Simulation agent
  • Specialist analysis agents

Example work

  • Troubleshooting
  • Root-cause investigation
  • Research and simulation
  • Iterative analysis

AG Arch role

  • Orchestrate iterative investigation
  • Preserve evidence and traceability
  • Support replanning and rerouting as new reality is discovered
  • Separate findings from verified conclusions
  1. Goal
  2. AG Arch control
  3. Research / diagnostic agents
  4. Iterative action
  5. Evidence trail
  6. Verdict
  7. Verified conclusion

Current truth · Authority · Constraints · Routing · Recovery

Cross-cutting orchestration pattern

This is not a sixth work type. Multi-agent orchestration can occur inside any of the five kinds of agentic work when several AI capabilities must route and hand off under one AG Arch control contract.

Multi-agent orchestration under one control contract

AG Arch does not merely send a prompt to one model. It orchestrates and governs work across interchangeable agents, models, tools, and systems without transferring authority to those capabilities.

  1. Goal
  2. AG Arch
  3. Planner agent
  4. Specialist agent A / B
  5. Tool / system action
  6. Verifier agent
  7. Evidence
  8. AG Arch verdict
  9. Verified outcome / recovery

Shared across all work types

Same control model underneath

Domains and agents change, but the control plane does not have to be reinvented for every job.

  • Current truth
  • Goal / outcome criteria
  • Authority + constraints
  • Capability selection / routing
  • Agent handoff / orchestration
  • Observation + evidence
  • Verdict
  • Recovery / replan / re-authorize / reroute / safe stop

Current proving domain

Proving the model against real autonomous work

Software Engineering

Software Engineering is where AG Arch is being proven today: real goals, real tools, state changes, permission boundaries, failures, evidence, verification, and recovery. It is not the product boundary — it is current proving evidence.

Cross-domain examples elsewhere on this page are applicability scenarios. They do not imply that every domain integration, connector, or Domain Pack already ships.

  1. Issue / goal
  2. AG Arch — repo/runtime truth, authority, constraints, proof requirements
  3. Planning / coding agent
  4. Tools / GitHub / runtime
  5. Testing / verification agent
  6. Observed real state
  7. Evidence
  8. AG Arch verdict
  9. Verified outcome / recovery

These are not separate AG Arch products. They are different kinds of agentic work governed through the same control plane.

How It WorksDiscuss a use case