Deployment

Make your company agent-ready, one workflow at a time.

Every Annex agent starts the same way: select the workflow, connect the systems it needs, set its permissions and approvals, validate it against agreed scenarios, then activate it. There is no rebuild of the whole company, and the next agent reuses what the first one set up.

1Selectone workflow agent2Connectthe systems it needs3Set limitspermissions, approvals4Validatedrafts, reviewed by people5Activatelive, then the next oneQuote requestsDiscounts up to 7%Drafts go to J. WeberNext: purchasing1Selectone workflow agent2Connectthe systems it needs3Set limitspermissions,approvals4Validatedrafts, reviewedby people5Activatelive, then the next oneQuote requestsDiscounts up to 7%Drafts go to J. WeberNext: purchasing1Selectone workflow agentQuote requests2Connectthe systems it needs3Set limitspermissions, approvalsDiscounts up to 7%4Validatedrafts, reviewed by peopleDrafts go to J. Weber5Activatelive, then the next oneNext: purchasing

What gets installed

What must a business install?

Nine things, all scoped to the first agent. Most of them already exist somewhere in the company; a deployment makes them explicit and checkable in the Company Brain and Decision Assurance.

  1. 01

    Source systems

    The systems the agent reads and writes, and which record is authoritative for each fact.

  2. 02

    Business roles

    Who owns the work, who may approve what, and who is told when the agent stops.

  3. 03

    Policies

    The rules the agent follows: price lists, discount limits, budgets, approved suppliers.

  4. 04

    Permissions

    The agent’s app permissions in each system, as narrow as its job allows and kept apart from its business authority.

  5. 05

    Commitment checks

    Which actions create a commitment, and the source-system path to reserve or confirm each one.

  6. 06

    First agent

    One agent with a bounded job: a clear start, a clear end and an owner.

  7. 07

    Test scenarios

    Routine cases, exceptions, stale data, conflicts and failed writes, run before the agent acts.

  8. 08

    Go/no-go criteria

    What must be true before the agent works on real requests, agreed with the owner in advance.

  9. 09

    Expansion

    What the next agent can reuse, and what it must add and test on its own.

Five steps

How does a first deployment run?

  1. Select a workflow agent.

    Pick the workflow with frequent, rule-based work, an owner and outcomes a person can check.

    e.g. Annex Buy for purchasing

  2. Connect your business systems.

    Connect the systems it needs, and record which one is authoritative for each fact and how actions are performed there.

    ERP · Inbox · Pricing

  3. Define permissions and approval rules.

    Agree what the agent may do on its own, who approves above its limits, its app permissions and what happens when something fails.

    Limits · Approvals

  4. Validate the workflow.

    Run it against agreed scenarios (routine cases, exceptions, stale data, conflicts and escalation) until it meets the go/no-go criteria.

    Test scenarios · Go/no-go

  5. Activate the agent.

    It starts on real work within its limits, with a person on every exception. The next agent reuses what is already connected.

    Live · then the next agent

What we don’t require

Keep the systems you have.

Annex works with your relevant systems of record. Nothing to replace: no new ERP, and no complete digital rebuild first.

  • No all-at-once rollout

    The first agent has a deliberately limited scope.

  • No free-roaming agent

    Every action has a boundary, and a person approves above it.

  • No network dependency

    A first agent inside one company is useful before any other company is involved.

Effort depends on the systems involved, data quality and how complex the policies are.

Choose the first agent.

One bounded role is enough to start.

Tell us the work, the systems it touches and who approves today. Test scenarios and go/no-go criteria are agreed before an agent acts.