Skip to content

AI AgentsThe work that runs itself

Agents that act on your data, in your systems, under your rules

Not a chatbot bolted to the side of your CRM. Agents that read your records, take the action, and hand off to a person the moment the rules say they should.

  • Acts inside the CRM
  • Scoped to a role or a queue
  • Human approval gates
  • Every action logged

Your team is not short of AI that can talk about the work. They are short of hours to do it.

Records go stale. Tickets sit in the wrong queue. Renewal quotes get drafted the week after the renewal. None of that is a thinking problem — it is a doing problem, and it happens in your systems, not in a chat window.

How it works

Four steps, and you can stop at any of them

Putting an agent to work takes four steps: scope it to a single role, write the rules, run it on real records, then read the log. Nothing happens behind your back — you set the scope, you approve the rules before it runs, and the log is open any day of the week.

  1. Scope the agent

    Pick the role it stands in for — an SDR, a support triage officer, a renewals coordinator. That role decides which records it can see and which it can change. Nothing wider.

  2. Write the rules

    The rules are plain, reviewable statements: what to act on, what to leave alone, what to hand to a person. You approve them before the agent runs, and you can change them any afternoon.

  3. Run it on real work

    The agent picks up records from a queue or a pipeline stage, does the work in your system, and writes the outcome back against the record. No side database, no shadow copy.

  4. Watch the log

    Every read, every write, every escalation is timestamped against the record it touched. If an agent did something you did not expect, you can see exactly what and when.

Governance

An agent with a boundary drawn around it

Every agent runs inside three bands: what it may read, what it may write, and what it must hand to a person. The boundary is set before the agent runs, not negotiated afterwards, and there is nothing outside it.

Agentone roleREADWRITEHUMAN APPROVALOutside this boundary: nothing.
Read
Accounts, contacts, tickets and notes inside its scope.
Write
Create, update, route and escalate — inside the same scope.
Approve
Anything that leaves your organisation waits for a person.

Agent activity log

Renewals · Tues
  • 09:02Enriched account record from the ERP
    Account · Harbour Industrial SuppliesActioned
  • 09:14Routed ticket to the Billing queue
    Ticket · #48219 — invoice mismatchActioned
  • 09:31Drafted renewal quote for review
    Deal · Renewal FY27 — Northline FreightAwaiting approval
  • 10:07Flagged duplicate contact before merge
    Contact · two records, same work emailEscalated to human
  • 10:22Updated stage after signed order was received
    Deal · Fit-out stage 2 — Corbett GroupActioned
  • 10:48Composed customer reply, held for sign-off
    Ticket · #48244 — delivery delayAwaiting approval

Illustrative log. Every line is written against the record it touched, so the history lives with the data, not in a separate tool.

What agents do not do here

  • An agent will not touch a record outside the scope you gave it. There is no global tenant key behind the scenes.
  • An agent will not send an email, a quote or a refund to a customer without a named person approving it first.
  • An agent is not a replacement for a bad process. If the workflow is broken, automating it makes the mess faster.
  • An agent will stop and escalate when the rules do not cover the case. It does not guess its way through ambiguity.

Capabilities

What you get, plainly

Every agent comes with scoped permissions, human approval gates on customer-facing work, routing and escalation, an action-level audit trail, continuous record hygiene, and connections that run on your own systems.

Scoped permissions

Read and write limits are set per agent, per role. The agent sees the same slice of data the person it stands in for would see.

Human approval gates

Anything customer-facing queues for sign-off. A named person approves, edits or rejects before it goes anywhere.

Routing and escalation

Tickets, leads and cases land in the right queue with the right owner. Anything outside the rules goes to a person, not to a guess.

Action-level audit

Every action is written against the record it touched, with a timestamp and the rule that triggered it. Exportable for review.

Record hygiene

Duplicate detection, field completion and enrichment from the systems you already sync. Cleanup runs continuously, not quarterly.

Runs on your systems

Agents act through the same authenticated connections OptiSync uses. No shadow copy of your database, no data leaving your stack unannounced.

Questions

The things buyers actually ask

Accountability, guardrails, and how these agents differ from the AI already in your CRM — answered the way we would answer them on a call.

How is this different from the AI already built into our CRM?

The AI built into most CRMs summarises and suggests: it writes a draft, and a person still does the work. These agents complete the work inside the record — create, update, route, escalate — and log what they did. The measure is whether the queue got shorter, not whether the copy read well.

What stops an agent doing something expensive or embarrassing?

Two things stop it: scope and approval gates. Scope means an agent can only touch the records in its lane. Approval gates mean anything leaving your organisation — an email, a quote, a refund, a credit — waits for a named person. An agent that hits a case its rules do not cover stops and escalates.

Who is accountable when an agent gets it wrong?

You are, the same as when a staff member gets it wrong. That is why the log matters. Every action carries a timestamp, the record it touched and the rule that fired, so you can see the cause and change the rule. We will help you fix it, but we will not pretend the agent is not yours.

Do we have to change our CRM or our process first?

No — agents work on the CRM and the data model you already run, so nothing has to change first. That said, if the underlying process is unclear, we will say so before we build. A confused process automated is just a faster mess.

How long before an agent is doing useful work?

A single scoped job — one queue or one pipeline stage — is doing useful work in weeks, not quarters. Start narrow, watch the log for a fortnight, then widen the scope once you trust what you are reading.

Pick one queue. Let an agent run it.

Bring the job your team keeps putting off — the enrichment, the triage, the renewal chase. Thirty minutes and we will tell you straight whether an agent should be doing it.

No obligation. No procurement process to start a conversation.