Skip to content

AI Workforce & Assistance

Operations assistant: why it's stuck, and the fix, ready to approve

A failed sync, a job closed without photos, a part that won't arrive in time. Working out why can take half a morning across four systems. An operations assistant reads the logs and records, explains the cause in plain words, and prepares the fix. Staff approve it, and the action runs and is logged. We connect it to your systems and decide with you which fixes it may prepare.
Talk through your operationsSee three exceptions explained ↓
09:12INV-2318 didn't sync to accountingSync logSY-344 rejected“tax status missing”Customer recordAcme Propertiescertificate gone Oct 21Change historytwo contacts mergedby the office, Oct 21WhyAccounting rejected INV-2318 because Acme's tax-exempt certificate was lostwhen two contacts were merged on Oct 21.Prepared fixRe-attach certificate TX-4471 from the merged contact, then retry SY-344.changes 1 customer record · retries 1 syncApproveEditoffice lead✓ synced at 09:31 · INV-2318 is in accounting

Something is always stuck. Finding out why takes the morning.

Every business that runs on several systems has exceptions: an invoice that didn't reach accounting, a job missing its photos, a delivery that moved. The error message rarely says why. Someone opens the sync log, the customer record and the change history, pieces it together, and only then fixes it.

An operations assistant does the piecing together. It explains each exception with the records that prove it and prepares the fix, so staff spend their time deciding, not digging. It prepares; people approve.

Today, and with an operations assistant

Today

  1. 01An error email or a red badge says something failed.
  2. 02Someone opens three or four systems to find out why.
  3. 03The fix depends on who happens to know that system.
  4. 04Nobody writes down the cause, so it happens again.

Exceptions wait for the one person who understands them.

With an operations assistant

  1. 01Each exception arrives explained, with the records that show why.
  2. 02A fix is prepared, with what it will change.
  3. 03Staff approve, edit or reject it, and it runs.
  4. 04Causes are logged, so repeat problems show up.

Less digging, and the same safe fix every time.

Three exceptions, explained

One morning's exceptions, as staff see them. Each comes with its cause, its records and a prepared fix.
09:12INV-2318 didn't sync to accounting

What happened

The invoice was rejected by accounting with “tax status missing”.

Why

Acme's tax-exempt certificate was lost when two of its contacts were merged on Oct 21. Every Acme invoice since then will fail the same way.

Records

  • SY-344 · sync log
  • Acme Properties · customer record
  • Oct 21 · merge in the change history

Prepared fix

Re-attach certificate TX-4471 from the merged contact, then retry SY-344.

If approved

1 customer record changed, 1 sync retried. No customer is contacted.

Approves: office lead

What it may touch

On its own

  • Reads logs, records and history in connected systems
  • Explains exceptions in plain words
  • Drafts fixes and messages
  • Sends internal reminders to staff

With a person's approval

  • Retries syncs and runs
  • Changes records
  • Reschedules work
  • Messages customers, with approved templates only

Never

  • ✕ Deletes records
  • ✕ Touches prices, payments or payroll
  • ✕ Acts in systems it isn't connected to
  • ✕ Approves its own proposals

Where it shows up

Not a blank chat box. It appears where the work already happens.

  1. Morning digest

    What's stuck, in order

    Every open exception, explained, with its fix ready.

  2. Inside the exception

    Next to the error

    Open a failed sync and the explanation and fix are already there.

  3. Ask it

    In plain words

    “Why didn't J-481 get invoiced?” answered from the records, with links.

  4. End-of-day handover

    What's still open

    A short list for whoever's on tomorrow, with what's been tried.

The parts it's built from

Each part is built and tested on its own, so you can start with one and add the rest.

  1. Notice

    Workflow Monitoring & Recovery

    Failures, delays and gaps from your systems.

  2. Explain

    LLM Application Development

    Reads the logs, records and history; writes the cause plainly.

  3. Prepare

    Tool-Using Agents

    • the exception
    • its records
    • the allowed actions

    no safe fix → explain it and assign it to a person

  4. Approve

    Human-in-the-Loop AI

    Staff approve, edit or reject; the action runs and is logged.

Underneath every stepAPI IntegrationConnects to your scheduling, accounting, CRM and supplier systems.

What changes from one business to the next

SystemsWhich it can read, and which it can act in
ExceptionsWhich kinds it watches, and who owns each
ActionsWhat it may prepare, and who approves each kind
Where staff see itA morning digest, inside your admin screens, or a chat workspace

A common starting point is a hosted language model with tool calling, the APIs of your existing systems, n8n or Trigger.dev for the runs behind each action, and Langfuse for tracing. These are starting candidates, not requirements.

Is an operations assistant the right system?

Good fit

  • Staff spend hours a week working out why something is stuck.
  • Your systems keep logs and history it can read.
  • The same kinds of fix come up again and again.

Avoid when

  • Exceptions are rare. A monitoring alert is enough.
  • Your systems can't be connected or don't keep history.
  • You want it to fix things without anyone approving.

How we'd approach it

Most of the early work is listing the systems, the exceptions that cost the most time and the actions staff may approve.

  1. 01

    Understand

    As much as the problem needs

    We map how the work runs today: who does what, which systems hold the records, and where it breaks.

  2. 02

    Design

    Decided before it's built

    Ownership, rules, integrations and exceptions, agreed with you and written down with acceptance criteria.

  3. 03

    Build

    Connected to what you use

    We build the scoped system, connect it to the tools you already run, and test it on the cases that matter.

Related systems

The capabilities underneath

Questions about operations assistants

Can it fix things on its own?

Only internal reminders, by default. Retries, record changes and anything a customer sees wait for a person, and you can loosen that one action at a time once it has proven itself.

How is this different from monitoring alerts?

An alert says something failed. The assistant says why, shows the records that prove it, and has the fix ready.

What if its explanation is wrong?

Every explanation links the records it's based on, so staff can check it in a click. Wrong ones are flagged and become test cases.

Which systems can it work with?

Reading needs an API or export. Acting needs an API that supports that action.

Find out why it's stuck, and fix it

Tell us which exceptions take the longest to work out today. We'll show you how the assistant would explain and prepare each fix.

Start a project