What is agentic AI for IFS?

How an agent reads, decides and acts in IFS.

The short answer

An agent works out the steps, then takes them in IFS.

Published
Updated
Reading time
3 minutes
Covers
IFS Cloud

How an agent differs from chatbots, screen bots and fixed flows

A chatbot answers. An agent is given a goal and a set of tools, then loops: read, pick a tool, check the result, choose the next step.

If an invoice’s currency differs from its order’s, the agent stops and asks a person.

How far it may go on its own is a setting you choose.

Chatbot, screen bot, fixed flow or AI agent
ApproachWhat it doesWhere it stops
ChatbotAnswers in a chatThe work stays undone
Screen botReplays clicks on screensA moved field breaks it
Fixed flowMoves data on mapped pathsThe unmapped goes to a person
AI agentPicks its tools as it runsIt needs limits and approvals

What an agent needs in IFS

Four things keep an agent safe to run against live IFS records.

  1. Typed tools

    Projections to read, official IFS actions to write.

  2. Safe writes

    An ETag on every IFS update.

  3. Limits

    The IFS permissions of the person it acts for.

  4. A person’s yes

    An approval before the writes you choose.

One invoice, worked by an agent

It acts when the match is clean and asks when it is not.

Read the invoice capture guide

NgageFlow 4 steps

New email in the AP inbox: invoice NC-24152 from Nordia Components

  1. 1 Read the email and the PDFNC-24152, 6 lines, 8,400.00 EUR Outlook
  2. 2 Find the supplier by tax IDNordia Components, active IFS
  3. 3 Compare the lines with PO-20461Line 3 is above the agreed price IFS
  4. 4 Ask the AP lead to decideApprove, correct or reject Teams
An illustrative run on a test landscape. Supplier, order and amounts are samples.

Questions IFS teams ask

Is an AI agent the same as a copilot?

No. A copilot helps a person while they do the work: it drafts, suggests and summarises. An agent does the work itself. It is given a goal and a set of tools, takes the steps, and hands over to a person when a decision needs judgment.

Do agents replace IFS workflows or customisations?

No. An agent works beside IFS, through IFS projections and official IFS actions, so the posting rules, validations and authorisations you already have still apply to everything it does. Nothing is installed in IFS.

How do we stop an agent doing something wrong in IFS?

Limit it the way you would limit a new colleague. It runs within the IFS permissions of the person it acts for, never through a shared service account. It uses the tools you allow and no others, within a budget, and the writes you choose wait for an approval. Every step goes into the audit trail.

What can start an agent in IFS?

A record created or changed in IFS, an email, a chat message, a schedule or a webhook. Before it writes, the agent reads the projection’s metadata, so it knows which fields are required.

Does this apply to IBM Maximo and SAP?

Yes. IBM Maximo, IFS and SAP work out of the box, through each ERP’s official interfaces, and one flow can use two of them, for example the asset in IFS and its cost in SAP. For any other application, including a homegrown one, we teach Ngage how it works in weeks. This guide uses IFS terms because it is written for IFS teams.

How quickly can we try an agent on our own IFS?

On IFS, a flow can be live in hours, depending on access to your test environment and the scope.

Try an agent on a process from your IFS

Describe it in a sentence or two. We run it with you, live, in a safe test setup.