For much of the SaaS era, the product was the center of the sale. Implementation helped customers put it to work. With AI agents, those two things are becoming harder to separate.
An agent does part of the work itself, so it has to learn how your company does that work: which documents matter, which exceptions are normal, who signs off. That knowledge lives inside your operation. So vendors are sending engineers to work inside it too.
On November 10, IBM and Microsoft announced Imagine Enterprise, a platform for building and running industry-specific AI agents. Customers can have IBM Consulting deploy and operate it, and both companies offer forward-deployed engineering support for qualifying engagements. The first industry frameworks cover financial services, insurance, and retail.
Mohamad Ali, who leads IBM Consulting, described the demand behind it:
Clients are no longer looking for AI that simply helps people do the same work a little faster.
We see the same demand at a smaller scale. Much of our work this year has been building agents with customers, with our engineers working directly alongside the people who run imports, compliance reviews, and the month-end close.
What this means for buyers
When engineers come with the software, the deployment becomes part of what you buy. That changes what is worth evaluating.
A demo shows what a product can do. It does not show how much of your deployment the product already handles and how much will be built for you from scratch.
That split shapes your cost, how fast you go live, and how the system keeps improving after launch.
Two kinds of work in every deployment
The first kind is what the product already does. In our products it falls into three areas.
1-1. Reading documents
Invoices, packing lists, bills of lading, certificates of analysis, labels, and ledger exports hold the information agents need. VaultSage reads these document types and connects their contents into one record that agents work from.
1-2. Checking for discrepancies
TraderSage compares quantities, terms, and dates across a shipment's documents and flags inconsistencies for review. Many of these checks can be reused across customers because they test relationships common to trade documents.
1-3. Tracing requirements to their sources
RegSage links each regulatory requirement to the public rule behind it, such as the FDA's dietary supplement manufacturing rules in 21 CFR Part 111.
2. Your own rules
The second kind is your own rules.
Which costs belong to which brand? What counts as a returned sale this month? Who approves a shipment before it is released? The answers live in spreadsheets, internal procedures, and people's heads. Two companies in the same industry can answer them very differently.
This is where forward-deployed engineers do their most useful work. They sit with your team, work through the exceptions, and turn decisions into explicit rules. We keep those rules in a separate layer for each customer, beside the systems they already run. Your team can read them and update them as the business changes.
We improve the shared product through deployment experience while keeping each customer's documents, data, and business rules private. Your documents, rules, and data are used only for your deployment and are not shared with other customers. We do not train models on customer data, and the model providers we use are contracted not to train on it either.
Watch how the balance moves
Both kinds of work are necessary. What tells you most about a vendor is how the balance moves over time.
When the product keeps getting more capable, each new deployment needs less building. Engineers spend their time on your rules, the part only your team can supply. You go live sooner, and product updates reach you whether you run in our cloud or on your own servers.
When most of each deployment is built fresh, you are paying for skilled custom work. That can be the right choice for a unique process. Plan and budget for it that way.
Three questions to ask an AI agent vendor
Ask these of anyone selling you agents, including us.
- What does the product already do? Which parts of my deployment are standard product capabilities, and which will be built for my business?
- Can we manage our own rules? Where do our business definitions and approval rules live, and can our team read and update them without your help?
- How is our information protected? How are our documents, business rules, and data isolated from other customers, and are they ever used to train models?
We work alongside customers because that is where the details become clear. The product should improve with every deployment, so the next customer takes less effort to serve.
If your import operations, trade documentation, or regulatory records still run on spreadsheets and email, we'd like to hear from you.
FAQ
Why do AI agent vendors send engineers with their software?
AI agents perform parts of a workflow, so they need to learn company-specific documents, exceptions, and approval rules. Vendors send engineers to work with the customer's team and turn that knowledge into rules the agent can follow.
What is IBM and Microsoft's Imagine Enterprise?
Imagine Enterprise is a platform announced by IBM and Microsoft on November 10, 2026, for building, running, and governing industry-specific AI agents. It combines IBM Consulting Advantage with Microsoft tools including Foundry, Copilot Studio, and Fabric. The initial industry frameworks cover financial services, insurance, and retail.
What do forward-deployed engineers do in AI agent deployments?
They work directly with customers to understand workflows, connect systems, define business rules, and put agents into daily use. IBM and Microsoft offer forward-deployed engineering support for qualifying Imagine Enterprise engagements.
How should I evaluate an AI agent vendor?
Ask what the product already does without custom work, where your business rules live and whether your team can maintain them, and whether your documents and data are used only for your deployment.
Which parts of a deployment are usually customer-specific?
Business definitions, cost allocation rules, approval processes, and exceptions are defined with each customer and belong to that customer. Document reading, cross-document checks, and tracing requirements to public regulations are standard product capabilities.
Does NURIE AI train models on customer data?
No. The external model providers we use are also contracted not to train on customer inputs. Customer documents, rules, and data are used only for that customer's deployment and are not shared with other customers. Product updates are delivered to cloud and on-premises customers alike.