The hardest part of AI in your ERP is not the AI. It is knowing where it fits.

Most conversations about AI in manufacturing start with what the technology can do. That's the easy part, and it's getting easier every month. The question that actually stalls projects is a different one: where does it fit in how this business already runs?

Manufacturers asking about explainable AI in their ERP are rarely asking a technical question. They're asking a practical one. If something is going to sit between our people and our operating data, we need to see what it did and why before we let it near a live process.

Danie Grobbelaar, co-founder of the Intak Group, the group Lancea Konsult belongs to, puts the blocker plainly:

"One of the biggest problems in utilising AI is that it's actually quite difficult to understand where your use cases are going to fit in and how it's going to fit into your processes, especially if you don't even really understand what it is and how it's going to operate and interact in your business."

That's a chicken-and-egg problem. You can't design good AI use cases until you understand the technology in your own context, and you can't understand it in your own context until you've used it on your own data.

There are two ways to start using AI on your own SYSPRO data, and they're different kinds of thing. One is AGx, developed by Lotus Labs. The other is SYSPRO Torque, built by SYSPRO. We implement both.

What AGx actually is

AGx is a connector, not an application. It is an MCP gateway that plugs into the AI assistant a business is already using (preferably Claude Desktop for best results and ease of use) and gives that assistant a governed route to SYSPRO data.

There is no separate AGx interface to learn, no new AI platform to buy, and no data warehouse to stand up first. The conversation happens where the user already works, and the answers come out of SYSPRO.

The questions look like the ones managers already ask each other in corridors:

• Give me a quick overview of this customer.
• What products are they buying the most?
• Are there any unusual purchasing trends this month?

This is what conversational analytics means in an ERP setting: a question, asked directly, answered from the system of record, with no dashboard to build first and no report request waiting in a queue.

Key takeaway: AGx does not copy your data somewhere else to make it answerable. Questions are resolved against SYSPRO itself, through SYSPRO's own understanding of how your entities, business objects and processes relate to each other.

AGx can do more than answer. When it acts, it acts as the operator: it can post only what that operator is authorised to post, and SYSPRO enforces that exactly as it would for the person. Reading is different. In the plug-and-play setup, AGx can see all of your SYSPRO data, whoever is asking. Narrowing what it can read is possible, but it takes configuration as part of the implementation. Out of the box, the data boundary is set by who you give AGx to, rather than by a SYSPRO setting.

How we recommend using AGx. The immediate, plug-and-play value is in talking to your data: analysis, feedback, and a straight answer on where to focus next. It is the lowest-risk and fastest-paying thing AGx does, and where we would start with anyone who does not yet know what AI could do with their own data.
For execution, we recommend working in a Test or Dev environment. The operator's execution rights apply either way; Test is where you design and prove a process before it carries real transactions, the same discipline any ERP change already goes through.

Who AGx is for, and why that is a governance decision

We want to be clear about this. AGx is not a tool for everyone in the business, and we would not recommend rolling it out that way. Because AGx reads all SYSPRO data by default, choosing who uses it is the main way you control what it shows.

Danie's definition of the user:

"It's specifically focused around IT leads as well as top management within companies, people that are unrestricted in what they are allowed to have access to in terms of SYSPRO information."

"AGx is very much like a CEO assistant, or a very skilled process designer or consultant involved in designing processes within the company."

That points to three users:

• The senior decision maker, who needs answers from the business without waiting on someone else's availability to produce them.
• The IT lead, who needs to understand what an AI ERP capability actually does before committing the organisation to a direction.
• The senior system architect, who is designing the processes, and who gains a genuine laboratory: point AGx at a Test or Dev company and explore firsthand how AI behaves against real operational complexity, without a single live transaction at stake.

That third use is the one most easily missed. For anyone designing processes, understanding the capability is the work itself.

What the three have in common is that they are already allowed to see everything. AGx is not for anyone whose SYSPRO role limits what they can see. Unless it has been configured to, AGx would not apply those limits for them, and an AI that answers easily from data someone was never meant to see is a real risk, however helpful it feels.

What SYSPRO Torque is

Torque is part of SYSPRO AI, SYSPRO's own set of AI capabilities. SYSPRO describes it as an industrial AI platform for manufacturing: agents that watch what is happening in the operation, recommend a response, and, within permissions the customer defines, carry it out. Their shorthand for the distinction is "AI that advises and AI that acts".

What that gives you is clearer in what SYSPRO says ships today. Torque is in controlled release, and this release covers insight, recommendation and supervised execution, where a person approves what the agent proposes before it happens. Fully autonomous operation is listed as coming later. So a Torque implementation today is an agent that watches a process you have defined, tells you what it would do and why, and does it once someone says yes.

On SYSPRO, Torque runs through Sidekick, which becomes its agent surface once the two are connected, and it acts only through SYSPRO business objects, running under the operator's existing SYSPRO permissions. The Torque experience needs the SYSPRO Web UI on SYSPRO 8 2026 R1.

The use cases SYSPRO has put on the public record are the repetitive, rules-based kind, which is where this sort of automation earns its keep first: "sales order adherence, job status monitoring, supplier follow-ups, and eliminating 'swivel chair' integration work".

The two are not competing. AGx is a connection: your LLM, reaching your SYSPRO data, acting under your operator's execution rights. Torque is an execution platform: agents operating inside a defined workflow domain, running business processes. AGx serves a small number of people who already see everything. Torque is the route when AI has to work across the operation, inside processes that involve people with different levels of access.

Who Torque is for. The operations leader whose workflows are already defined and whose exceptions are already understood, where the question has moved from what could AI do here to which of these should run without us.

Danie's qualifier is the practitioner's half of the answer:

"But in order to do that, there's a whole design journey that you need to work through. And your people need to be educated and upskilled in what's possible. And you need to, almost firsthand, first experiment and understand certain angles on what AI is able to do, how it integrates, how it interacts, and all of that."

How we recommend using Torque. Start with one workflow where the exceptions are already well understood, and design it properly before it runs anything real. An agent that executes the wrong workflow flawlessly is a more expensive problem than no agent at all.

Explainable AI is the thread that runs through both

Put the two side by side and the same principle is doing the work in each, through different mechanisms.

In AGx, explainable AI comes from how close everything stays to a person. Every question and every action starts with a named user in a conversation, so nothing runs unseen. Answers come from the system of record, not from a copy of it. Anything posted goes through SYSPRO's own authorisation for that operator. And the reach of the data is a decision you make explicitly, by choosing who gets AGx and how it is configured.

In Torque, SYSPRO has made explainable AI an explicit design principle and given it a name. Glass House is SYSPRO's term for it: every recommendation carries the reasoning, the rules and the data behind it, and every action the agent takes is logged. You can see why it wanted to do something, and afterwards you can see that it did.

This matters, because AI agent governance decides whether a manufacturer can put an agent anywhere near a live process. Governance is knowing where the boundary sits before you switch anything on. For AGx, you draw it by choosing who uses it. For Torque, SYSPRO draws it into the platform.

What you actually need

Plenty of businesses need both, for different parts of the operation at the same time: the executive team asking questions of live data while one production workflow runs under agents.

Not knowing where to start is not a reason to wait. If nobody can yet say which process AI should touch, a senior executive asking questions of their own SYSPRO data is one of the quickest ways to find out, and AGx makes that close to plug-and-play. What no AI product shortens is the design work before an agent runs a process: describing that process the way it actually runs today, not the way it is documented.

Key takeaway: If your team can't yet name the workflows, start with a smaller project, aimed at understanding rather than execution. If your team can already name them, don't spend half a year proving what you already know.

Where Lancea sits in this

AGx was developed by Lotus Labs and is distributed by Lancea. Both are part of the Intak group. Lancea implements AGx, and SYSPRO Torque, which SYSPRO built, and we're actively piloting Torque projects with clients as one of a few partners globally chosen by SYSPRO. We have a focused unit dedicated to AI projects with customers.

We're equally invested in both, and that's deliberate. What matters to us is that you end up with the one that fits: the one closest to what your business actually needs, how your processes really run, and what makes you different from the company down the road. Sometimes that's AGx. Sometimes it's Torque. Often it's both.

Where to start

The next step isn't a product decision. It's a conversation about what your business needs.

Talk to the Lancea team about which one fits your environment, or explore AGx and SYSPRO Torque on their own terms.

Related reading

• Are you frustrated that AI can do everything except talk to your SYSPRO data?
• Ask your ERP a question in plain English: what conversational analytics looks like in practice
• How agentic AI is reshaping ERP in manufacturing
• AI agents in the enterprise: what it means for SYSPRO users in South Africa