This demo shows Claude operating SAP Business One itself — navigating, querying, and working the system end-to-end with no human at the keyboard. It’s a demonstration of where AI-plus-ERP is heading: from answering questions about the system to operating it under instruction.
Most of what TEKAI ships today is the first half of that sentence — ask, automate, predict, deliver. This video is a look at the second half. For the difference between an agent that acts on live data and a chatbot that returns a script, see our guide to AI in SAP Business One.
On a demonstration system, Claude received plain-language instructions and carried them out in SAP Business One — completing the workflow a trained user would perform, unattended. No macro was recorded, no script was written for the run; the model read the screen, understood the task, and worked the application the way a person does.
That framing matters more than any single step in the video. The demonstration isn’t “an integration was built for this workflow” — it’s “the model can operate the workflow because it understands what the workflow is for”. The same capability that completes this sequence generalises to the next one without new development.
Classic RPA breaks when a screen changes; scripted macros do one path and only that path. Every SAP Business One upgrade, form change or added field is a maintenance ticket for a script-based bot — which is why RPA estates tend to grow a permanent support team of their own.
A model that understands the task, not the pixels, adapts the way a person does. Move a field, rename a button, add a confirmation dialog, and it re-reads the screen and continues, because it was never following coordinates in the first place — it was following intent. That is why this class of automation survives the updates that kill script-based bots, and why it changes the economics of automating long-tail workflows that were never worth a scripting project.
Capability and permission are separate questions, and the distinction is the whole governance story. What ran here ran on a demonstration system. In production, the same principles that govern all TEKAI work apply: scoped authorisation — the agent can touch only what its user could touch; human approval on consequential actions — posting, paying, committing; and your decision, workflow by workflow, about which sequences are automated at all.
An agent that can run SAP Business One unattended is not an argument that it should run yours unattended. It’s an argument that the choice is now yours to make, with the controls to make it safely (/tekai/security/).
The end-state this previews is specific: routine transaction chains delegated to an agent, with humans setting the rules up front and reviewing the exceptions afterwards — the supervisory shape most operations teams already use for junior staff, applied to software.
The practical entry point isn’t unattended operation — it’s the TEKAI patterns already live: ask questions of live data, automate document drafting, predict what’s coming, deliver reports where people read them. Those patterns carry the same architecture shown here, with humans firmly in the loop.
The robot demo shows the ceiling; the Fit Test scores your floor — five minutes to see which of the shipping patterns fit your operation today. The value of the video is not the spectacle: it’s that every repetitive sequence your team runs daily has just become a candidate for delegation, on your terms, when you decide it’s ready.
The demo shows Claude’s computer-use capability on a demonstration system; TEKAI’s shipping patterns are the query, automation, prediction and delivery workflows.
Technically demonstrable; operationally, approval gates are yours to set — and we recommend them on anything consequential.
It follows intent, not screen coordinates — no scripts to maintain, and no bot that breaks when a screen changes.
Same answer as all TEKAI work: scoped permissions, human approval where it matters, nothing trained on your data — architecture at /tekai/security/.
Something reversible and measurable — reconciliation drafting and report delivery are the proven first steps.
New to the category? Start with the overview: AI for SAP Business One, explained.