Skip to content

The news behind the software that runs companies, explained for everyone.

The ERP LEADERS brief · Issue 02LinkedIn edition · 27 October 2026

SAP developer: AI writes tests in seconds, with about an hour of corrections

One SAP developer says AI proposes tests in seconds, but correcting them takes about an hour. This week: what that claim can establish, and a small trial your team can measure.

4-minute read · 5 reporting sources · A practical trial checklist

Exterior of an SAP Labs office building
SAP Labs office building · Photo: Saturnalia0 / Wikimedia Commons · CC0

The case: quick generation, human review

Paul Hardy, an ABAP developer at Heidelberg Materials, described his experience to SAPinsider. ABAP is a programming language used in SAP systems; unit tests are automatic checks on individual pieces of code.

He estimates that writing those tests manually doubles development time. AI proposes them in seconds, but he has found syntax errors and puts the correction work at about an hour. He still requires human review.

The limit: this is one developer's account, not a measured productivity study. It supplies no comparable sample, tool-by-tool results or evidence of how many real faults the tests caught. An hour of corrections is not the total cost of testing.

Read the interview in ERP Today, 21 October, first published by SAPinsider on 19 October.

Take this to your next programme review

Our suggested trial: choose one small, well-understood custom program in a test environment. Name a developer to review the tests and a business owner to approve the expected results. Agree the time budget before starting.

For example, if the program calculates invoice totals, write down the expected tax and rounding results independently of the existing code. Include ordinary inputs, boundary cases and an invalid input. A test that merely repeats today's calculation can preserve today's mistake.

Record three things:

  1. Coverage before the trial. Which agreed business rules already have automatic checks? List the rules still untested; avoid treating a code-coverage percentage as proof of correctness.
  2. Total human effort. Log time spent specifying, generating, correcting, reviewing and running the tests separately. Include setup time and any later maintenance. Compare with a documented manual baseline if one exists; otherwise say the saving is unknown.
  3. Fault detection. In the isolated test environment, use known faults or deliberately changed logic to check whether the tests fail for the right reason. Record missed faults as well as successful detections.

The decision: extend the trial only if the agreed rules are checked, a named reviewer accepts the tests, and the effort fits the budget set beforehand. Keep generated tests out of the live system until the normal approval process is complete.

At the review, ask who will maintain those tests after the next software change. A fast first draft is useful only if somebody owns what happens next.

The week in one screen

  • SAP's first €10 billion quarter. SAP reported €10.27 billion in quarterly sales; CFO Dominik Asam called the fourth quarter critical. Strong vendor sales do not establish the return from a customer's own project. SiliconANGLE, 21 October.
  • Leonhard Weiss replaces a retired SAP module. Its equipment operation covers about 150,000 machines, devices and tools. The replacement, smART, was built with cbs on SAP's Business Technology Platform and connected to S/4HANA. This is a change within its SAP landscape, not an exit from SAP. CIO, 20 October.
  • Gartner puts disposable applications on its 2027 list. These are applications deliberately retired when their purpose ends. For an ERP team, our question is who owns the data and access permissions after a temporary application closes. Gartner, 20 October.

Dispute watch: Microsoft's licensing case

The Court of Appeal refused Microsoft permission to appeal to the UK Supreme Court in the ValueLicensing dispute, The Register reported on 19 October. Microsoft could still apply directly to the Supreme Court. ValueLicensing's £270 million is a claim, not an award; Microsoft denies wrongdoing. Both companies declined to comment on the latest development.

Help choose the next case

Which ERP decision should we examine next? Comment on the LinkedIn article with the programme and the decision you want investigated. Please leave out confidential project details.

Weekly: one case examined, a practical review checklist and a short news roundup. Subscribe to The ERP LEADERS brief on LinkedIn for future editions.

Updated 27 October: clearer structure and source links; the Leonhard Weiss item now specifies that its replacement uses SAP BTP. The roundup has been shortened.

Discuss this case on LinkedIn