ERP & Enterprise

ERP Roundup: Business Central 26 Goes Out of Support on Monday, and SAP Puts a Risk Tier on Every Agent

Sharif George6 min read

A quiet week for announcements, a loud one for deadlines. A Business Central version stops being serviced on Monday, SAP spent two days explaining how much autonomy it will sell you, and a billing default moved a month to the right. No named ERP failure, lawsuit or restated filing surfaced in the last seven days, so this is a three-story roundup rather than a padded five.

1. Business Central 26 goes out of support on Monday, and BC29 on-premises is now downloadable

Two things happened to the Business Central upgrade path this week, pointing in opposite directions. Yun Zhu's October cumulative update summary, published 5 October, lists CU00 for BC29 with on-premises download packages attached — the build that had not released when we wrote up BC29's GA last Friday. Alongside it, 28.6, 27.12 and 26.18 shipped.

26.18 is the part worth reading twice. Zhu's lifecycle table, updated 2 October, puts the end of support for version 26.x at 13 October 2026 — next Monday. Treat that precise date with care: Microsoft's own released-updates page for 2025 release wave 1 lists 26.18 as the October update and states no end date at all, and the trackers that do publish one mostly derive it from the standard eighteen-month rule. Two sources agree on the month; one gives the day.

Scope is where this gets misread. These are modern-policy dates governing on-premises servicing: after 26.18, a 26.x deployment stops receiving cumulative updates. An online tenant hits no support cliff at all — it is governed by the mechanic in Microsoft's update rollout timeline: a five-month update period you can reschedule inside, a one-month grace period where you cannot, then an enforced update period in which, in Microsoft's words, extensions that cause the update to fail "might be automatically uninstalled". Data is not deleted and a compatible version can be reinstalled afterwards. That is not a support deadline. It is a worse one: it arrives on Microsoft's schedule and removes your extensions to get there.

What it means for testing

Split the client list on Monday. On-premises 26.x clients need an upgrade scheduled, not a reminder: an unserviced version does not fail on a screen, it fails the first time a security fix is not there. Online clients need the opposite — a sandbox rehearsal of the enforced update, to find which extension breaks it before the enforced path answers that by uninstalling it in production. A UI-led regression pass finds neither. Scope it like the BC29 pass with no release plan behind it: from what the upgrade touches, not what the user sees.

2. SAP Connect: autonomy stops being a setting and becomes a test-design decision

SAP Connect ran this week, and SAP's own keynote write-up of 7 October is the primary worth reading rather than the conference coverage. Joule Work rolls out to all customers in October, the SAP Autonomous Suite covers finance, procurement, supply chain, HR and customer experience, and SAP counts over 200 Joule Agents since May against a target of 400 by year end. The productivity figures quoted — above 20% in HR, finance and procurement — are SAP's own and self-reported, so take the direction, not the number.

Most readers here will never deploy Joule. Read it for the governance vocabulary, which is the part the rest of the market copies. SAP named three oversight levels: human-in-the-loop approval, human-on-the-loop monitoring, human-in-command override. Chief Security Officer Marielle Ehrmann said every agent should carry a verified owner, a risk tier and scoped access rights, with tighter oversight the more autonomously it runs. The operative claim is that embedded agents run inside existing workflows, on the same data and rules, leaving the same audit trail.

What it means for testing

An autonomy level is a test condition, not a configuration note. The same process at human-in-the-loop and at human-on-the-loop is two acceptance scenarios with two sign-offs, and the second has no approval step for a tester to observe — so the only evidence it worked is the audit trail. Test the trail as a deliverable: can a business owner, six months on, see which agent acted, under whose ownership, on what data, and what a human could have overridden? That is what a sign-off has to record, asked of a participant who never attends the UAT session.

3. Copilot Business default-on billing slips to 1 December, with a 4,000-credit cap nobody chose

Microsoft's October Partner Center announcements moved the usage-based billing change for new CSP purchases of Microsoft 365 Copilot Business from 2 November to 1 December 2026. Pay-as-you-go is on by default, covering Copilot Cowork, Work IQ APIs and GitHub Copilot Harness, with a default spending limit of 4,000 Copilot Credits per user per month that an admin can adjust. Sandbox environments open on 2 November, and eleven markets are excluded at launch.

A date moving a month later is not, on its own, a story. The shape of it is: a default that is on, a limit set for you, and a meter attached to products the buyer did not separately purchase. That turns up on an invoice, not in a test result. The sandbox date is the useful one — four weeks to see the billing behaviour before a client does.

What it means for testing

Commercial defaults belong on the pre-go-live checklist with an owner against them, beside licence keys and tenant settings, because no UAT script has ever caught one. Write the check as a question somebody answers in writing — what is this tenant's spending limit, who set it, on what date — not a line saying "review licensing". It is the pattern we keep hitting: a prerequisite correct on the day you look, wrong by the time anyone depends on it.

The through-line: three changes, three dates, and nothing on screen to test

A version that stops being serviced. An agent whose autonomy level is chosen rather than observed. A billing default that switches on by calendar. None of the three alters a field, a page or a posting routine, which is why a feature-led pass walks past all of them — the same blind spot as the SOAP endpoints BC29 removed behind the UI. Each is governed by a date rather than a behaviour, and a date is the one thing a test case cannot assert. Monday's post made the adjacent point about tooling: a record-and-replay script proves the system still does what it did, not that anybody decided it should. Dated risk needs a named owner, a recorded check and a date on the record — which is all a Business Central UAT pass ever really produced.

→BC26 stops being serviced on or around 13 October; 26.18 is its last cumulative update. On-premises clients need an upgrade scheduled; online clients need an enforced-update rehearsal.
→SAP named three oversight levels and gave every agent an owner and a risk tier. Treat an autonomy level as a test condition; test the audit trail as the deliverable.
→Copilot Business default-on billing moves to 1 December, with a 4,000-credit default cap and a sandbox from 2 November. Put the spending limit on the pre-go-live checklist, with a name and a date.

Free ERP UAT checklist (Excel)

A structured workbook covering the full UAT cycle: pre-UAT preparation tasks, test execution tracking, an issue log with severity guide, and post-UAT wrap-up. Built from real ERP implementations.

No spam. We will occasionally send UAT and ERP implementation resources. Unsubscribe any time.

Sources and further reading

Free Plan Available

A Deadline Is Not A Test Case

LogicHive gives ERP implementation teams the acceptance layer that automated tests do not cover: structured test cases, named business owners, linked defect tracking, and sign-off built from real execution data rather than a screenshot of a passing build. When a version falls out of support or a default switches on by calendar, running a pass is a scheduling decision rather than a project. The free plan runs a real project end to end.

Written by Sharif George

LogicHive Frontman

Sharif is the frontman for LogicHive. He writes the ERP news roundups and the Business Central release coverage, following each Microsoft, SAP and Oracle change through to what it means for regression scope and acceptance testing on live implementation projects.

View all articles by Sharif George
  • ERP & Enterprise

    Business Central's Record-and-Replay Test Tool Reached GA This Week. It Is Not UAT.

    Microsoft moved the Business Central page scripting tool from preview to general availability with version 29, thirty months after the preview opened, adding multi-row grid selection and dialog-text validation — neither of which the reference documentation describes yet. What a recording actually proves, what it is documented not to reach, and why a replay pipeline is a regression asset rather than an acceptance one.

    5 October 2026 · 6 min read

  • ERP & Enterprise

    ERP Roundup: BC29 Arrives With Two Testing Features Its Own Docs Do Not Describe, and a Licence Key Where the Download Date Decides

    Business Central 29 went generally available this week with two AL testing capabilities Microsoft lists as launched — and nothing in the reference documentation showing how to use either. Alongside it, a licence key change that reads like simplification and behaves like a dated certificate, and a survey where 46% shipped AI code that failed in production and 69% of them stayed confident. Three ERP stories and what each changes about testing.

    2 October 2026 · 6 min read

  • ERP & Enterprise

    A Go-Live Decision Taken in 2022 Is Still Setting the Audit Opinion in 2028

    Europe's largest council finally has a working Oracle finance system, with no outstanding priority 1 or 2 tickets — and an auditor who does not expect to sign off a set of its accounts until 2028/29. The gap between those two facts traces back to which modules went live untested in 2022, and to what UAT sign-off is actually supposed to produce.

    28 September 2026 · 6 min read