On 25 August 2026, Microsoft answered a question in its own FAQ with one word. Asked whether there would be a September 2026 release wave 2 announcement, the answer was “No.” Not delayed, not merged, not renamed. Cancelled as a concept.
Two weeks later the community had decided what that meant: updates are now unpredictable, regression testing is impossible, and every partner should panic. Read the actual announcement and the opposite is true. Microsoft changed how it talks about what is coming. It did not change how it ships. Those are different problems, and only one of them affects your integration.
For anyone who has built an interface into Business Central, the change matters for a reason nobody in the announcement names directly: the release plan was never really a marketing document for us. It was the scope document for regression testing. Twice a year it told you what to retest. That document is gone, and something has to replace it.
What Microsoft retired, and what it did not
The announcement is short and unusually clear on this point. Starting September 2026, roadmap content for Dynamics 365, Power Platform and Dataverse moved to the AI at Work roadmap, the renamed Microsoft 365 roadmap. The twice-yearly wave model was replaced by continuous publishing: capabilities appear when they are committed, and then move through three states – In Development, Rolling Out, Launched.
The transition ran on a tight calendar. Existing roadmap items with a public preview or general availability date of 1 June 2026 or later migrated across. Release Plans on Microsoft Learn were archived. Release Planner retired by 15 November 2026. Partners can pull roadmap data through a Release Communications MCP server, with no cost and no sign-in.
What the announcement explicitly leaves alone is the part that matters to production systems: “This transition changes how Microsoft communicates upcoming innovation. It doesn’t change how our products are built, released, or deployed.” Products with established release schedules continue to follow them. Message Center remains the source for tenant-relevant change notifications. Microsoft Learn remains the documentation home.
You can read the full post on the Dynamics 365 blog, and the partner-facing summary in the Partner Center announcements for September 2026.
The distinction that resolves most of the confusion: a release wave was two things at once, a communication package and a release boundary. Microsoft removed the communication package. The boundary survived, because Business Central still has versions, and versions still arrive on a schedule.
The product clock never stopped
Business Central is not a continuously deployed web app where features simply appear. It runs a documented cadence, and that cadence is unchanged. From Microsoft’s update cycle documentation:
| Event | When | What it means for an integration |
|---|---|---|
| Major version | April and October | Schema, API and platform changes. This is your real retest trigger. |
| Preview window | March and September, one month before | Sandbox on the new version while production is still on the old one. |
| Update period | Five months from GA | You choose the date inside the window, not whether. |
| Grace period | One month after the window | No more postponing. Incompatible extensions block the update. |
| Minor update | Monthly, except April and October | Regulatory and critical fixes. Lower risk, not zero. |
Version 29.0 landed in October 2026 and Microsoft’s own release notes still call it “2026 release wave 2”. The naming persists inside the product even though the roadmap stopped using it. Version 29.1 followed in November. If you are on 28.x and your partner has not started planning the move inside the five-month window, the wave retirement is not your problem. The calendar is.
This is worth stating plainly because it is the most common misreading I have seen since August: the wave is gone, so the release is gone. The release is not gone. What is gone is the document that arrived two months early and told you what was inside it.
Why the wave plan mattered more to integrations than to anyone else
A finance team can absorb a new field on a journal line. An integration cannot absorb it silently, because an integration is a contract with assumptions baked in: this field is required, this value is an enum with five members, this endpoint returns these columns, this event fires once per transaction.
Under the wave model, the release plan gave you a two-month head start on three questions. Which fields are changing? Which APIs are affected? Which of my mappings assumes the old shape? That is not browsing. That is scoping, and it decided whether the regression window was two days or two weeks.
There is a second loss that nobody has written about yet. The wave plan was also the communication artifact. “Here is what is coming in October” is a sentence a partner can put in front of a client. “Things will appear on a roadmap and roll out on their own schedule” is not. Client change management just got harder, and that work lands on whoever owns the interface.
Continuous disclosure does not remove the information. It changes its shape: many small entries instead of one bundle, published when committed rather than on a calendar, with statuses that update as delivery progresses. You now read a stream, not a document. If you were relying on a person remembering to check a roadmap twice a year, that person is now the single point of failure.
What replaces the wave plan in practice
Four signals do the job, and they are all free.
- The AI at Work roadmap, filtered. You can filter by product, save the view, export to CSV and follow items by feature ID. The practical move is a saved filter for Business Central plus the specific areas your interface touches, reviewed on a fixed day each month.
- Message Center. The announcement is explicit that Message Center remains the source for tenant-relevant change notifications. This is the signal that actually applies to your environment, as opposed to the product generally.
- Preview sandbox. A month before each major version, the preview is available. This is the only way to test against the real thing rather than a roadmap entry. If you are not using the preview month, you are testing in production, just later.
- Telemetry. Update availability, failures and extension incompatibilities emit signals. If nobody is watching them, the first person to notice an update problem will be a user.
None of this is exotic. It is the difference between a calendar reminder and a process.
The regression pack you should have had anyway
Here is the uncomfortable part. If the loss of the wave plan breaks your testing, your testing was already fragile. The wave was doing work that automation should have been doing. Microsoft shipped a tool in the same cycle that points at the answer: page scripting reached general availability, a generative AI capability that creates and replays client-side user acceptance tests without AL code, documented in Microsoft Learn.
The regression pack we build on every integration project looks like this, and it is not new. It is just now mandatory.
| 1 |
A written interface contract Every field, its type, whether it is required, who owns it, and what happens when it is rejected. One page that both sides sign. Without it, “did the release break us” is unanswerable. |
| 2 |
A golden set of transactions Twenty to forty real payloads, including the ugly ones: a unit conversion, a partial delivery, a replayed batch, a record with a missing optional field. Replayed on every update. |
| 3 |
An idempotency check Send the same transaction twice and prove nothing duplicates. Most integration incidents after an update are not API failures, they are retries that created a second record. |
| 4 |
A replay path A documented way to reprocess a failed transaction without manual SQL. If a bad update produces a dead-letter queue nobody can drain, the regression test passed and the business still stopped. |
With that pack in place, a release is a scheduled event you handle. Without it, a release is an outage you discover. The wave plan was never a substitute for the pack. It was a reminder that the pack was due.
Four clauses for the support agreement
This is where most integration contracts are silent, and where the cost shows up eighteen months later. If you are renegotiating support, these are the four lines worth adding.
- Version currency. Name the versions you support and how far behind a customer may fall. A five-month update period plus a one-month grace period means “we will get to it” has a deadline whether the contract says so or not.
- Regression scope. State what is retested on a major version and who pays. Otherwise the honest answer is “whatever we had time for”.
- Change notification. Say who watches the roadmap and Message Center, and how fast the customer hears about something that affects them. A named role, not a team.
- Third-party breakage. Marketplace apps and per-tenant extensions can block an update, and Microsoft can remove an incompatible app from Marketplace. Decide in advance who is responsible when the blocker is not your code.
What this looked like on a live integration
We built the billing integration behind Baltic Ground Services’ aviation ERP, where fuelling, ground handling and de-icing all bill off the same aircraft movement into Business Central. Five modules, one flow, and a hard requirement that every charge is auditable when an airline disputes it.
That project could not run on a wave calendar, because the operational data never stops and the billing cycle does not pause for a release. What made it survivable was not a release plan. It was an interface contract, a replay path for rejected postings, and the discipline of testing against the preview environment before the update window opened.
The same principle applies to the less glamorous work: the tyre tracking system that reconciles 35,000 serials across axle and vehicle, or the field app integration we wrote about in what breaks when the ERP and the field app disagree. In each case the risk was never the release itself. It was a key, a unit, or a retry that nobody had written down.
Where this connects to the rest of your stack
One thing worth saying, because it is where integration work and computer vision work meet. The same interface that carries a work order out to a technician also carries an inspection result back. A vision system that flags a defect at 27 times manual speed is only useful if that flag reaches the ERP as a record someone can act on, with the right asset, the right batch and the right timestamp attached.
That means the release question is not only an IT question. When a platform update changes a field shape, it can quietly break the path that carries quality data. The team that built the vision system should know how its output is keyed before anyone signs the regression scope.
Questions we get asked
If your integration was built around the wave calendar
The wave plan is gone and nothing replaced it automatically. That leaves a gap between what Microsoft publishes and what your interface assumes, and the gap is where incidents live.
If you are running a Business Central integration and cannot answer “what changed in the last major version and did we retest it”, that is worth an hour of someone’s time before the next update window opens.
Get a clear path forward.
Share your challenge and walk away with concrete next steps - whether you're exploring AI, automating a process, or scoping a custom software project.
- Honest assessment of your situation - no sales pitch
- Actionable next steps you can use right away
- AI, automation, and custom software expertise