Think about the last time you watched someone copy a number from one screen and type it into another. A purchase order total keyed into a spreadsheet, a customer email retyped as a sales quote, or an approval chased in Teams because nobody opens the workflow. I would guess it was not very long ago.
At Directions EMEA 2026 in Paris, Microsoft gave that kind of work a name: people as middleware. The ERP, the CRM and the other apps each do their own job, and a person sits in the middle holding them together by hand. Microsoft wants to turn those people into managers instead, where agents do the joining and people decide what should happen.
The ERP becomes a service
To make that work, Microsoft is changing what Business Central is. People will work in Copilot and whatever comes after it, with Business Central behind all of it as the trusted layer. Microsoft described Dynamics 365 as "headless business services on MCP", which means the ERP is no longer only an app a person opens. It becomes a set of services an agent can call, with the same posting rules, approvals and permissions applied every time.
That changes where a partner's knowledge matters most. For years, a lot of Business Central expertise has been knowledge of the screens, like where a setup lives or which field the user always forgets. When agents do the clicking, what matters is the rules behind the screens: what may be posted, by whom, and after which approval. Those rules were always the hard part of an implementation, but an experienced user could cover a gap by simply knowing not to post something yet. An agent will not know that, so the rules have to be written down and set up properly.
Agents are not integration
There is one place where I think this idea gets stretched too far, and it was the subject of my own session in Paris. If agents can call the ERP, it is tempting to let an agent be the integration as well, and I think that is a mistake.
An agent is built for judgment, and it may make a slightly different call next time. An integration needs guarantees: a webstore order has to arrive exactly once, and when the warehouse system is down at two in the morning, the message has to wait and try again. The patterns behind those guarantees, like logging every message, catching duplicates and retrying safely, are not new. I have been building integrations around those ideas for ten or twelve years, and AI has not made a single one of them less important.

At Equerra those patterns live in Elevate Connect inside Business Central, and in Elevate Integrations, which runs in the customer's own Azure subscription and connects any system, not only Business Central. The agents sit on top, reading the message log, explaining why something failed overnight and suggesting a fix for a person to approve. That is the shift from middleware to manager that Microsoft was describing.
Where partners fit
Microsoft backed its argument with McKinsey's State of AI in 2026 report. Nine in ten organisations now use AI somewhere, but only about 6 percent say it has added real profit, at least 5 percent of EBIT. What sets that group apart is that they changed how the work is done, rather than adding AI to the old process.

That is where partners come in, and it is also where the pressure is. Customers can now build simple agents and apps for themselves, so some of the work we used to do for them is moving to them. What they cannot easily do on their own is step back and decide how their business should run now. Microsoft said the one thing it will never ship is a partner's judgment. I take that as a compliment, and also as a warning, because a partner who mostly sells work the customer can now do alone has a shrinking business ahead of it.
What it looks like from where I sit
A lot of what Microsoft described is already how we work at Equerra. We have started a headless ERP journey with customers, where Business Central stays the system of record and the portals and agents people use are built on our Frontier platform. The pages, the APIs and the agents all share one permission model, so an agent can never do more than the person it is working for.
Frontier is also where our customers can build for themselves. It lets them create apps and connect their systems around the way their business actually works, instead of bending the business backwards to fit the system. It gives them a base layer for agentic development that goes well beyond permissions. A development harness holds every AI-built change to the same standard and defines what done means, a governance layer makes every agent act as the user, ask before it changes anything and leave one audit record, and a model router sends each step to a cheaper model once testing shows it does the job just as well, so simple work does not pay top prices. That makes it a great place to start, and it takes away much of the risk that comes with building with AI.

Copying numbers from one screen to another was never a good use of anyone's time. If this works the way Microsoft described, people stop holding the systems together and start deciding what the systems should do. That is a much better job, and I am looking forward to helping our customers get there.


