The MVP Summit left me chewing on one question: when does the “ChatGPT moment” hit Dynamics ERP?
ChatGPT wasn’t a faster search engine. It changed what users expected from their computer. They stopped formulating queries and started stating intent. The approach changed from “how do I get the system to give me what I want” to “what do I actually want.”
Dynamics ERP is approaching the same inflection point. When ERP stops being a system you operate and becomes a system you direct.
Here is what I think that actually looks like, in three stages.

1. Intent replaces navigation

Running ERP today is navigation. You click into finance, then procurement, then warehousing. You complete forms. You post transactions. The skill is knowing where to go and what to fill in. Companies hire and certify around that skill, and it’s real, but it’s also the part of the job that is slowly being deprecated.
The tipping point is when users do not need to care which module they’re in.
  • “List POs that will miss promised dates but keep margin whole.”
  • “Why did operating margin dip last week, and what should I do about it?”
None of those map cleanly to a single form or module. They’re cutting across and providing answers that today may require a Teams meeting with a planner, a buyer, and a finance lead.
Forms are optimized for transactions. Agents are optimized for outcomes. That’s a different kind of software, and it changes what “knowing your ERP” means.
I think of this as being a programmer who is unaware of assembly and the complexity of managing registers, directives, system calls.  Today most developers lean on modern programming languages with compilers which optimze for the compute architecture on which the program will eventually run.  The complexity of memory management is still handled, but no longer slowing down developers.  Part of this was enabled by the increased compute capabilities, like AI is being enabled similarly.

2. The agent becomes the operator of record

The next stage is bigger. AI moves from assistant, to automation, to employee. The unit of work shifts from “task” to “objective.” Agents own the work unless escalated.
A few examples of what that means in practice:
  • An agent creates, adjusts, and confirms planned orders nightly with little human involvement.
  • Another manages vendor commits and only interrupts buyers when constraints are unsolvable.
  • Finance close tasks are executed autonomously and reviewed afterward, not performed step by step.
Humans shift their focus:
  • Exception handling
  • Policy definition
  • Economic tradeoff decisions
  • Market approach
That’s a different job description than “operate the ERP.” It’s closer to managing a team that happens to be made of agents. The org chart eventually catches up. So do the metrics, because performance stops being measured in tickets closed and starts being measured in business outcomes the agent was directed toward.
In addition to technology and compute enablement, this is going to take a shift of human trust and habits, hence will take time.

3. ERP reasons across modules natively

This is the part bolt-on automation cannot fake. Many of the most valuable questions a business asks have always required cross-functional meetings, spreadsheets, and trust to answer:
  • “If we accept this rush order, what downstream costs will surface in 60 days?”
  • “Which customers are unprofitable because of warehouse behavior, not pricing?”
  • “What’s the least disruptive way to free $5M in cash this quarter?”
Each one needs multiple key leaders Finance, Supply Chain, Operations, and Policy reasoned about together.  At a strategy level, not just analysis, automation, or taking an action.
Classic ERP integrated the data across those modules. Agents integrate the meaning. When the system can hold all of that in view and produce a defensible recommendation, the meeting culture starts to shrink. So does the spreadsheet sprawl that sits underneath it. That’s not a feature release. It’s a different operating model.

What quietly makes the moment possible

This shift isn’t going to happen on its own. It only works when a few things are already in place:
  • Process is standardized. Agents tolerate ambiguity poorly.
  • Policy is explicit. Agents need boundaries written down, not lived in tribal knowledge.
  • Data quality is boringly reliable. The kind nobody talks about, because nothing is broken.
  • Governance moves from approval to authorization. Humans set what an agent is allowed to decide, then get out of the way.
Skip any of those and the system will appear unreliable and untrustworthy.

What this means for the next 18 months

The ChatGPT moment for Dynamics ERP will not arrive immediately. It will build, and will match human comfort in AI. Each Dynamics 365 Finance & Supply Chain Management release adds another piece. Copilot Studio, agent-building primitives, planning copilots, finance copilots, the deeper data fabric underneath.  Random acts of AI and the future of work.
Eventually the use of AI becomes mature and forms become the familiar and comfortable fallback, not the default.
The companies that benefit first are the ones that have already done the unglamorous work. Clean masters. Consistent processes. Policy and processes documented properly. Data lineage they can defend. None of that is new advice. It’s just more important.
The ChatGPT moment for Dynamics ERP is when it becomes a system that reasons about the business, rather than one that records what humans already decided.
Forms don’t reason. Workflows don’t reason. Agents do.