The Future of ERP: Licensing, Build vs Buy, and Agentic-first Architecture

AI agents are becoming increasingly capable, and the cost of building new systems is dropping significantly. We are hearing more and more bold headlines like that recently — but how does this reflect on ERP? How might it impact implementations, usage, and architecture, and what does it mean in our context? Let me share a few of my thoughts as an experienced Solution Architect with a deep technical background who has personally adopted some of these tools recently.

Licensing: Enterprise software has traditionally been licensed per seat — any user who touches the system needs a licence or pays a subscription fee. However, as AI agents become increasingly capable of performing the routine tasks that end users typically carry out in the system (such as creating and posting journals, settling invoices, and so on), per-seat licensing is becoming an inefficient model for major vendors.

Here is a Deloitte article that predicts that by 2030, at least 40% of enterprise SaaS spend will shift toward usage-, agent-, or outcome-based pricing. But the transition won't be overnight — the seat-based model is evolving rather than disappearing (source).

Clearly Microsoft is looking into it as well - recently announced license simplifications, improved monitoring tools and bundled per-seat licences with AI (Copilot) (source).

However, it's a gradual shift and for the next year or two it will most likely be hybrid models, combining seat-based subscriptions with usage.

Buy vs. Build: Why buy expensive licenses when you can build something in-house, tailored to your specific needs, relatively quickly and at a reasonable cost (and that's not necessarily the ERP itself, but any ISVs or extensions)? For big corporates — who remain the primary ERP users and implementers — it still makes more sense to buy. When a CFO needs an urgent analytics report for a board meeting and the system is unresponsive, there is someone on the other end who is accountable: with SLAs, with legal liability, and with commitment to resolve the issue. Those SLAs, vendor relationships, and legal frameworks remain critically important for large organisations — with or without AI, and arguably even more so with it.

Of course there is more to it than just support, things like compliance and regulatory requirements for many industries (financial services, pharmaceutical, manufacturing), as well as established ecosystem effects and total cost of ownership.

At this moment IT leaders in large organisations are practically approaching this as follows - "buy what scales and build what sets you apart". And this means that most of them will continue buying ERP from major vendors for the reasons stated above. However, the nature of ERP projects is shifting from monolithic implementations toward composable architectures where vendor platforms serve as the foundation, augmented by custom-built AI agents and integrations at the edges.

Implementations: So big organisations will still need ERP systems and implementation projects and the level of automation within organisations will likely go even further (more software will be built along the way, because it's more feasible, and so why leave small things out of scope or defer it as "technical debt").

Now, looking at a typical ERP implementation story — fit/gap analysis, configuration, data migration, extensions and integrations — in the context of AI automation and coding agents what will remain the main bottleneck?

I would argue it is knowing what to build, the ability to architect it in a wider landscape of systems an organization may already have, and the capacity to specify it in sufficient detail for coding agents to execute. This is not something organisations will be able to do themselves, and it's not exactly a traditional functional consultant's job, which is largely used to be focused on how things are done within a specific ERP, from a user-experience perspective rather than an agent-first one.

Moreover, AI assistants are already able to automate roughly up to 40% of routine ERP consulting tasks, including generating project documents, mapping data fields, and doing repetitive configurations (source).

So the consultant role will be shifting from translating business needs into system configurations to defining how human-agent hybrid workflows should operate. As AI can hallucinate or behave unpredictably, the new consultant must be able to help define failure modes, acceptance criteria, safety checks, etc.

The future ERP implementer is not simply a traditional consultant who can also write prompts — they will be a new professional archetype combining deep domain expertise, specification discipline, AI governance capability, and orchestration thinking. The critical insight for D365 F&O specifically is that functional and technical knowledge doesn't become less valuable — it becomes the essential input that determines whether AI agents produce reliable business outcomes or expensive errors.

With all that said — then what future hiring will look like? I think this is something we cannot yet properly define. ​

*** Here is a summary of supporting research materials for this article.