Your ERP Doesn't Need AI. It Needs Relief.

Satya Nadella spent $80 billion on data centers last year so you could summarize an email. Jensen Huang wore a leather jacket on stage and convinced the world that inference is the new oil. And somewhere in a conference room in Walldorf, Germany, a product manager at SAP just added "AI-powered" to a dropdown menu that was already there and called it innovation.

We are in the "pets.com" phase of artificial intelligence. Everyone has a strategy. Nobody has a business model. Venture capital is flowing into companies whose entire product is a wrapper around an API that OpenAI could deprecate on a Tuesday. The median enterprise "AI transformation" consists of a chatbot that hallucinates about your own product documentation and a Copilot license nobody asked for.

Meanwhile — and this is the part that should terrify every CTO who just approved a seven-figure AI budget — the vast majority of enterprise workflows are running the same way they did in 2019. Same ERP. Same batch jobs. Same config spreadsheets passed between teams on SharePoint. The only thing that changed is the press release.

Here's the uncomfortable truth about artificial intelligence in the enterprise: it doesn't matter how smart your model is if the system it talks to was designed in the 1990s to resist change at the molecular level.

That system is called ERP. It runs the world. And it doesn't care about your agent framework.

ERP: The $500 Billion Trap Everyone Pretends to Love

Let's talk about the elephant. Not the one in the room — the one that is the room.

61% of SAP ECC customers have yet to move to S/4HANA. More than a decade after release. A decade. That's not a migration timeline, that's a geological epoch. Only 8% complete migrations on schedule — and "on schedule" in SAP-land already means "two years and three budget revisions from now."

This is not a technology problem. This is an organizational hostage situation. ERP exerts gravitational force. Every process, every price, every partner function, every order workflow — it all gets sucked into the core. And once it's in there, extracting it requires the organizational equivalent of open-heart surgery on a patient who's running a marathon while filling out a change request form.

The AI industry looked at this and said, with a straight face: "Let's put a language model on top of it."

Right. And I'm going to put a spoiler on my minivan and enter Le Mans.

You cannot make a system designed for stability behave like a system designed for agility by adding a chat interface. You cannot make a 30-year-old monolith "agentic" by pointing Claude at an RFC call and hoping for the best. You cannot fix architecture with vibes.

The problem was never intelligence. The problem is that execution is trapped inside systems that weren't designed for speed or change. And no amount of prompt engineering fixes a bad architecture.

Make ERP Great Again

Here's the thing that drives me slightly insane about the AI-versus-ERP discourse: ERP is not the villain. ERP is the backbone. It runs payroll. It manages inventory. It keeps Fortune 500 companies from accidentally shipping 40,000 units of the wrong product to the wrong continent. ERP is the most boring, most essential, most quietly competent category of software ever built.

The problem isn't that ERP exists. The problem is that we've spent twenty years stuffing every revenue motion, every pricing model, every partner function, every order workflow, every subscription logic tree into a system whose core job is stability. We took the world's best filing cabinet and asked it to also be a race car. Then we got mad when it couldn't turn.

That's not SAP's fault. That's ours.

There is a company in New Jersey called viax that understood this — quietly, without a TED talk — years before the current AI mania. Their thesis is unglamorous, architecturally rigorous, and correct: you don't replace the ERP. You don't fight the ERP. You don't slap a chatbot on the ERP and call it transformation. You liberate the ERP from work it was never designed to do well.

viax builds a revenue execution layer that operates alongside ERP. The volatile stuff — pricing determination, product configuration, order orchestration, subscription management, partner functions — gets modeled, tested, and deployed in viax. The stable stuff — financials, inventory, master data — stays in ERP where it belongs.

ERP does what ERP does best: be the system of record, be stable, be auditable, be the thing that makes your CFO sleep at night. viax handles the part that needs to move at the speed the business actually makes decisions. Which, for those keeping score, is considerably faster than "next quarter's release train."

The result isn't ERP replacement. It's ERP relief. Your S/4HANA migration just got simpler because you extracted the hardest, most volatile logic into a layer purpose-built for it. Your ERP team stops being the bottleneck for every revenue change. Your business team stops writing passive-aggressive emails about why a pricing update takes eleven weeks.

This is not disruption. This is the architectural equivalent of telling your ERP: "You're great at your job. Stop doing everyone else's job too."

viax doesn't compete with ERP. It makes ERP viable again by removing the burden ERP was never designed to carry. That's not a pitch. That's structural engineering.

The Architecture That Agents Actually Need (And Nobody Else Built)

Now. Here's where it gets good. Because viax didn't just build a revenue execution layer. They built — and I genuinely believe this was only partially intentional — the ideal operating environment for autonomous AI agents.

Let me explain why this matters by explaining why everyone else's "AI agent" is a joke.

The reason enterprise AI agents are mostly theater is that enterprise systems are hostile terrain. An ERP isn't a clean REST API with Swagger docs and a sandbox. It's a 30-year-old organism with custom fields nobody documented, business rules encoded in ABAP routines that one developer in Walldorf wrote during the 2007 Champions League final, configuration nested seven layers deep, and an implicit understanding that "turn-key implementation" means 18 months and $40 million if you're lucky.

Point an AI agent at that. Go ahead. Watch it hallucinate a pricing calculation and post it to production. I'll wait.

Now look at what viax actually built. A configurable execution engine with a strongly-typed GraphQL API. A hierarchical determination engine — think SAP's condition technique, except rebuilt for the modern stack so it's actually introspectable. A constraints engine for product configuration that doesn't require a PhD in Variant Configuration to understand. A state machine for every business interaction — orders, quotes, subscriptions, invoices — with well-defined states, hookable transitions, and traceable logic paths.

Every business process is API-accessible. Not "we exposed an endpoint" accessible — natively programmatic. Every state transition is observable. Every pricing determination can be queried, replayed, and explained. Every configuration constraint can be validated without executing a full order cycle.

The system was designed to be operated by machines before "agentic AI" was a phrase that appeared in a single pitch deck.

You know what an AI agent can do with a system like that? Its actual job. It can draft a complete tenant configuration from a requirements brief — determination models, state transitions, partner functions, constraints — and hand a human a reviewable output instead of a blank canvas. It can validate that every customer's unique pricing logic survives a platform version bump before deployment, not after production goes sideways at 2 AM. It can trace a determination hierarchy and tell you exactly where pricing diverged from expectation, because the platform was built to make that path visible.

It can do these things because the architecture allows it. Not because someone fine-tuned a model. Not because someone wrote a really good system prompt. Because the substrate was engineered for programmatic operation from day one.

viax didn't bolt AI onto a legacy architecture. They built an architecture that agents can natively operate. Everyone else is writing prompt wrappers around systems that actively resist automation. The architecture is the moat.

The Flywheel (a.k.a. The Part Where It Becomes a Business)

Here's the business model that most AI companies can't articulate because they don't have one. They have "usage." They have "seats." They have "we'll figure out monetization after Series C." viax has something more interesting: a compounding loop.

They run agents internally first. Not in a sandbox. Not against synthetic data. Against real production complexity — the actual mess of enterprise software in the wild. Medical device order flows. Distribution logic. Publishing subscription models. The agents that triage support, validate deployments, and draft configurations get sharper every day against real customer environments with real edge cases that no benchmark dataset would ever capture.

An agent that operates across three Fortune 500 customers' revenue execution environments doesn't just get faster. It gets categorically smarter about how enterprise revenue logic actually behaves. That's not a dataset you can buy. That's an asset you earn by doing the work.

Then those same agents become a customer-facing product.

Platform add-on. Subscription. Infrastructure cost plus usage plus margin. The customer doesn't buy "AI" — nobody knows what that means anymore. They buy operational capability. Measurable reduction in time-to-revenue. Elimination of configuration drift. Weeks of implementation work compressed into hours of human review.

The enterprise software industry has spent two years desperately asking: "Where does AI actually create value that someone will pay for?" Most of the answers have been embarrassingly thin. Summarization. Co-authoring. Semantic search. Important, maybe. But not the kind of thing that moves a P&L or survives a CFO asking "what exactly am I getting for this?"

viax's answer is thicker: AI agents that execute revenue operations autonomously, validated against real enterprise complexity, sold as a subscription alongside the platform customers already trust.

That's not a feature announcement. That's a business model.

The Category No One Named Yet

There's a pattern in enterprise technology that repeats about every fifteen years, and if you've watched it happen twice, you can see it forming a third time.

A new capability emerges. The incumbents try to absorb it into their existing products. They add a tab. They acquire a startup and suffocate it. They announce a "suite" and ship a checkbox. A small number of companies build the capability natively — not as an add-on to an existing architecture, but as the architecture itself. The native builders eventually define the category. The incumbents eventually concede it.

SAP absorbed CRM. It took Salesforce building CRM natively — on cloud, multi-tenant, API-first — to create a category worth $300 billion. Oracle absorbed data warehousing. It took Snowflake building it natively on object storage to break the grip. In both cases, the incumbent had more money, more customers, more engineers, and more conference keynotes. In both cases, the native builder won because architecture matters more than installed base when the paradigm shifts.

The same thing is forming now with agentic revenue execution. SAP will add "AI agents" to S/4HANA. They'll demo it at Sapphire. It'll work on stage with pre-loaded data. Oracle will add "autonomous operations" to their cloud suite. Larry will say something about it being the most important thing since the database. Both will be constrained by architectures designed for human operators navigating menu structures — not autonomous agents executing business logic programmatically at machine speed.

viax is in position to define this category because they already built the substrate. They didn't have to retrofit. They aren't bolting agents onto a monolith and calling a press conference. They're running agents through a system that was purpose-built to be operated by machines — and that system's reason for existing is to make ERP work better, not to replace it.

That last part is what makes the positioning so elegant. viax isn't asking enterprises to rip out their ERP investment. They're asking enterprises to stop abusing it.

The companies that win the agent era won't be the ones with the best models. They'll be the ones whose architectures make agents effective. Models are commoditizing. Architecture is not.

The Uncomfortable Truth

Every enterprise CTO reading this is thinking: "We're already doing AI." You are. You're doing it the way Blockbuster was "already doing" online video in 2007. You've got a Copilot that drafts emails nobody needed, a chatbot that confidently fabricates your own return policy, and an "AI strategy" deck that's been in review since Q3. Congratulations. You're transforming.

The companies that will matter in five years are the ones building systems where AI doesn't assist humans — it replaces entire categories of work. Not because the humans aren't needed. But because the work itself — the configuration validation, the regression testing, the root cause analysis, the revenue modeling, the pricing determination tracing — was never work that required human judgment. It required human patience with bad tooling. It required someone willing to click through fourteen screens to verify that a condition record didn't break when someone updated a partner function.

That's not knowledge work. That's penance.

viax removed the bad tooling. The agents handle the rest. And ERP finally gets to do what it was always supposed to do — be the system of record, not the system of everything.