The Third Awakening of ERP: How Agentic AI Makes Composable ERP Finally Practical

The oldest debate in enterprise software is back, and this time it is more accessible

The long arc of ERP

Enterprise Resource Planning did not begin as a monolith. It began as a collection of narrow, fiercely practical tools. Material Requirements Planning (MRP) tackled the factory floor. Demand planning wrestled with forecast accuracy. Inventory management, capacity planning, and shop-floor scheduling each lived in its own corner of the business, solving a specific problem with specific math.

What happened next is the pattern every enterprise technologist recognizes: success-bred ambition. These discrete capabilities were woven together into integrated suites, and the monolithic ERP stack was born. SAP R/3 became the archetype: a single, tightly coupled system spanning finance, logistics, manufacturing, and HR on a common data model. JD Edwards carved out the mid-market, particularly in manufacturing and distribution. Infor (through successive acquisitions of Baan, SSA, Lawson, and others) assembled an industry-specific portfolio. Oracle, PeopleSoft, and Baan rounded out a generation of suites that promised one thing above all: one system, one vendor, one version of the truth.

The promise was real. So was the cost. These stacks were powerful, but they were also rigid, expensive to upgrade, and slow to absorb innovation that originated anywhere outside the suite vendor's own roadmap.

Two trends that broke the monolith open

By the late 1990s and early 2000s, two forces converged to challenge the all-in-one premise.

The first was the maturation of application integration. CORBA gave way to web services, which gave way to service-oriented architecture (SOA). For the first time, systems could speak to one another through standardized interfaces rather than brittle, point-to-point custom code. Integration moved from an art practiced by specialists to an engineering discipline that enterprises could scale. The technical barrier to running a different system alongside your ERP—and making it actually work—began to fall.

The second was a leap in narrow, specialized capability that the suite vendors could not match. Nowhere was this clearer than in supply chain. i2 Technologies pioneered constraint-based optimization, using mathematical modeling to simultaneously reconcile material availability, production capacity, and delivery logistics—capabilities far beyond what SAP or Oracle offered in their native planning modules. Manugistics competed alongside i2, and together the two essentially built the supply chain planning and optimization market. At its peak in August 1999, i2 carried a market capitalization of over $28 billion on a customer roster that included Dell and General Motors [1]. Other specialists—Kinaxis in concurrent planning, Logility in demand and inventory optimization, JDA in retail and supply chain—extended the same principle: in a narrow domain, a focused vendor could out-innovate a suite vendor by years.

The great debate: best-of-breed vs. integrated suite

These two trends ignited a debate that would define enterprise architecture for two decades: best-of-breed versus integrated solutions.

The best-of-breed camp argued that no single vendor could lead in every functional area. You bought the best demand planning tool, the best warehouse management system, the best transportation optimizer and you integrated them. The payoff was superior functionality and the freedom to swap components as the market advanced.

The integrated camp countered with a rational yet powerful objection: integration is the problem. Every best-of-breed bolt-on meant another vendor relationship, another upgrade cycle, another interface to maintain, and another seam where data could diverge. The suite vendors sold certainty and confidence. One data model, one support contract, one throat to choke. For many CIOs, the operational risk of a fragmented landscape outweighed the functional upside of specialization.

For years, the debate was genuinely unresolved. Suite vendors steadily absorbed specialized functionality (SAP, Oracle, and JDA all made inroads into supply chain planning), while best-of-breed vendors fought to stay ahead [2]. The pendulum swung, but neither side won decisively because the underlying trade-off was real, and the integration tax on best-of-breed remained genuinely painful.

Public cloud changes the equation

The public cloud and the rise of cloud-native, API-first applications fundamentally shifted the calculus.

Suddenly you could adopt a piece of innovative functionality in the cloud and connect it to your core ERP through standard, well-documented system calls. A modern OData or REST API against SAP S/4HANA, a Microsoft Dynamics 365 SDK call, or an event-driven integration through middleware like Infor's ION meant that "integration" was no longer a multi-year, multi-million-dollar custom project. It became a configuration exercise [3].

Concrete examples abounded: a cloud-based concurrent supply chain planning tool (e.g.,  Kinaxis or o9 Solutions) reading order and inventory data from a core ERP through APIs and writing optimized plans back; a cloud transportation management system calling the ERP for order details and returning shipment confirmations; a cloud procurement marketplace sitting in front of the ERP's purchasing module. Each was a focused, best-of-breed capability that was now reachable from the core without the integration tax that once made best-of-breed impractical.

As one industry observer put it: in the cloud era, integration "becomes the vendor's problem to get all these bits and pieces to run," rather than the customer's [4].

Composable ERP emerges

This is the soil from which composable ERP grew. Gartner coined the term in the early 2020s as the successor to its earlier "postmodern ERP" concept, defining it as an adaptive technology strategy that enables foundational, administrative, and operational digital capabilities through which an enterprise can keep up with the pace of fast business change [5].

The essence is simple: rather than one monolithic suite, you assemble a portfolio of modular, API-first capabilities, often from different vendors, around a stable core. Gartner's research suggests organizations that adopt a composable approach implement new features roughly 80% faster than those on monolithic stacks [6].

But composable ERP is not merely best-of-breed rebadged. The critical difference is the architecture: components are designed to be independently deployed, upgraded, and replaced, connected through APIs and unified by a data fabric. The integration burden that crippled best-of-breed in the 1990s has been engineered away.

The best of both worlds

Here is what makes composable ERP strategically attractive rather than just technically interesting: it is a best-of-both-worlds scenario.

Legacy investments do not have to be thrown away. The business logic painstakingly built over decades, the tailored processes encoded in custom code, and most importantly the data accumulated in the system of record can all be retained in the core ERP. That core remains the authoritative system of record for financial accounting, compliance, master data, and the transactions where rigid control is a feature, not a bug.

Meanwhile, new functionality extends and innovates around the edge of the core. Where you need to move fast or differentiate; predictive analytics, specialized optimization, intelligent automation. You can plug in modern, composable capabilities through APIs. The stable center holds; the edges evolve at the speed of the market.

This resolves the old debate not by picking a side, but by dissolving it. You keep the integration certainty and data integrity of the suite where it matters, and you capture the innovation velocity of best-of-breed where it counts.

Agentic AI: the catalyst that makes composability strategic

Today, a new force is arriving that will promote and facilitate the same concepts of composable ERP. That force is agentic AI.

Until now, composable ERP demanded something scarce: skilled integration engineers who could wire APIs, orchestrate workflows across systems, and maintain the seams between components. That labor cost was the hidden tax on composability, and it is why many organizations treated composable ERP as an aspirational architecture rather than an active strategy.

Agentic AI changes this in three ways:

1. Agents become the orchestration layer. Rather than hand-coding every cross-system workflow, AI agents can coordinate processes across disparate systems that were never designed to talk to one another. Agents function as a UX and orchestration layer that turns multi-step, cross-platform processes into automated operations [7]. The agent reads from the ERP, reasons about the next step, calls the specialized cloud service, and writes the result back.
2. Agents can be built, tailored, and connected to existing ERP systems to facilitate novel decision-making, predictive analytics, and process automation. A planning agent can pull demand signals and inventory positions from the core ERP, invoke a constraint-based optimizer, evaluate scenarios, and recommend or execute a replan. A procurement agent can monitor exceptions, negotiate routine orders, and update the system of record. The pattern is the same: the core stays lean and composable, and agents act as the flexible application layer on top [8].
3. The integration barrier collapses. Emerging standards like the Model Context Protocol (MCP) are standardizing how agents connect to ERP tools, APIs, and data sources, often without custom integration code [9]. What once required an integration project now increasingly requires an agent configuration.

The result is profound. Composable ERP stops being an architecture that only the most engineering-rich organizations can afford. It becomes an approach that any enterprise with a modern, API-accessible core can pursue because the agents do the integration work that humans used to do by hand.

Why this time is different

The best-of-breed debate of the 1990s was settled by default: integration was too hard, so most enterprises chose the suite. The composable ERP conversation of the early 2020s was real but bounded: the architecture was sound, but the integration labor limited how far most organizations could go.

Agentic AI removes that last constraint. When agents can orchestrate across systems, connect to ERP functions through standardized protocols, and drive decision-making and automation at the edge of the core, composable ERP shifts from a technical architecture to a strategic initiative: one that is more accessible, more affordable, and more consequential than it has ever been.

The oldest debate in enterprise software is back. This time, thanks to agentic AI, it finally has a clear answer: keep the core, compose the edges, and let the agents run the seams.

[1] i2 Shows Supply Chain Software Isn't What It Used To Be
https://www.informationweek.com/software-services/i2-shows-supply-chain-software-isn-t-what-it-used-to-be

[2] ERP Vendors Expand Offerings, Make Inroads Into SCM Market
https://www.supplychainbrain.com/articles/1064-erp-vendors-expand-offerings-make-inroads-into-scm-market

[3] Infor Positions Industry Context as the Foundation for Agentic ERP
https://moorinsightsstrategy.com/research-notes/infor-positions-industry-context-as-the-foundation-for-agentic-erp/

[4] Composable ERP may break up monolithic systems
https://www.techtarget.com/enterprise-software/news/252511618/Composable-ERP-may-break-up-monolithic-systems

[5] Enterprise Resource Planning to Optimize Operations
https://www.gartner.com/en/information-technology/topics/enterprise-resource-planning

[6] Why CIOs must reimagine ERP as the enterprise’s composable backbone
https://www.cio.com/article/4100401/why-cios-must-reimagine-erp-as-the-enterprises-composable-backbone.html

[7] Reimagining ERP for the agentic AI era
https://www.technologyreview.com/2026/01/20/1129965/reimagining-erp-for-the-agentic-ai-era/

[8] Lean, composable, and agile: How ERP is evolving in the agentic AI era
https://www.deloitte.com/us/en/what-we-do/capabilities/applied-artificial-intelligence/articles/how-erp-is-evolving-in-agentic-ai-era.html

[9] Building Intelligent ERP Systems with Agentic AI (Part I)
https://softwarearchitectblogs.wordpress.com/2026/04/19/building-intelligent-erp-systems-with-agentic-ai-part-i/