Remember the ERP vs. Best-of-Breed debates from years ago?
If not, here’s the short version:
ERP vendors, such as SAP and Oracle, argued that a single integrated platform spanning finance, human resources, customer relationship management, order management, logistics, inventory management, and supply chain planning was the best approach. Why buy separate WMS, TMS, OMS, CRM, and planning systems — and deal with the cost and complexity of integrating and maintaining them all — when one platform could do it all?
Best-of-Breed vendors argued that beyond finance and HR, the capabilities offered by ERP vendors were too limited and not sophisticated enough to meet the complex needs of manufacturers and retailers. Deep specialization and domain expertise were what customers wanted and needed, they argued.
This often resulted in a battle between logistics teams and the CIO. As I wrote back in November 2013 in “The Rising Power And Influence Of Enterprise Software Users”:
I can’t tell you how many times, over the course of my career, I’ve come across the following situation: The logistics team spends several months evaluating software vendors, including the incumbent ERP vendor, and they ultimately select a best-of-breed solution, but when they present their decision to the CIO (who was disconnected from the process), he responds by asking why the ERP vendor wasn’t selected. This then triggers another round of evaluations, this time with corporate IT involved. Or the CIO simply overrules the logistics team and selects the ERP vendor in the name of “IT simplification and standardization.”
The ERP vs. Best-of-Breed debate ultimately became less relevant, as time and acquisitions (and the move away from proprietary architectures) blurred the distinction between the two camps. Generally speaking, there is much more feature-function parity between applications today, so companies are placing more emphasis on other important factors when evaluating solutions and vendors: time-to-benefit; how often new functionality is released (innovation cycle); how quick and easy those innovations are to deploy; and ease of use (see “Will Software Vendors Start Competing on Design?”).
Fast forward to today, and a similar debate is emerging around AI: Do you use one AI agent that does it all, or deploy many specialized AI agents?
The arguments for each approach echo those of the ERP vs. Best of Breed days, with each side presenting compelling reasons why their approach is better.
For example, the cost and complexity of integrating and maintaining many different applications was a key selling point of the ERP argument. Today, the analogous argument is the cost, complexity, and risks of coordinating — or orchestrating, if you prefer — many different agents that need to share data and coordinate with each other to make decisions and execute actions.
On the other hand, companies have always been multi-agent organizations — just with human agents. Specialists in procurement, transportation, planning, finance, customer service, manufacturing, and other functions work together every day, each bringing deep domain expertise while coordinating across the enterprise. From this perspective, AI should mirror — not replace — that organizational structure.
I’m not enough of an AI architect to know which approach will ultimately prove superior from a technical standpoint. But if history is any guide, there probably won’t be a one-size-fits-all answer. Some organizations will gravitate toward a single enterprise agent (or as few agents as possible), while others will deploy many specialized agents coordinated through orchestration layers. Either way, governance, accountability, interoperability, and trust — not simply the number of agents — will likely determine success.
One question still lingers for me, and I suspect for many others: If all the promises of AI and AI agents in supply chain management come true, what will be left for us humans to do?
