The distinction between front office and back office originated in financial services as far back as the 1720s—particularly in the City of London and later on Wall Street, where revenue-generating, client-facing roles were separated from the operational functions that processed and settled transactions. Over time, this practical and often physical separation became shorthand for where value was seen to be created, embedding assumptions about status, visibility, and importance that spread far beyond banking.
The concept of a front office and a back office is so deeply ingrained in business language that we rarely pause to question it: it’s a convenient way to distinguish between customer-facing activities and the operational work that supports them.
Yet the divide is not just organizational shorthand. It reflects a deeper belief about where value is created and who gets credit for it: the front office is where customers are won, and revenue is generated, while the back office is where operations, finance, HR, and technology keep the business running. Which might be another way of saying the front is where the action is, and the back is where the plumbing is hidden.
As customer-facing roles are celebrated and operational work recedes from view, the front/back-office divide continues to shape decision-making, investment, and culture. But while it once made sense when front- and back-office staff literally occupied different floors, does this division still hold up in today’s digital, automated world?
Why don’t we just have one office?
At least a few commentators argue that a hugely important new player in the business game—artificial intelligence, and particularly the new wave of autonomous agents—may be the factor that will render the front- back-office distinction meaningless. The Economist points out that classic back-office giant SAP (ERP) and traditional front-office leader Salesforce (CRM) have made significant, and expensive, moves to expand their portfolios. The goal: to leverage AI agents in ways that allow each to effectively match the capabilities of the other, creating a uni-office stack that can handle all the functions a business might need, effectively collapsing the old divide.
I’m not so sure. Many journalists, led by at least some analysts, are suggesting that everything will soon be replaced by agents, that classical software categories are disappearing, and that the future is all agents. That feels premature to me, and it’s unlikely to unfold as quickly as some predict.
That doesn’t mean the broader trajectory is wrong. The lines between categories are undeniably blurring, and there is more convergence—convergence that benefits both customers and brands. In today’s AI age, setting up a new account should take seconds with online ID and verification checks. The days of faxing details back from a website capture for manual entry into a mainframe database back at base are consigned to a slower and more inconvenient past.
Nonetheless, we should be cautious about any idea that agents, amazing as they are, will on their own magically transform an already complex tech stack into a single one-stop solution that can be relied on for every aspect of corporate data processing.
See also: Agentic AI in Industry: The Technologies That Will Deliver Results
One Ring to Rule them all? Not for business
The reality is that very few organizations put all their eggs in one basket. There are sound commercial reasons for this—even in the cloud, regulators discourage excessive concentration, and most commentators recommend multicloud commercial arrangements wherever possible. But there are also simple functionality problems: the days when you just assumed an IBM or a DEC, or even an ICL, would provide all an organization needed are long gone. Today, a best-of-breed approach is a much more effective way to extract real value from the market.
Some of the biggest enterprise software suites rely on many third-party additions to function as intended, and many market leaders are expensive to install and slow to adapt.
So if you already need third-party additions to make a system work, the idea that AI will turn an expanded SAP, Salesforce, or similar platform into a single source of truth for the entire enterprise feels a little overambitious.
My skepticism is also fuelled by what customers ask my team every day. They understand the need for back-office ERP systems and front-office customer relationship management, but they also recognize that neither is particularly good at handling the real fuel of day-to-day business: the documents and information that constantly flow back and forth between front and back office.
To take one core business function, HR, we are often brought into environments that already use market-leading, end-to-end HR platforms. Yet as a document management software provider, we help these customers store people’s information, manage HR processes, enable employees to ask questions, and archive records. Beyond that, we support searching for information, automatically generating employee documents, communicating those documents, and allowing employees to interact with them.
See also: Agentic AI and the Death of SaaS
AI, but not as a way to sell a monolithic solution
I could go on. These are core processes that category-leading applications are supposed to excel at, yet in practice, you still need interfaces, support, and “glue” in the form of document intelligence to make the whole picture work. And don’t get me started on invoices. To me, it sounds suspiciously like the large ERP companies are trying to use agents to recreate the monolithic giants of yesteryear, which is simply not how modern businesses operate.
To make business IT truly work, you need a linking element between enterprise platforms—something you might call “document intelligence.” It acts as a lingua franca, surveying, organizing, and helping firms leverage everything they need from their data and applications. And it doesn’t care what floor it sits on: it will help organizations wherever they need it.
Something tells me we’ll still be talking about document intelligence long after we’ve moved on from agents.
