You can almost hear the hum. It’s not coming from a server farm or the air conditioning in a sprawling open-plan office. It’s the sound of a silent, systemic change taking place inside the very architecture of enterprise software. For years, we’ve talked about AI as a feature, a nifty add-on to make our existing tools a bit smarter. But if you were listening closely to what Box CEO Aaron Levie had to say at the recent TechCrunch Disrupt 2025 conference, you’d realise we’ve been thinking far too small. The conversation is no longer about features; it’s about a fundamental rewiring of AI enterprise workflows that signals the most significant platform shift in a generation.
Levie, who has long been a sharp observer of the enterprise software landscape, didn’t pull any punches. He laid out a vision not of AI replacing Software-as-a-Service (SaaS), but of a future where the two enter into a powerful, and necessary, symbiotic relationship. This isn’t just another incremental update. This is a complete re-imagining of how organisations function, how software is built, and even who—or what—the primary user of that software is. And if he’s right, the very foundations of the multi-trillion-dollar enterprise software market are about to be reshaped.
The Next Act: SaaS Gets a Brain
For the better part of two decades, the SaaS model has been the undisputed king of enterprise IT. Predictable, scalable, and accessible from anywhere, it moved businesses from clunky on-premise servers to the cloud. But its primary strength has also been its main limitation: it’s largely deterministic. You click a button, and the same, predictable action happens every single time. It’s a system of record, a reliable but rigid framework for business processes. What Levie articulated—and what many in the industry are beginning to build—is the next stage of the SaaS evolution: a hybrid model.
The Hybrid Architecture: System of Record Meets System of Intelligence
Imagine your core business platform—your Salesforce, your Workday, or indeed, your Box—as the sturdy, reliable skeleton of your company. It holds all the critical information, the structured data, the financial records. It’s the source of truth. Now, imagine a fleet of intelligent, autonomous agents layered on top of that skeleton, acting as the company’s nervous system. These agents can access the skeleton’s data, but they operate with a degree of autonomy, making decisions, executing tasks, and reacting to new information in real-time.
This is the hybrid future. It’s not about ripping out the old systems. Instead, it’s about augmenting them with a new, non-deterministic layer of intelligence. As Levie aptly put it, you need a clear separation between the two. “You want to have some sort of ‘church and state’ between the deterministic side of your software and the non-deterministic side,” he noted, according to a report from the conference in TechCrunch. This separation is crucial for maintaining trust and control. You need your financial reporting system to be rigidly deterministic; you don’t want an AI getting creative with your balance sheet. But you absolutely want that same AI to analyse that balance sheet, cross-reference it with market trends, and draft a report on potential risks and opportunities.
This model allows for the best of both worlds. The deterministic core ensures stability and auditability, while the non-deterministic layer of autonomous software provides the dynamism and intelligence that businesses now crave. The system of record remains the foundation, but the system of intelligence is where the real action will be.
A New Kind of User is Logging On
Perhaps the most startling prediction Levie made concerns the very definition of a “user”. For decades, software has been designed for humans. We’ve obsessed over user interfaces (UIs) and user experiences (UX) designed for eyeballs and mouse clicks. That era is rapidly drawing to a close. The new primary user of enterprise software will not be a person. It will be an agent.
From a Thousand Employees to a Million Users
Levie’s statement on this was unambiguous: “We’ll have about 100 times more, maybe 1,000 times more, agents than we have people.” Let that sink in. A typical mid-sized enterprise with 5,000 employees could soon have 500,000 or even 5 million “users” actively working within its systems. These users won’t be logging in to check their email or fill out an expense report. They will be tireless digital workers, executing complex workflows 24/7.
This sea change has profound implications for software design. The focus will shift from graphical user interfaces to application programming interfaces (APIs). The most important question for a software developer will no longer be “How does this look to a human?” but “How easy is it for an AI agent to understand and interact with this service?”. This heralds an era of process automation on a scale we’ve never seen before.
Consider these common workflows transformed by agents:
– Sales Prospecting: An AI agent could monitor industry news, social media, and CRM data to identify potential leads, research their business needs, draft a personalised outreach email, and schedule a follow-up, all before a human salesperson has had their morning coffee.
– Financial Auditing: Instead of teams of junior auditors manually checking invoices against purchase orders, a fleet of agents could perform the task in minutes, flagging only the true exceptions that require human judgment. The accuracy would be higher, and the cost drastically lower.
– Supply Chain Management: An agent could constantly monitor weather patterns, shipping lane congestion, and supplier inventory levels, proactively re-routing shipments and adjusting orders to prevent disruptions before they even occur.
These aren’t just efficiency gains; they represent a move toward a more resilient and predictive operational model, all driven by AI enterprise workflows. The human worker is elevated from a doer of repetitive tasks to a supervisor of agents and a handler of exceptions—the truly complex problems that require creativity and strategic thought.
The Economic Shake-Up: Rethinking How We Pay for Software
If your primary users are no longer people, then the traditional “per-seat” licensing model that has dominated SaaS for twenty years becomes utterly obsolete. Charging per agent would be nonsensical and economically unviable when their numbers scale into the millions. This forces a complete rethink of software pricing, one that aligns cost directly with value generated.
Goodbye, Per-Seat. Hello, Consumption.
The future is consumption-based. Just as we pay for electricity by the kilowatt-hour or cloud computing by the CPU cycle, we will pay for enterprise AI by the task completed, the API call made, or the tokens processed. This model offers far greater flexibility and fairness. A small business running a few agents for basic automation will pay a fraction of what a global corporation running millions of sophisticated agents will.
This shift presents both challenges and opportunities. For customers, budget predictability becomes more difficult. An inefficiently coded agent could rack up a colossal bill, much like a leaky tap running up a water bill. This will create a new C-suite role—someone responsible for managing the “AI burn rate.” For vendors, it means their revenue is tied directly to the utility and performance of their product. If their agents aren’t creating value, they won’t get used, and revenue will dry up. It’s the ultimate form of accountability.
A Gold Rush for the Agent-First Startup
This entire platform shift, as Levie described it, creates a massive opening for newcomers. He stated, “We are in this window right now…there’s a complete platform shift happening in tech.” Legacy SaaS companies, with their monolithic codebases and business models built around per-seat licenses, will find it difficult to pivot. Retrofitting a system designed for human interaction to serve millions of autonomous agents is a monumental engineering challenge.
In contrast, a new wave of “agent-first” startups can build their entire architecture, pricing model, and go-to-market strategy around this new paradigm from day one. They aren’t burdened by legacy technology or customer expectations. They can leapfrog the incumbents by creating platforms designed specifically for building, deploying, and managing these new AI workforces. We are already seeing the early signs of this, with companies emerging that focus exclusively on providing the tools to orchestrate and monitor multi-agent systems.
Much like the shift to cloud gave us Salesforce and the move to mobile gave us Uber, this transition to an agent-driven enterprise will create its own titans. The question is, who will they be? And will the current giants of tech be able to adapt quickly enough to remain relevant in a world where their main customers are bots?
The silent takeover is already underway. The hum of autonomous software executing millions of tasks is getting louder. For businesses, the choice isn’t whether to adopt this new model, but how quickly they can do it. Those that embrace the hybrid architecture and begin cultivating their AI workforce will build a formidable competitive advantage. Those that cling to the old ways, designing software only for humans and selling it by the seat, risk becoming relics of a bygone era.
What are your thoughts on this agent-driven future? How can organisations prepare their human workforce for a world where they collaborate with, and manage, armies of AI agents? The floor is open.
