The shift from human-led workflows to AI agents is reopening the market for the first time in a decade, transforming necessary labor into scalable code.
The End of the Evangelism Era
In 2005, when we started Box, the challenge was one of evangelism. We had to convince skeptical IT leaders that the cloud was safe, that the internet was fast enough for business, and that their data wouldn't vanish into the ether. It was a slow, uphill battle against a lack of infrastructure. Today, the landscape for AI is the polar opposite. There is no 'why' left to answer; the debate ended the moment executives used ChatGPT and realized it could outperform their own marketing teams. The challenge has shifted entirely from persuasion to implementation.
We are currently in a rare window—roughly a four-year span that began a year ago—where the next hundred great companies will be born. This period is unique because AI is finally addressing the 'unstructured' world. For decades, enterprises mastered structured data—the rows and columns of revenue and IDs. But the vast majority of corporate knowledge is trapped in contracts, invoices, and presentations. AI agents are turning these static files into active knowledge bases, allowing computers to finally 'read' and automate the workflows that define modern work.
The Strategic vs. Necessary Labor Theory
A common misconception is that AI is a zero-sum threat to employment. In reality, most corporate time is currently consumed by 'necessary but non-strategic' activities—manual data extraction, information retrieval, and administrative friction. There is a massive backlog of work that companies want to do but cannot justify at a $120,000 annual salary. If it costs too much to translate an ad into 100 languages or to review every single vendor contract for minor compliance risks, those tasks simply don't happen.
AI agents lower the economic threshold for these tasks. When the cost of labor drops toward the cost of compute, the volume of what a company can achieve expands. In a decade, the majority of work performed will consist of tasks that are currently considered economically non-viable for human labor. This isn't just about efficiency; it’s about growth. A 50-person company can now operate with the sophistication of a 500-person firm, entering more markets and conducting deeper research than was ever possible on a startup budget.
Reopening the Nouns and Verbs of Business
From roughly 2008 to 2022, the 'core nouns and verbs' of the digital economy felt solved. Food was DoorDash, music was Spotify, and payroll was Gusto. Incumbents were using modern tech stacks, making it nearly impossible for startups to do anything other than derivative work. AI has effectively reopened the map. For the first time in a decade, there is a new set of complex workflows—specifically in professional services like legal, accounting, and specialized consulting—that software previously couldn't handle because they required human judgment.
However, the trap for new founders is building a simple 'AI wrapper' around existing categories. If you build a 'CRM with AI,' you are competing with incumbents like Salesforce who are exceptionally good at executing on obvious additions. The real opportunity lies in categories where there is currently no incumbent technology because the work was previously entirely manual. By building software that 'contains' the labor, startups can move away from seat-based pricing—which is limited by headcount—to consumption models based on the actual volume of work performed.
Core, Context, and the Liability Shift
To navigate this new landscape, founders must understand the distinction between 'Core' and 'Context.' Core activities are those that drive a company’s unique value—for Disney, that is animation; for a pharmaceutical firm, it is drug discovery. Context refers to necessary operations like HR or payroll. Most companies will never build bespoke AI for Context tasks because of the liability. If an internal AI system mismanages payroll, you cannot sue your own IT department. Companies will always prefer to pay a vendor like Workday to shift that liability.
The massive opening for AI startups is in building custom software for the Core parts of a business that were previously too expensive to automate. While storage and security are largely solved problems, the value extraction layer is wide open. We are moving from a world where we simply store documents to one where we use AI to turn those documents into active intellectual property. The goal for the modern founder is to find these 'green field' opportunities where incumbents are too focused on their existing enterprise install base to notice the shift in how value is created.
The Deflationary Advantage
Technology is a unique industry because its raw materials—storage, compute, and now tokens—get cheaper over time. This creates a supply-side deflation that benefits the software provider. If you charge a steady price while your internal costs drop, your margins expand naturally. This is why companies like Dropbox or Box remain highly profitable despite intense competition; users value familiarity, craft, and workflow over the marginal savings of switching to a slightly cheaper commodity provider.
For those starting today, the advice is to be 'super commercial' and highly ambitious. We are in a 10-to-20-year cycle where the fundamental economics of business are being rewritten. Ride the tailwinds of AI economics rather than fighting the headwinds of stagnant industries. Assume your competitors' AI is perfect, then find a strategy that wins anyway. The winners of this era won't just be those who provide the smartest 'tokens,' but those who build the most elegant workflows on top of them.