The biggest shift in enterprise AI adoption isn’t a technological bottleneck; it’s a human one. The real challenge lies not in whether the AI exists, but in how fast companies can redesign decades-old workflows and organizational charts to use it. According to Aaron Levy, CEO of Box, today’s enterprise leaders aren’t debating if AI will take over—that’s assumed. Instead, they’re racing to ensure it happens within their companies faster than at their competitors, a dynamic starkly different from the early, skeptical days of cloud adoption.

While the consumer world rapidly embraced generative AI, enterprise adoption faces unique hurdles: legacy IT systems, ingrained processes, and complex governance. However, unlike the cloud transition, which required rewriting software from the ground up, AI often acts as a new consumption layer atop existing systems. Incumbent SaaS companies with robust APIs may be surprisingly well-positioned, as AI agents can become super-users of their platforms. This allows for a sustaining innovation, expanding TAM by enabling automation of tasks where human users were previously impractical.

Yet this doesn’t spell the end for startups. AI unlocks entirely new software categories in fields like legal, healthcare, and financial services, where work has historically been unstructured and resistant to digitization. In these verticals, there are no natural software incumbents, creating a greenfield for innovation. Furthermore, the very nature of work is set to transform, with employees increasingly shifting from direct execution to managing and orchestrating AI agents—reviewing outputs, setting strategy, and integrating work.

Surprising Insights

• Incumbents may have an edge: Existing SaaS companies with strong APIs could be the natural home for AI agents in their category, as the agent can simply consume the API. This contrasts with the cloud era, where incumbents often had to rebuild everything.
• The biggest disruption is organizational, not technical: The primary bottleneck for enterprise AI value isn’t model capability, but the speed of human change management—rewiring processes, budgets, and job roles.
• AI cost is a marginal expense: The license cost for powerful AI developer tools is a tiny fraction (around ~1%) of an engineer’s salary, making it an easy budget decision for companies and unlikely to trigger zero-sum budget wars.
• Vertical SaaS durability comes from domain expertise, not code: The moat for vertical software companies isn’t complex technology (often simple CRUD), but deep, embedded knowledge of industry-specific workflows—expertise that translates well to building effective AI agents for that niche.

Practical Takeaways

• Focus on capacity, not just cost-cutting: Approach internal AI adoption as a tool to increase what your team can do—research more markets, prototype faster, run more campaigns—rather than purely as a lever for headcount reduction.
• Prioritize AI-native talent: Seek out and empower employees who instinctively use AI to work in new ways; they can teach the entire organization to operate at a new speed.
• Experiment in decentralized pockets: Instead of a top-down, company-wide transformation mandate, encourage decentralized experimentation with high-impact workflows to learn what works before scaling.
• Augment, don’t fully abstract: Don’t assume AI will replace all software interfaces. For many core operational functions (HR, ticketing), employees still want a standard, pre-built interface—they are buying the solved workflow, not a blank canvas.
• Treat entry-level roles as agent managers: For new hires, especially in technical fields, emphasize skills in prompting, reviewing, and integrating AI-generated work, preparing them for a role as an orchestrator rather than just a coder.