Why enterprise AI still feels so hard

Enterprise AI spending is accelerating in 2025, but May Habib describes it as reeking of desperation, with tools like Copilot, ChatGPT Enterprise, and Coworker failing to deliver real ROI. She frames the core problem as an autonomy gap, a large gulf between what AI can technically do and what organizations are actually extracting from it. Rassan Shears characterizes the situation not as a stall but as an overload, with enterprises overwhelmed by too many options and unsure where to begin.

Both speakers argue the obstacle is organizational rather than technical. Shears says the real challenge is refactoring processes to take advantage of AI capabilities rather than doing old things with new tools. Habib adds that no number of LLM-funded deployment services firms will solve the organizational rewiring problem, and that AI implementation strategies sold to senior leaders get delegated down multiple layers, undermining transformation. Organizations are building small agents that replicate what individual employees used to do rather than rethinking operations from scratch.

Incentive structures compound the problem. Shears argues employees cannot be expected to act against 20 to 30 years of career orientation without new incentives, and that larger enterprises struggle to create the cultural movement and psychological safety that smaller organizations can achieve more easily. Habib says the mandate for transformation requires telling employees their new job is to wire themselves out of their old job, and that organizations framing AI adoption around cutting jobs scare people into inaction, while leading organizations frame it around growth.

A Rider study released around December 2024 showed roughly 75 percent of leaders said their organization had a clear AI strategy versus around 45 percent of individuals, illustrating a seismic perception gap. A separate finding shows three quarters of executives now describe their AI strategy as being for show, a phenomenon the speakers call AI theater. Organizational siloing around AI pet projects is described as calcifying into dysfunction, with leaders actively hiding their AI work from peers.

A small but growing group of end users has become what the speakers call an AI elite, described as five times more productive than peers and three times more likely to receive a promotion in 2025. Among Rider's AI Champions, customers who reach one billion tokens per month, roughly one third are in marketing, one third in sales and operations, and one third in finance, HR, and customer success, with 20 percent now carrying AI or agents in their job title. Token leaderboards are described as an effective mechanism for driving organizational focus on the autonomy gap.

Where agents have actually worked and scaled, the speakers observe that there was typically nothing before, meaning few enterprises are succeeding by replacing existing table-stakes processes. A regional bank in EMEA achieved five times the number of meetings set for in-branch bankers using autonomous agents going into the community. A 150-year-old financial institution used an agentic team to handle creative, compliance, and omnichannel work for a new product launch without touching legacy stacks or processes. Both speakers suggest that demonstrating AI success at the edge on a smaller scale is a concrete way to bring the entire organization along.

Even straightforward agent deployments bump up against whether IT will prioritize requests and whether internal teams know how to connect to MCP servers. Habib identifies a product gap in AI's current inability to maintain and connect context across interactions, describing agents as capable of reasoning like a teammate today but lacking persistent memory. Broad data connectivity for agents is described as not yet in everyone's hands, producing episodic and choppy AI experiences, though the speakers consider it solvable in the near term.

The speakers argue that writing code is now near zero marginal cost, making process understanding and organizational alignment more important than the mechanics of how something is built. Organizations that have broken away from the pack are described as treating learning, context, and experience as the valuable asset rather than the code itself, and as not suffering from sunk cost fallacy. Enterprises running pilots will likely need to abandon what they built and adopt newer external solutions as the technology shifts, though the accumulated learning and context from those pilots retains value even when the underlying tooling must be replaced.