Agent builders face a widening gap between enterprise and open models

The race to deploy AI agents in workplaces is creating two distinct realities. On one side, commercial platforms are rapidly consolidating features through acquisitions [1], while most organizations still lack the infrastructure to develop their own agentic systems [2]. For builders working with open models, this divergence presents both challenges and opportunities.

Commercial suites leave little room for customization

Superhuman's acquisition of Fathom [1] exemplifies how productivity platforms are absorbing specialized agent functions into unified commercial offerings. These packaged solutions appeal to enterprises seeking quick ROI - Brazilian companies expect 15% returns from agentic AI by 2028 [2] - but they provide limited flexibility for teams needing tailored workflows. The 3% of companies that feel prepared to scale AI agents [2] likely have dedicated engineering resources to build beyond these off-the-shelf tools.

Open models show unexpected versatility

Meanwhile, open-weight models continue demonstrating capabilities beyond their original design. The newly released Iris agents [3] exhibit strong performance on tasks they weren't specifically trained for, including general tool use and office work. This emergent versatility makes them valuable building blocks for developers creating specialized agents, though they still require significant implementation work compared to commercial alternatives.

For builders, the practical takeaway is clear: focus on niches where commercial platforms can't compete. Open models like Iris [3] work best when solving problems too specific for generalized productivity suites, particularly in domains where existing tools show rigidity. The coming years will likely see commercial platforms dominate horizontal applications while open models power vertical innovations - a division that plays to the strengths of each approach.