Agents: The New, New Kingmakers

The piece argues that autonomous AI agents are rapidly supplanting human developers as the primary decision-makers in software design and infrastructure. Tracing a line from early spell-checkers through neural code completion (TabNine) and GitHub Copilot to ChatGPT’s prompt-driven code generation, it shows how models now produce not only code but also choices about languages, libraries and architectures - sometimes ignoring explicit constraints. That shift changes cost dynamics (token spend vs. developer salaries), decision-making authority (agents versus human consensus), and perceived velocity (faster prototyping versus mixed DORA/METR outcomes). Vendors must choose whether to blend AI into human workflows, design strictly for agents, or support both, and many risks exist from superficial “AI-washing” versus genuine agent-first products.

Numerous companies are already optimizing for agents: Neon (now part of Databricks) reports agent-created database instances rising from 30% to 80%; Vercel saw agent-triggered deployments jump from under 3% to over 50%; Daytona pivoted entirely to agent runtimes; Observable added agent-first notebooks; Netlify coined “Agent Experience”; E2B reports over a billion sandboxes; Pamir.ai and others build agent-specific hardware and runtimes. The consequence is a market reorientation: developer influence shifts to whoever controls a handful of dominant models. Developers who can choose and guide those models retain power; others risk becoming advisors to agent-driven choices that may be narrower and more conservative than human-driven ecosystems.