Four Routes to Building an AI Agent: What They Cost and Where They Break

A technical analysis published in mid-2026 outlines four distinct approaches to building AI agents, arguing that feature sets across platforms have largely converged and are no longer the key differentiator. The piece uses OpenAI's Agent Builder — launched at DevDay in October 2025 and deprecated by June 2026 — as a cautionary example of how platform choices can force costly migrations. The author identifies four critical questions teams should answer before choosing a build path: what the agent can affect, who is accountable when it errs, how frequently the underlying process changes, and whether the capability is a competitive advantage or routine infrastructure. No-code platforms such as Chatbase and Relevance AI are highlighted as useful for rapid proof-of-concept work, with entry-level plans typically around $24 per month, but they struggle when agents must follow conditional, multi-step logic. The analysis warns that Gartner expects over 40 percent of agentic AI projects to be cancelled by end-2027, attributing failures to escalating costs, unclear business value, and inadequate risk controls rather than technical shortcomings.