Why AI coding tools like Cursor and Replit are doomed - and what comes next (www.zdnet.com)

Seasoned exec Jeremy Burton argues that well-funded AI coding startups like Cursor and Replit face an existential threat because their core value—automatic code generation—is becoming indistinguishable from the capabilities of foundation-model providers (OpenAI, Anthropic, Google, Microsoft). These giants control the LLMs that do the heavy lifting (Burton singles out Anthropic’s Claude and its Claude Code IDE as a likely “good enough” solution) and have the compute and capital to keep improving them (e.g., Amazon dedicating vast Trainium2 capacity to Claude’s next iterations). Startups have attracted roughly $3 billion collectively, but they largely sit on top of third-party LLMs, making it hard to sustain unique product differentiation or the engineering spend required to train competing models. Burton sees observability—the deterministic, telemetry-driven view of how code behaves in production—as the more defensible frontier. His company Observe builds knowledge graphs over app telemetry (Snowflake, AWS Iceberg, etc.) to pinpoint runtime issues and suggest fixes, a capability harder for LLM-first firms to replicate. Practical outcomes for pure-play code tools include building their own models (Cursor’s move), embedding observability, merging with DevOps players (Harness, Datadog, Dynatrace, Splunk), or being acquired or liquidated if funding dries up. The broader implication: the market will consolidate around either deep-model owners or platform players that combine code-gen with production observability and deterministic tooling.