Labs are pursuing a long-term bet: AI coding could empower agents that self-improve the code inside AI models, potentially edging toward artificial general intelligence.

The debate unfolds amid external pressures, including looming IPOs for major AI players and the need to justify massive capital expenditures with strategic upside.

Over the last eight months, AI coding tools have matured enough to reliably build entire software projects from plain-language prompts, attracting enterprises that spend heavily on development.

Labs are pouring resources into frontier AI models that are expensive to develop, with current model-access sales not yet offsetting the costs.

Leading labs like OpenAI, Anthropic, and Google are making AI-generated coding tools—such as Claude Code, Codex, and AlphaCode 2—a central research focus.

Profitability isn’t the only driver; the broader aim is to transform AI development and potentially accelerate progress toward AGI.

Autonomous AI coding agents could operate at scale with minimal human supervision, accelerating model improvements and pushing toward more capable AI systems.