Despite online discussion, the actual penetration of coding agents among engineers in companies outside the Bay Area is still very low.

By the end of 2026, there will be a rise in 'background' or 'proactive' agents that understand user workflows to anticipate needs and complete work autonomously.

The knowledge gained by a company through the difficult process of iterating and solving problems with AI, described as 'pain', is what translates into a competitive moat.

The concept of multi-agent systems is currently misunderstood, as the idea of functional sub-agents coordinating via a 'gossip protocol' is not yet feasible with current model capabilities.

The years 2025 and 2026 are predicted to be incredible for the adoption and optimization of coding agents.

The adoption of AI agents within companies is still in a very early stage as of 2025, with penetration remaining low.

OpenAI's code review product has gained significant traction, catching numerous bugs both internally at OpenAI and for external customers.

The OpenAI Codex team uses a balanced approach to product development that includes both formal evaluations and close monitoring of customer feedback and usage patterns.