Version 1

A cross-industry convergence on AI coding agents is crystallizing in May 2026, driven by four reinforcing signals: major AI labs are explicitly abandoning consumer features to replicate Anthropic's coding-agent success [1]; Oracle's Larry Ellison says AI now writes the company's own production code [2]; OpenAI is packaging the trend into enterprise tooling via Codex/GPT-5.5 [3]; and Anders Hejlsberg — creator of C# and TypeScript — says the primary developer task has already shifted from writing code to reviewing agent-written code [4]. The main actors are frontier AI labs (unnamed but described as a cohort), enterprise titans Oracle and Salesforce, OpenAI, and credentialed industry voices commenting on what the shift means for the profession.

When the CEO of Oracle says AI writes his company's code, the CEO of Salesforce quotes labs abandoning entire product lines to chase the same bet, and a foundational language designer says engineers are becoming project managers, the signal is no longer speculative. The commercial and professional structure of software development is being renegotiated in real time — with implications for developer hiring, enterprise software procurement, and which AI labs survive the next competitive cycle.

In the third week of May 2026, a cluster of high-profile statements from enterprise CEOs, AI labs, and foundational technologists revealed a shared inflection point: AI agents are moving from coding assistants to primary code authors, and the industry is reorganizing around that fact.

The competitive realignment is most visible at the lab level. Salesforce CEO Marc Benioff described a uniform strategic pivot across major AI labs, quoting their posture directly: "They're all like, 'We're only going to do coding agents too.'" [1] The framing is striking — it positions the convergence not as independent discovery but as deliberate mimicry of Anthropic, which built early commercial traction in the coding-agent space. Consumer-facing features like video models and conversational personas are being deprioritized in favor of enterprise coding workflows, which carry clearer revenue paths and enterprise contract structures.

At the production-software layer, Oracle CEO Larry Ellison made the adoption concrete: AI is now writing Oracle's own codebase. The workflow change he described is fundamental — developers no longer write step-by-step procedures but instead declare intent in natural language, with AI generating the executable logic. "We don't write the procedure, we declare our intent," Ellison said. [2] This is not a pilot or a sandbox experiment; framing it as current practice at one of the world's largest enterprise software companies signals that the threshold from 'AI-assisted' to 'AI-authored' has been crossed at scale.

The tooling layer is formalizing the shift. OpenAI published a case study documenting fintech company Ramp's deployment of Codex with GPT-5.5 for automated code review, reducing feedback time from hours to minutes. [3] The case study is promotional and lacks independent verification, but it illustrates how the agentic coding wave is being packaged into enterprise procurement: not as a developer toy, but as a replacement for time-intensive human review steps with measurable cycle-time payoffs.

The deepest signal may be the conceptual one. Anders Hejlsberg — who created C# and TypeScript and has spent decades shaping how developers think about code — framed the shift not as incremental productivity gain but as a role redefinition: the primary software engineering task has moved from writing code to reviewing and overseeing agent-written code. Architecture and judgment are now the dominant contributions. His summary — "We are all turning into project managers" [4] — carries particular weight precisely because it comes from someone who built the tools the current generation is being displaced from.