Coding assistants are becoming the new platform
Developer-facing AI tools are shifting from autocomplete helpers to the de facto interface where engineers discover APIs, run tests and coordinate multi‑step work. The Verge argues that OpenAI, Google and Anthropic are competing to own that interface, which means internal platforms must be consumable by agents as well as humans. That suggests platform teams should prioritise machine‑readable SDKs, schemas and safe execution surfaces so internal standards remain authoritative rather than advisory. (theverge.com)
Coding assistants are moving from autocomplete sidekicks to the screen where developers find tools, edit files, run tests, and ship code. (theverge.com) The shift started with GitHub Copilot’s technical preview on June 29, 2021, which suggested code inside Visual Studio Code. By 2025 and 2026, GitHub, OpenAI, Google, and Anthropic were all selling agents that can change multiple files and execute steps, not just finish a line. (wikipedia.org, github.com, developers.openai.com, anthropic.com, codeassist.google) OpenAI says its Responses application programming interface is a “unified interface” for agent-like apps with tools including web search, file search, computer use, code interpreter, and remote Model Context Protocol servers. Anthropic says Claude Code can read a codebase, make changes across files, run tests, and deliver committed code, while Google says Gemini Code Assist now includes a terminal agent through Gemini Command Line Interface. (developers.openai.com, developers.openai.com, anthropic.com, codeassist.google) That changes where developers encounter a company’s internal platform. If an engineer asks an agent to add payments, fetch customer data, or deploy a service, the agent will reach first for the software development kit, schema, command, and policy it can parse automatically. (theverge.com, modelcontextprotocol.io) Anthropic’s Model Context Protocol, announced on November 25, 2024, is one sign of that redesign. The company described it as an open standard for connecting assistants to repositories, business tools, and development environments, and OpenAI’s current documentation says its Responses application programming interface can connect to remote Model Context Protocol servers too. (anthropic.com, developers.openai.com) GitHub and Microsoft are making a similar bet inside the editor. GitHub said on February 6, 2025 that Copilot’s agent mode could implement changes across multiple files, and Visual Studio Code now describes Copilot as bringing “AI agents” into the editor, including third-party agents from Anthropic and OpenAI. (github.com, code.visualstudio.com) Google is pushing the same pattern across enterprise development tools. Its current Gemini Code Assist materials say the product works in integrated development environments, offers a command-line agent through Gemini Command Line Interface, and supports Model Context Protocol setup so the assistant can use current documentation instead of stale training data. (codeassist.google, docs.cloud.google.com, ai.google.dev) The result is a new requirement for platform teams: the internal platform has to work for humans and for bots. A wiki page and a tribal convention are weaker inputs than a typed software development kit, a machine-readable schema, and a safe command surface with clear permissions. (theverge.com, anthropic.com, developers.openai.com) The companies building these assistants are also competing to become the default doorway into development work. If that doorway belongs to the agent, the platform that gets called most often may be the one that writes the rules developers actually follow. (theverge.com, github.com, anthropic.com, developers.openai.com)
