AI coding = platform war
AI tools are shifting from neat demos to owning the workflows that run software, and vendors are racing to capture that control. Observers point to examples like Shopify’s AI Toolkit and AI systems operating existing apps through their interfaces as evidence that coding features are becoming a route into the software stack rather than just developer tools (The Verge). Shopify’s toolkit and a Business Insider report showing OpenAI’s Codex automating Lightroom are cited as concrete cases of this pattern (nadcab.com, businessinsider.com).
Artificial intelligence coding tools are turning into control panels for software work, and Shopify, OpenAI, Google, and Anthropic are all pushing into that layer. (theverge.com) On April 9, 2026, Shopify said its new AI Toolkit lets outside assistants use Shopify documentation, application programming interface schemas, code validation, and command-line store actions. Shopify published the launch in its developer changelog, not a keynote. (shopify.dev) OpenAI’s Codex is pitching the same shift from helper to operator. OpenAI says Codex can handle planning, feature building, refactors, reviews, and releases across tools, while Business Insider reported on April 11 that Codex denoised 50 Adobe Lightroom photos without an application programming interface or plugin. (openai.com, businessinsider.com) Google and Anthropic are building similar command surfaces around coding agents. Google’s Gemini command line interface is an open-source terminal agent that uses local or remote Model Context Protocol servers for tasks like fixing bugs and improving test coverage, and Anthropic says Claude Code can read a codebase, edit files, run tests, and deliver committed code. (developers.google.com, anthropic.com) The contest is no longer just about who writes the best snippet. It is about who becomes the layer that sits between a worker and the software stack, whether that stack is a code repository, a store back end, or a desktop app. (theverge.com, openai.com, shopify.dev)) That helps explain why vendors are adding ways for agents to use existing tools instead of waiting for every service to expose a clean interface. Anthropic’s March 23, 2026 release notes said Claude users on Pro and Max plans can give the system “computer use,” letting it open files, run development tools, click, and navigate what is on screen. (support.claude.com) Anthropic also said in March that it acquired Vercept to push those abilities further, describing computer use as a way for Claude to complete multi-step tasks inside live applications “the way a person at a keyboard would.” CNBC reported on March 24 that the feature can open apps, browse the web, and fill in spreadsheets after a user sends a task. (anthropic.com, cnbc.com) Google is making a parallel bet on standards and plumbing. Its Gemini application programming interface docs published last week tell developers to connect coding assistants to Gemini Docs Model Context Protocol and “Skills” so agents stay current with product changes instead of relying on old training data. (ai.google.dev) The older model was autocomplete inside an editor. The newer model is an agent with access to repos, terminals, schemas, cloud sandboxes, and sometimes the screen itself, which gives the vendor that hosts the agent a better shot at owning the workflow around the code. (theverge.com, openai.com, developers.google.com, anthropic.com) That is why a Shopify toolkit entry, a Codex Lightroom demo, and a terminal agent release all point to the same fight. The product that starts as “AI coding” is increasingly trying to become the software layer users work through every day. (shopify.dev, businessinsider.com, theverge.com)
