Major tech companies are building internal AI coding agents while continuing to use commercial tools, signaling a shift in how enterprises approach AI-assisted software development. According to a report published by The New Stack, Coinbase, Shopify, and Ramp have each developed proprietary coding agents for their engineering teams, yet none have replaced commercial products like Claude Code, Codex, or Cursor. The report finds that companies are choosing to own the "agent harness" — the execution environment that handles context, permissions, workflow orchestration, tool access, and verification — while continuing to rely on frontier models from Anthropic, OpenAI, and Google for reasoning.
Coinbase's internal agent, Forge, connects Slack, GitHub, and Linear into a workflow that can turn a bug report into a pull request with limited developer involvement, retrieving repository context, producing code, validating changes, and creating mobile builds for review. Shopify's River operates on top of a reproducible engineering environment supported by a company-wide monorepo, durable execution sessions, sandboxed environments, credential management, and reusable engineering skills, with the company disclosing that River now contributes to one out of every eight merged pull requests. Ramp's Inspect runs inside sandboxed virtual machines that closely replicate developers' local environments, integrating with Datadog, Sentry, LaunchDarkly, Buildkite, and other engineering systems while remaining model-agnostic and supporting multiple frontier models alongside MCP servers, custom tools, and proprietary workflows. LangChain recently open-sourced Open SWE, describing it as the public version of the same pattern already running internally at Stripe, Ramp, and Coinbase, acknowledging that this architecture has become repeatable.
The report states that internal agents primarily handle asynchronous engineering workflows that start in Slack or issue trackers, interact with internal systems, and validate changes against production telemetry, while commercial assistants continue to dominate interactive development sessions in which developers work directly within their editors or terminals. According to Chintan Turakhia, Senior Director of Production Engineering for Wallet and AI Engineering at Coinbase, the company's infrastructure now processes more tokens than before while reducing overall spending, with co-founder and CEO Brian Armstrong attributing the improvement to multiple architectural optimizations including intelligent routing, lower-cost default models, prompt caching, smaller context windows, and better visibility into AI spending. The report notes that these optimizations are only possible because Coinbase owns the gateway sitting between developers and foundation models, allowing platform teams to update routing policies centrally rather than asking thousands of engineers to switch tools.
The report explains that this pattern mirrors what happened with internal developer platforms over the past decade, where few enterprises built their own cloud infrastructure after adopting AWS or Azure but instead built opinionated platforms on top of public cloud services that standardized deployment pipelines, security policies, and operational practices. The same approach is now emerging around AI, with the language model becoming another infrastructure dependency while the enterprise-owned harness determines how that intelligence is applied, making model selection a configuration choice instead of a workflow decision made by individual developers. Owning the harness doesn't eliminate infrastructure costs, however — Walmart recently introduced usage limits for its internal coding assistant after demand exceeded available budgets, while Uber exhausted its annual AI coding budget within four months, and recent research from Stanford University and Microsoft Research found that autonomous coding workflows can consume up to one thousand times more tokens than interactive code generation, with even identical tasks frequently exhibiting order-of-magnitude differences in token consumption.
The report concludes that enterprises are beginning to treat AI agents as part of their internal developer platform, with the language model increasingly resembling interchangeable infrastructure while the enterprise-owned harness becomes the control plane that governs context, security, execution, verification, and economics. Model providers will compete to become the preferred reasoning engine inside platforms they don't control, and if common agent runtimes and open protocols continue to mature while enterprises keep their policy and context layers proprietary, competition will increasingly move above the foundation model. The strategic advantage lies in moving away from the model itself and toward the platform that determines how, when, and where it's used. This architectural choice may redefine procurement negotiations, as vendors who once sold complete solutions now find themselves bidding to power a single layer within systems they can't fully observe or influence.