Why Tech Giants Are Building AI Coding Tools In-House

The cost of AI is forcing a reckoning at the world's most data-heavy companies. Rather than paying premium prices to OpenAI and Anthropic for off-the-shelf coding tools, Coinbase, Shopify, and Ramp have each built their own internal AI coding agents, according to reporting from The Information.

Coinbase launched its agent, Forge, in April and distributed it to all 2,500 engineers. Forge connects via Slack, GitHub, and an internal web interface, and can handle everything from brainstorming new features to writing and reviewing code inside production systems like Datadog and Sentry. The key innovation is a model router that dynamically assigns tasks to the most cost-efficient model available, including open-source options alongside Anthropic and Google.

The results are striking. Coinbase CEO Brian Armstrong reported cutting AI costs by nearly half while continuing to increase token usage. The company uses a prompt-caching system to avoid reprocessing repeated inputs, and defaults cheaper open-weight models for routine tasks while reserving high-performance models for complex planning.

Key takeaways:

• Shopify's internal agent, River, was used by 75% of employees as of May and delivers more coding output at lower token cost than external tools.
• Ramp's agent, Inspect, integrates with DevOps platforms and is model-agnostic, supporting any state-of-the-art model alongside custom workflows.
• The real strategic edge may not be cost savings alone: owning the orchestration layer between developers and foundation models could become a durable competitive advantage.

Not every in-house build pays off. Walmart capped usage of its internal agent, Code Puppy, after costs surged. The lesson: cost control requires intentional architecture, not just internal development.

Read the full article on Digital Today