This week’s news from Zed, Anthropic, and OpenRouter shows why better harnesses matter more than better models

This week’s news from Zed, Anthropic, and OpenRouter highlights the importance of better harnesses for models over just better models themselves. Harnesses are the software around the model that supply context, connect tools, route work, and check results.

Zed made headlines with its new Delta feature, which rebuilds how users review agent-written code. It introduces shared threads instead of pull requests, making it easier for teams to collaborate without opening pull requests. Zed claims its team members landed 570 changes to Delta’s main branch without opening a single pull request, showing the efficiency of this approach.

Anthropic is also improving the user experience by unifying its interface. It’s merging its Cowork and Claude Chat capabilities into one interface to reduce confusion for users. This change is aimed at making it easier for users to understand the capabilities and workflow possible with these tools.

OpenRouter is focusing on giving companies more control over where requests are processed. Its US in-region routing ensures that requests are decrypted, processed, and served inside the country, or rejected if that guarantee can't be met. This level of control matters, as Deloitte’s survey found that 77% of companies consider an AI solution’s country of origin when selecting a vendor.

In terms of cost, OpenRouter’s US routing is now generally available to business and enterprise customers, with Open-weight models accounting for roughly 60% of OpenRouter’s US-originating token consumption in August. The economics of inference are also improving, with Vercel reporting a 23.2% decrease in the average price per token in August, the third straight monthly decline. This trend is making companies invest more in the harness around inference, as they seek to reduce costs and improve efficiency.