Hey friend,

Today, we’ll explore

• How AI is writing 80% of code inside Anthropic

• Mistral Medium 3 + Le Chat Enterprise

• Google’s Gemini 2.5 pro upgrade

• OpenAI’s $3B Windsurf acquisition

Let's dive in.

AI Is Writing Most Of Anthropic's Code

According to Boris Cherny, who works on the Claude Code team at Anthropic, approximately 80% of their code is written by AI. This came up when discussing how much of Claude Code was written by Claude itself. As a software engineer, I found this mind blowing.

While people will understandably question that number, there’s no doubt AI is changing how we think about coding.

Boris estimated Claude Code makes him personally 2× more productive as an engineer who "codes all day every day." For some Anthropic engineers, the productivity boost might be as high as 10×. This explains why VPs and CTOs are excited about the tool despite initial concerns - the ROI on engineer salaries makes the token costs worthwhile.

AI News

Here’s my pick of AI news this week:

1. Mistral Medium 3: 90% Of Claude's Power At One-Eighth The Cost

Mistral AI has launched Mistral Medium 3, a high-performance AI model delivering 90% of Claude Sonnet 3.7's capabilities at 8X lower cost, alongside Le Chat Enterprise, a privacy-focused productivity platform for business deployment.

• Mistral Medium 3 offers exceptional efficiency, performing at near-Claude levels while requiring only four GPUs for deployment and costing significantly less at $0.40/$2 per million input/output tokens versus Claude's $3/$15 rates.

• The model excels in professional applications, scoring above 90% on coding benchmarks (HumanEval, MultiPL-E) and outperforming Llama 4 Maverick in multilingual tasks.

• Le Chat Enterprise provides comprehensive business features including no-code agent builders, hybrid cloud/on-premises deployment options, and native integrations with Google Drive and SharePoint.

2. Google's Gemini 2.5 Pro (I/O Edition) Wins At Coding

Google has released Gemini 2.5 Pro Preview (I/O Edition), featuring enhanced coding and web development capabilities. I personally use Gemini more than any other model and feel that Google is starting to lead the way.

• Gemini 2.5 Pro significantly reduces function-calling errors compared to its March release and introduces new video-to-code transformation capabilities that improve developer productivity for visual front-end tasks.

• The model has demonstrated superior performance in code abstraction tasks compared to Claude 3.7 Sonnet

• Available immediately through the Gemini API and Vertex AI at the same price tier ($1.25M input tokens), with existing 03-25 model IDs automatically upgrading before a full showcase at Google I/O.

3. OpenAI's $3B Windsurf Acquisition Reshapes Developer Tooling Landscape

OpenAI has agreed a $3 billion acquisition of Windsurf, an AI-powered coding tool. The strategic purchase integrates Windsurf's coding editor with OpenAI's models to strengthen its position in developer tooling, with integration expected within months.

• Windsurf only has a reported $40 million in annualised revenue. This move reflects OpenAI’s need to counter big challenges from Google and Anthropic and to secure a dominant position in the AI coding space.

Your Opinion Matters

What did you think of today’s email? Your feedback helps me create better emails for you!

Got more feedback or want me to cover a specific topic? Reply to this email and let me know.

Owain