Welcome to The Daily AI Briefing! In today's rapidly evolving AI landscape, we're tracking major developments across multiple fronts. From Windsurf's new in-house developer models to shifting user preferences on Poe, plus breakthrough research on LLM conversation capabilities and practical automation solutions. These innovations continue to reshape how we interact with artificial intelligence and what we can expect from these systems in both enterprise and consumer contexts. Today's topics: - Windsurf launches SWE-1 AI models for software engineering - Poe's usage report reveals shifting AI popularity trends - How to automate legal document analysis with Zapier - New study shows LLMs struggle with extended conversations - Latest AI tools and job opportunities Windsurf has made a significant move in the developer AI space with the release of its SWE-1 family of models. These in-house AI systems are specifically designed for the software engineering lifecycle and include three versions: the full-size SWE-1 for paid users, SWE-1-lite replacing Cascade Base for all users, and SWE-1-mini. What makes these models stand out is their ability to work across multiple interfaces—editors, terminals, and browsers—with a "flow awareness" system that creates a shared timeline between users and AI. Internal benchmarks show SWE-1 outperforming most competitors, sitting just behind models like Claude 3.7 Sonnet. This release comes shortly after reports of a $3 billion acquisition by OpenAI. In the broader AI ecosystem, Poe's Spring 2025 Model Usage Trends report provides fascinating insights into shifting user preferences. GPT-4.1 and Gemini 2.5 Pro quickly captured 10% and 5% market share respectively within weeks of launch, while Claude saw a 10% decline during the same period. Reasoning models have surged from just 2% to 10% of all text messages since January. The image generation landscape is also evolving rapidly, with GPT-image-1 gaining 17% usage and challenging established leaders. In video, China's Kling family has become a top contender with approximately 30% usage shortly after release, while ElevenLabs dominates the audio segment with 80% usage. For those looking to put AI to practical use, a new tutorial demonstrates how to build an automated system that analyzes legal documents uploaded to Google Drive. The process uses Zapier Agents to trigger automated workflows when new documents are added to a dedicated folder. The system leverages Google Drive to retrieve files, ChatGPT to analyze documents and identify concerning clauses, and Gmail to send summary emails. While this represents a powerful automation solution, the tutorial wisely notes that users should always double-check AI answers and consider hiding sensitive information. However, a new study from Microsoft and Salesforce researchers reveals important limitations in current AI systems. They found that leading LLMs including Claude 3.7 Sonnet, GPT-4.1, and Gemini 2.5 Pro significantly underperform during multi-turn conversations where instructions are gradually revealed. While achieving 90% success in single-turn settings, this drops to approximately 60% in multi-turn conversations. Models tend to "get lost" by jumping to conclusions or building on initially incorrect responses. Neither temperature adjustments nor reasoning models improved consistency, exposing a major gap between evaluation metrics and real-world usage. Among trending AI tools this week are Salesforce's enterprise-ready xGen Small, AlphaEvolve's coding agent making mathematical discoveries, Stable Audio Open Small for text-to-audio music generation, and Nous Research's Psyche open infrastructure. As we wrap up today's briefing, it's clear that AI continues to advance rapidly across multiple domains. From specialized developer tools to platforms tracking real-world usage patterns, the ecosystem is maturing and revealing both new capabilities and limitations. The gap between single-turn and multi-tur