OpenAI–Windsurf Deal Canceled; Key Team Joins Google DeepMind, Cognition Steps In
PLUS: Former ByteDance PM (Strategy), Currently Building In Stealth
Editor’s Note - July 15th, 2025: Since this article was first published, OpenAI’s proposed $3 billion acquisition of Windsurf has fallen through. Instead, Windsurf CEO Varun Mohan, co-founder Douglas Chen, and several core researchers have joined Google DeepMind in what appears to be a high-profile reverse acquihire.
Meanwhile, Cognition, the AI company behind the widely discussed coding agent Devin, has signed a definitive agreement to acquire the remaining Windsurf team and IP. The deal allows Cognition to expand its position in the AI-powered IDE space - an area of growing competition led by players like Cursor, whose ARR recently crossed $500 million.
This edition is brought to you by VARA Network
Built on the Gear Protocol - Vara Network is a revolutionary Web3 application platform, removing many of the barriers associated with building, deploying, and operating dApps. Vara’s computer science paradigms like persistent memory and actor model, along with parallel data processing, allows for deep scalability, new design patterns and features. Vara’s unique architecture makes it completely frictionless for users. Learn more at Vara.Network
Venture Radar
AI Adoption in 2025: Shift Toward Workflow Integration
As of the first half of 2025, generative AI companies have accounted for over 53% of global venture deal value, with the majority of capital flowing toward applied use cases. Adoption is no longer centered on general-purpose models alone, but increasingly focused on domain-specific applications across industries such as healthcare, legal services, manufacturing, and logistics. AI tools are being integrated directly into operational workflows - from clinical documentation assistants and legal contract review platforms to AI-powered factory inspection systems and intelligent routing software. These applications are often embedded within vertical SaaS platforms, where they perform context-aware tasks like automated data entry, regulatory checks, and real-time recommendations. The shift reflects a growing demand for AI systems that are tightly aligned with industry-specific standards, decision processes, and compliance requirements.
Operational and Technical Realities of AI at Scale
Deploying AI at scale in 2025 introduces new technical and operational requirements. Models increasingly rely on dedicated GPU infrastructure, driving up demand for compute resources and creating supply constraints across cloud providers. Maintaining performance and accuracy now involves continuous fine-tuning, prompt engineering, and the use of RAG to anchor outputs in reliable context. Real-time AI tools depend on high-bandwidth data pipelines, low-latency inference, and scalable back-end systems. In high-stakes sectors such as healthcare and finance, AI models are frequently used in human-in-the-loop workflows to mitigate risk and ensure regulatory compliance. Accuracy, explainability, and fallback logic are built into production systems to address hallucination and reduce error rates. Integration into legacy enterprise stacks - including ERP, CRM, and workflow management platforms - is also becoming standard, increasing interdependencies and technical complexity in operational environments.
Geeks of the Week
Startup Name: CoffeeBlack AI
Geography: US
One-liner: Deploy “computer use” bots that click, type, scrape and upload in any desktop or web app - no coding required.
Founder(s) Background: Staff Solutions Engineer at HashiCorp.
Thoughts:
CoffeeBlack AI’s platform operates at the intersection of browser automation, agent-based AI, and enterprise task orchestration. By allowing users to automate repetitive desktop and web tasks - such as scraping data, filling out forms, uploading files, and navigating multi-step processes - it eliminates the need for API integrations or RPA setups. Its visual-language model-powered agents can even solve CAPTCHAs, making it particularly effective for scraping-heavy and high-friction environments.
The company’s go-to-market focus aligns with industries where operational systems are outdated, fragmented, or siloed - such as manufacturing, logistics, BPOs, and compliance-heavy functions. These sectors still rely heavily on human labor to bridge digital gaps between tools, giving CoffeeBlack a practical wedge into workflows where AI-native solutions are too early or costly to implement. In this way, the product acts as a transitional automation layer, helping businesses boost efficiency without overhauling their entire stack.
The broader market trend toward “agentic workflows” and AI copilots also supports CoffeeBlack’s positioning. As organizations experiment with task-specific AI agents, CoffeeBlack’s visual interface control offers a real-world pathway to deployment - especially in environments where API-based solutions are not viable. This positions them to ride the current AI tailwind while solving a real, immediate problem: how to automate the messy parts of everyday software use.
Founder(s) building in stealth
This edition is brought to you in partnership with Stella Capital.