June 2026 Summaries

Maintaining a long-lived fork of an actively developed project is challenging, as each upstream release introduces changes that can disrupt the fork's functionality. The described method leverages AI coding agents to automate the cycle of syncing, measuring, fixing, and shipping updates to such forks, significantly reducing the time required for these tasks from weeks to days. This approach, applied to Cohere's fork of the vLLM project, uses a control theory framework to treat upstream changes as disturbances and employs a feedback loop to restore the fork to a working state with minimal human intervention. The process involves detecting upstream releases, rebasing the fork, running tests, and applying fixes until the desired outcome is achieved. This method, which enhances efficiency and reduces manual effort, is open-sourced at cohere-ai/vllm-skills and can be adapted to other codebases by defining measurable criteria for maintaining a "healthy" fork.

Cohere has developed a sophisticated AI-driven solution to streamline incident response in regulated industries and government sectors by integrating its North enterprise AI agent platform with the cloud security platform Wiz through a custom Model Context Protocol (MCP) server. This integration automates incident response workflows, handling tasks such as triaging critical findings, drafting reports, creating tickets, and updating statuses, significantly reducing the time from insight to action. The system addresses the challenge of transforming high-fidelity signals from Wiz into actionable steps without human intervention by using North's reasoning capabilities to analyze attack chains and prioritize risks. Automated processes, such as weekly security posture briefs, ensure continuous visibility and situational awareness for security teams while maintaining data fidelity and avoiding over-reliance on personnel, thus offering a scalable and precise security operations framework.

A survey conducted to explore cultural and linguistic experiences with AI systems revealed significant cultural awareness gaps in large language models (LLMs), especially when they default to dominant cultural perspectives. Participants from over 22 countries highlighted issues such as the language switching problem, where non-English speakers often switch to English for perceived higher response quality, and AI's lack of understanding of diverse cultural norms, affecting tasks like creative writing and localized information look-up. Notably, 63% of respondents experienced cultural norm violations, often due to the AI's Western-centric bias, and 67% expressed concern that AI might marginalize their cultures in the future. The study emphasizes the need for culturally aware AI that respects regional nuances, handles mixed languages effectively, and provides guidance aligned with users' cultural contexts to prevent exacerbating existing inequalities and misinformation.

Cohere has developed a solution called "Serving Fairness" to address the issue of "noisy neighbors" in multi-tenant SaaS platforms, where large language model inference requests can cause latency problems for other users due to uneven traffic bursts. The approach combines architectural patterns and a scheduling algorithm to ensure fair distribution of compute resources among tenants, using mechanisms such as rate limiting, performance tiering, and the Deficit Round Robin (DRR) algorithm. These mechanisms work in a structured flow where rate limiting controls request admission, while performance tiering prioritizes requests based on service-level agreements. DRR ensures equitable compute distribution by granting each tenant a budget of requests or tokens, preventing any single tenant from monopolizing resources. Within each tenant, requests are further ordered by priority, deadline, and arrival time, maintaining fairness and urgency. This system allows Cohere to provide a balanced and predictable service experience, ensuring that each tenant receives their fair share of resources without being affected by others' traffic spikes. Serving Fairness is now available for all users of Cohere models via their SaaS API and third-party marketplaces like AWS, with the company inviting feedback for continuous improvement.

Cohere is emphasizing its commitment to the development of sovereign AI in the UK, aiming to aid businesses and government agencies in deploying AI technology securely and autonomously. The company's London office, which is central to its engineering, research, and commercial operations, is expanding to accommodate its growing team, particularly in AI engineering and applied machine learning. Cohere's strong presence in London aligns with the city's proficiency in financial and professional services, making it an ideal location for its sovereign AI solutions, which are well-suited for highly regulated sectors such as finance, healthcare, and telecommunications. The UK is presented as a global AI hub with a unique combination of research talent and progressive regulatory frameworks, which supports Cohere's vision of contributing to the projected $600 billion sovereign AI market. The expansion includes a new office fit-out by Thirdway Design, aimed at fostering collaboration and innovation among Cohere's team members.

Frontier labs are examining the impact of technologies like Claude, Copilot, and ChatGPT in the workplace, highlighting the need for a comprehensive evidence base that goes beyond initial scores. The future of work debate centers around whether AI capabilities will advance, their economic implications, and the optimal policy responses, with the first question being the most evidenced. Policymakers are encouraged to consider AI exposure scores as one of many signals and to implement robust policies like worker protections and reskilling infrastructure. It's crucial for workers to be involved in shaping the future of work, using their firsthand experience to inform the process. Researchers should focus on developing measurement tools that evolve with AI capabilities and address the questions policymakers need to answer, treating workers as partners in research. The 80% exposure figure is a technical feasibility estimate, not a definitive forecast, and the future of work will be determined by collaborative decisions made by researchers, policymakers, and workers. The analysis emphasizes the importance of a solid evidence base for these decisions, urging readers to consult the full report for detailed insights and recommendations.

Cohere has introduced North Mini Code, its first open-source model designed specifically for developers, marking a significant step in creating a more sovereign AI ecosystem. This mixture-of-experts model is notable for its efficiency and agentic coding capabilities, featuring 30 billion total parameters with only 3 billion active, allowing it to perform robust software development tasks without requiring extensive hardware resources. North Mini Code is available under an Apache 2.0 license and can be accessed via platforms like Hugging Face and Cohere's Model Vault. It is optimized for code generation and agentic software engineering tasks, demonstrating competitive performance and speed advantages over similar models, such as achieving up to 2.8 times higher output throughput and a 30% advantage in inter-token latency compared to Devstral Small 2. The model supports agentic workflows, enabling developers to maintain control and flexibility over their coding infrastructure, with community feedback playing a crucial role in guiding its ongoing development.

Cohere Labs has developed a tool called co/plot to enhance the research process by improving data visualization, recognizing that clear visual representations are crucial for understanding and communicating research findings. The tool addresses limitations in existing workflows, such as the difficulty of iterating quickly with Matplotlib and the accuracy issues with Figma, by allowing fast prototyping with customizable styling that remains true to the underlying data. As part of their commitment to open science, Cohere Labs has released co/plot to the public, enabling independent researchers to iterate faster and share their work with more confidence. The development of co/plot has demonstrated its value in improving decision-making during research and is seen as an important step towards raising the standard of visualization tools available to the scientific community.