This article presents a method to automate software fork maintenance using AI coding agents, framing it as a closed-loop feedback system in control theory. Applied to Cohere's fork of vLLM, it reduces the time to absorb upstream releases from weeks to days. The approach includes automated rebasing, measurement collection, and iterative fixing, with a case study on the Cohere Transcribe model.
AI agents can automate the full fork maintenance cycle: sync, measure, fix, repeat.
Fork maintenance is modeled as a feedback control system, with the agent as the controller.
Cohere automated incident response by connecting its enterprise AI agent platform North to cloud security platform Wiz via a custom Model Context Protocol (MCP) server. The security agent handles the complete workflow from triaging critical findings to generating IR reports, creating tickets, and updating Wiz status, reducing processing time from 30 minutes–2 hours per finding to 20 seconds. The article details architecture, three use cases (toxic combination analysis, assisted incident response, and autonomous weekly posture briefs), results, and implementation steps.
Cohere integrated North with Wiz via MCP to automate security incident response
Eight atomic tools exposed through MCP enable end-to-end workflow automation
This article discusses the importance of cultural awareness in AI systems, highlighting survey findings that many users face language barriers, cultural misunderstandings, and violations of norms. It calls for AI to be designed with cultural sensitivity to avoid marginalization.
Many non-English users switch to English for better AI performance, indicating a language gap.
Over a third of respondents felt AI lacks cultural understanding, and 63% experienced cultural norm violations.
Cohere introduces a new solution for fair scheduling of inference requests across tenants in multi-tenant LLM platforms, combining rate limiting, performance tiers, Deficit Round Robin, and priority selectors to prevent the "noisy neighbor" problem and ensure equitable GPU resource sharing.
Multi-tenant LLM platforms suffer from the noisy neighbor problem where one tenant's traffic burst impacts others.
Cohere's solution uses four mechanisms: rate limiter, performance tier, Deficit Round Robin, and priority selector.
Cohere is moving to 100 New Oxford Street, nearly tripling its London office footprint, to support growing R&D and commercial operations. This expansion reflects the company's commitment to the UK AI ecosystem and accelerates its sovereign AI strategy in Europe.
New office at 100 New Oxford Street provides over 14,000 sq ft, capacity for 100 people.
Expansion builds on recent partnerships with Aleph Alpha, Reliant AI, and UK government.
A 2023 paper estimating that 80% of U.S. workers have tasks exposed to large language models has been widely cited by major institutions. However, these scores are based on an older model and U.S. taxonomy, with limitations that compound when applied to policy. Better evidence tools exist but are not reaching policymakers fast enough.
80% exposure figure from 2023 paper cited by IMF, European Parliament, etc.
Scores based on GPT-4 era model and U.S. occupational taxonomy, with acknowledged limitations
Cohere and Mila announced a new academic research collaboration focused on improving AI evaluation across languages and cultures, starting with French-language cultural context in Quebec. The work aims to help frontier AI models better reflect the linguistic, social, and institutional nuances of Quebec French, moving beyond standardized language performance toward more culturally relevant and trusted AI systems.
Cohere and Mila partner to research AI evaluation for Quebec French cultural context.
Goal: make frontier AI models reflect linguistic, social, and institutional nuances of Quebec French.
As AI adoption expands beyond controlled pilots, mismatches between governance frameworks and actual use can arise. This article explores common AI governance challenges and failure modes, and outlines steps enterprises can take, including building an AI inventory, defining clear ownership, applying risk-based controls, and continuous monitoring.
AI governance gets harder as adoption scales, with loss of visibility and accountability being key risks.
Common issues include one-time approvals, unclear ownership, controls not matching risk, and sensitive data lacking appropriate controls.
Enterprise AI adoption typically follows a predictable five-phase progression: experimentation, tool adoption, internal platforms, strategic integrations, and AI-native transformation. Most organizations get stuck between Phase 2 and Phase 3, facing challenges like data access, trust gaps, and fear of model obsolescence. This article focuses on bridging the gap from pilot to production, emphasizing the need for internal platforms, unified data fabric, observability, and model optionality.
Enterprise AI maturity consists of five phases: experimentation, tool adoption, internal platforms, strategic integrations, and AI-native transformation.
Many enterprises hit a 'production wall' between Phase 2 (tool adoption) and Phase 3 (internal platforms).
Cohere open-sources Command A+, a 218B-parameter (25B active) mixture-of-experts model under Apache 2.0. Optimized for enterprise agentic workflows, it supports 128K input context, 64K generation, and text, image, and tool use. It significantly outperforms prior Command A models in reasoning, multimodal understanding, and multilingual tasks, while enabling efficient deployment via low-bit quantization and speculative decoding. Available on Hugging Face and Model Vault.
Command A+ is Cohere's latest open-source MoE model with 218B total and 25B active parameters, released under Apache 2.0. Designed for agentic tasks, it supports 128K input context and 64K generation.
Compared to Command A Reasoning, it achieves 85% (up from 37%) on Telecom benchmarks and 25% (up from 3%) on agentic coding, with gains across multimodal and multilingual tasks.
Model Context Protocol (MCP) is an open standard that connects AI applications to enterprise systems, simplifying data access and action execution. This guide explains how MCP works, its differences from APIs, RAG, function calling, and agents, common use cases, and security considerations.
MCP is an open protocol for connecting AI apps to enterprise systems, not a model or database.
Uses client-server architecture with resources, tools, and prompts as core features.
AI is increasingly applied to business intelligence to make data more accessible and useful. This article explains what AI in BI means, where it creates value, and key considerations for enterprise adoption.
AI in BI enables natural language queries, automated summaries, and anomaly detection.
AI-powered BI supports predictive analytics, root cause analysis, and role-specific insights.
RWS and Cohere collaborate to build a specialized translation model for Language Weaver Pro, leveraging Cohere's LLM and RWS's language expertise. The model outperforms competitors in 31 of 32 languages, offering cultural intelligence, security, and compliance for enterprise use.
RWS and Cohere co-developed a specialized translation model powering Language Weaver Pro.
The model outperforms competitors in 31 out of 32 languages, including DeepL.
Research reaches its full impact when researchers can see what they're doing clearly. Visualizations are key. This article discusses the challenges with existing tools like Matplotlib and Figma, and introduces co/plot, a tool built at Cohere Labs for quick iteration while preserving reproducibility and accuracy. co/plot was tested on Tiny Aya with 70+ languages and released as open source to support the research community.
Visualization is crucial for clear communication and progress in research.
Existing tools like Matplotlib and Figma have bottlenecks in iteration and accuracy.