TechPrufer
Get started
Cohere
Cohere logoCohere · Feature trail

Agentic Workflow Orchestration

26 posts · Nov 2023 – Oct 2026

What this trail covers

Cohere's North Automations platform enables enterprises to build and orchestrate complex, end-to-end AI agent workflows. This post details the release of North Automations, which provides a centralized layer to coordinate diverse agents across the enterprise, delivering enterprise-grade governance and security at scale. Key features include turning complex workflows into simple outputs with plain language descriptions, scheduled runs, and auditable execution paths with loops and branching. Users can now create reusable agents and automations that can be shared across the organization, eliminating the need to rebuild similar workflows repeatedly. Prototype documents or apps directly in chat, or connect North to your favorite existing tools. North 2 features an advanced agent orchestration system that manages complex, multi-step processes independently. Core features include Skills, Libraries, Memory, Applications, Automations (with a drag-and-drop visual workflow builder), and Connectors to enterprise data sources. North 2 also emphasizes enterprise-first security with flexible deployment options (self-hosted, VPC, hybrid, or on-premises), SOC 2 Type 2, ISO 27001, and ISO 42001 certifications, guardrails, autonomy policies, rigorous security testing, and system observability. North Admin provides granular controls for cost management, rate limits, user quotas, and org-wide caps, with features like detailed analytics, token spend monitoring, and efficient token consumption with NVIDIA hardware. The platform is model-agnostic, allowing users to leverage Cohere's models or bring their own.

All Cohere trails
  1. North 2: Enterprise AI Without CompromisesThis post announces North 2, an evolution of Cohere's agentic AI platform. Key technical contributions include a redesigned agent harness for more sophisticated and reliable multi-step automations, enhanced intelligence through memory, libraries, skills, artifact, and app creation, and model agnosticism. It details the integration of enterprise connectors (Slack, SharePoint, Outlook, Jira, Notion, GitHub, etc.) and planned financial data providers. Security enhancements include flexible deployment options (on-prem, cloud, air-gapped), SOC 2 Type 2, ISO 27001, ISO 42001 certifications, autonomy policies, and red-teaming exercises. Granular controls in North Admin are expanded with user/agent-level visibility, token spend monitoring, and efficient token consumption with NVIDIA hardware. The post also highlights the technical benefits of using Cohere models on NVIDIA Blackwell and Hopper architectures for reduced token consumption.
  2. Automation’s Early Footprint: The ATE DatasetThis post introduces the Agentic Task Ecosystem (ATE) dataset, a novel corpus of approximately 696,000 published tools across 123,000 MCP servers. It presents a quantitative analysis of the agentic AI tool landscape, revealing that only 2.6% of published tools can fully automate recognized work tasks. The analysis categorizes the ecosystem into three main patterns: existing work represented at a finer grain, infrastructure for running agents, and a small amount of genuinely new work primarily focused on agent management. It also identifies that nearly half of U.S. occupations have no agentic tools represented and that technical feasibility, rather than worker preference, predicts which occupations receive tools. The post further details how the distribution of agentic tools across occupational tasks varies by industry, with specialized tasks in healthcare and computing receiving tools, while routine tasks in legal, production, and sales remain largely untouched by automation.
  3. Cohere Releases Command A+ for Sovereign AI | CohereThis post announces the release of Command A+, an open-source Mixture-of-Experts (MoE) model. It details the model's architecture (218B parameters, 24B active), efficiency for enterprise deployment (e.g., on two H100s or one B200 GPU), and suitability for specific workloads like RAG, SQL generation, and financial document analysis. Multimodal capabilities for visual data are also highlighted. The model's multilingual support (48 languages) and its design for sovereign AI, enabling on-premises and air-gapped deployments, are emphasized.
  4. Ensemble Brings Agentic AI to RCM Platform | CohereThis post details the development of a healthcare RCM-native LLM by Cohere and Ensemble. It highlights the creation of a custom model based on Cohere's Command family, fine-tuned with Ensemble's RCM-specific data, operational logs, and expert annotations. The contribution focuses on domain adaptation for healthcare complexity, establishing a domain-specific benchmark for evaluation, and enabling agentic reasoning for tasks like claim review, denial resolution, and clinical appeals. The post emphasizes the trade-offs of generic LLMs versus domain-specific models for regulated industries and showcases opportunities for automation in financial and clinical orchestration within RCM.
  5. What Are AI Agents? How They Work and Examples | CohereThis post defines AI agents as software systems that use an AI model to pursue a goal across multiple steps, use tools, and adapt to their actions. It details the core components of an AI agent: the AI model, instructions, task context/state/memory, tools, and the agent harness. It also discusses the benefits of AI agents, such as higher throughput and adaptive task execution, and the challenges, including compounding errors and control of access/actions.
  7. A Day in the Life of a Wealth Manager, With and Without AI | CohereThis post details the application of Cohere's North platform and its AI agents within the wealth management domain. It showcases how agents can automate client intelligence gathering, market research synthesis, and client communication drafting, replacing manual workflows with auditable, client-ready processes. It also highlights the use of a regulatory compliance agent for accelerating reporting and an operations intelligence agent for streamlining settlement and reconciliation. The post emphasizes the success factors for implementing AI in wealth management, including defined approval boundaries, grounded connectors, and time/quality metrics.
  8. Orchestrate Agentic Workflows with North Automations | CohereThis post announces the launch of North Automations, Cohere's platform for orchestrating complex AI agent workflows. It addresses challenges like narrow workstreams, lack of governance, and cost by providing a centralized layer for coordinating agents, enabling model selection per step for cost/performance balance, and offering features for control and governance (plan mode, versioning, approvals, usage analytics, token monitoring). The platform integrates with existing tech stacks and allows flexible model choices. Use cases demonstrate its application in marketing, customer success, and sales operations.
  9. Agentic Workflows | CohereThis post enhances the Agentic Workflow Orchestration capability by integrating agentic search with intent recognition. This allows chatbots to more accurately understand user goals and context, improving the system's ability to plan and execute multi-step retrieval processes. The system leverages agents to understand complex queries, break them down into sub-tasks, and orchestrate the retrieval of information from various sources, leading to more accurate and context-aware information retrieval.
  10. Creating a Security Agent with Cohere North and Wiz | CohereThis post details the integration of Cohere North with the Wiz security platform via a custom Model Context Protocol (MCP) server to create a security agent. It outlines the architecture, the eight atomic tools exposed by the MCP server to interact with Wiz (listing issues, getting details, identifying toxic combinations, searching vulnerabilities, getting security posture, querying assets, checking compliance, and updating issues), and demonstrates three use cases: blast radius analysis of toxic combinations, assisted incident response (including IR report generation, ticket creation, and status updates), and an autonomous weekly security posture brief. It also shares key technical lessons learned during development and provides a guide for replicating the implementation.
  11. Modernizing FOI Systems with AI Agents | CohereThis post details the application of AI agents to automate the maintenance of software forks, specifically syncing with upstream releases, resolving merge conflicts, running tests and benchmarks, and fixing regressions. This approach has compressed the time to absorb a new upstream release from weeks to days, with humans only reviewing the outcome.
  12. North Mini Code: Agentic Coding for Developers | CohereIntroduces North Mini Code, Cohere's first open-source agentic coding model. It's a 30B parameter MoE model (3B active) optimized for code generation and terminal tasks, offering competitive performance and efficiency. Benchmarks show up to 2.8x higher output throughput and 30% lower inter-token latency compared to Devstral Small 2. It's available under Apache 2.0 license for local or on-prem deployment, supporting sovereign AI development.
  13. What Is Model Context Protocol (MCP) | CohereThis post introduces the Model Context Protocol (MCP) as an open standard for connecting AI applications to enterprise systems. It details MCP's client-server architecture, its components (host, client, server, transport mechanism), and the types of features it can expose (resources, tools, prompts). The post clarifies how MCP differs from APIs, RAG, function calling, and agents, positioning it as an integration layer. It outlines common enterprise use cases and emphasizes critical security and implementation considerations such as access control, authentication, authorization, server trust, tool safety, logging, and monitoring. The post also highlights Cohere's North platform as a solution for governed AI workflows.
  14. Introducing Command A+ | CohereIntroduces Command A+, a new open-source MoE LLM (218B total, 25B active parameters) designed for agentic workflows, RAG, and multimodal document processing. It offers improved performance across reasoning, agentic tasks, multimodal understanding, and multilingual capabilities compared to previous Command A models. Key technical advancements include support for 48 languages, 128K context length, low-bit quantization (W4A4, FP8, BF16) for efficient inference on hardware like B200/H100s, speculative decoding for accelerated text generation, and a new tokenizer for improved tokenization efficiency across languages. The model is integrated with vLLM and Transformers frameworks.
  15. Agentic RAG: A Practical Guide for EnterprisesThis post provides a practical guide for enterprises on implementing Agentic RAG, detailing the integration of LLMs with retrieval systems for enhanced knowledge retrieval. It focuses on practical implementation strategies, best practices for document segmentation, semantic coherence, and contextual relevance to improve model processing and retrieval accuracy. The post likely elaborates on the technical aspects of orchestrating agents for RAG, including prompt engineering, tool use, and data handling within an enterprise context.
  16. Command A Reasoning: Control for AI Agents | CohereThis post introduces Command A Reasoning, a system that provides fine-grained control over AI agent behavior and decision-making. It details how agents can be instructed to follow specific reasoning paths, use tools deterministically, and adhere to predefined constraints, thereby enhancing the reliability and predictability of complex AI agent workflows. This system allows for explicit control over agent actions, moving beyond implicit decision-making to a more structured and auditable approach.
  17. Command R Fine-Tuning: High Performance, Fraction of the Cost | Cohere BlogThis post details the fine-tuning process for Cohere's Command R model, focusing on achieving high performance with reduced computational cost. It introduces strategies for efficient data preparation, including data cleaning, formatting, and augmentation specific to fine-tuning tasks. The post elaborates on hyperparameter optimization techniques, such as learning rate scheduling, batch size selection, and regularization methods, to maximize model accuracy and generalization. It also discusses hardware considerations and best practices for leveraging specialized hardware configurations to accelerate the fine-tuning process and reduce overall training expenses.
  18. Introducing Command A: Max performance, minimal compute | Cohere BlogThis post introduces Command A, a new model optimized for high performance with minimal compute. It is an open-source, mixture-of-experts (MoE) model that unifies capabilities from across the Command family, supporting multimodal inputs, tool use, and advanced reasoning. Command A is available under an Apache 2.0 license and offers efficient inference.
  19. How to Create Custom Command ModelsThis post introduces the concept of creating custom Command models by fine-tuning Cohere's existing models. It details the process of preparing custom datasets, selecting appropriate base models, and the technical considerations for fine-tuning, including hyperparameter tuning and evaluation metrics. The goal is to enable users to tailor LLM behavior for specific tasks and domains, enhancing their utility in enterprise applications.
  20. Best Practices for Deploying Language ModelsThis post outlines best practices for deploying language models, covering aspects such as model selection, hardware considerations, inference optimization techniques (e.g., quantization, batching), monitoring strategies, and strategies for ensuring model reliability and scalability in production environments. It emphasizes practical advice for engineers integrating LLMs into their systems.
  21. Cohere Enables Enterprises to Connect Third-Party Datastores to CommandThis post details how Cohere's Command model can now connect to third-party datastores, enabling enterprises to build and orchestrate complex, end-to-end AI agent workflows. This integration enhances the platform's ability to perform advanced reasoning and tool use by allowing agents to access and process information from external data sources.
  22. Running Large Language Models in ProductionThis post details best practices for deploying large language models (LLMs) in production, focusing on practical considerations for running these models efficiently. It covers aspects such as hardware selection (e.g., GPU types, memory), software stack optimization (e.g., inference servers, libraries), and inference techniques (e.g., quantization, batching, speculative decoding) to maximize throughput and minimize latency. The goal is to provide engineers with actionable guidance for operationalizing LLMs in real-world applications.
  23. Updates to the Command R SeriesThis post announces updates to the Command R series, specifically highlighting Command R+ and Command R. Command R+ is an open-source, mixture-of-experts (MoE) model optimized for high-performance agentic tasks with minimal compute overhead, unifying capabilities from across the Command family and supporting multimodal inputs, tool use, and advanced reasoning. Command R offers a more powerful LLM for generation and improved RAG capabilities, including tool use and citations. Both models are designed for efficient inference.
  24. Introducing Tool Use With Command R: Seamlessly Automate Business WorkflowsThis post introduces tool use capabilities for the Command R model, enabling AI agents to seamlessly automate business workflows by interacting with external tools and APIs. It details how Command R can be configured to use specific tools, process tool outputs, and integrate these actions into its reasoning and generation process. The post highlights the practical application of this feature in automating tasks such as data retrieval, system interaction, and complex business process execution.
  25. Using the Command Model in Amazon SageMaker StudioThis post details the technical steps and considerations for integrating Cohere's Command model within Amazon SageMaker Studio, providing a practical guide for deploying and utilizing the model for enterprise AI workloads in a managed cloud environment.