Agentic Workflow Orchestration

7/27/2026

This post details the application of Cohere North's agentic AI platform within the wealth management domain. It showcases specific multi-agent workflows for client intelligence, market research synthesis, and client communication generation. Additionally, it highlights the use of standalone agents for regulatory compliance (MiFID II, Dodd-Frank) and operations intelligence (settlement, NAV reconciliation). The implementation details emphasize connecting to financial systems (OMS/PMS/CRM), defining human-in-the-loop approval boundaries, and tracking time and quality metrics for impact assessment.

7/27/2026

Introduces North Automations, a new capability for orchestrating agentic workflows. Details include turning complex workflows into simple outputs via plain language, scheduling, and branching logic. It emphasizes control through model selection, plan mode, versioning, and testing. Governance features include colleague approvals, usage analytics, and cost monitoring. The post also outlines integration capabilities with existing tech stacks and use cases from Marketing Operations, Customer Success, and Sales teams.

6/25/2026

This post details the integration of Cohere North with the Wiz security platform to automate incident response workflows. It describes the architecture involving a custom Model Context Protocol (MCP) server that exposes Wiz's GraphQL API as eight atomic tools. Specific use cases demonstrated include analyzing toxic combinations for blast radius, performing end-to-end assisted incident response (including report generation and ticket creation), and generating autonomous weekly security posture briefs. Key technical lessons learned regarding prompt placement, API validation, designing for partial failures, and context-aware enrichment are also shared. The full implementation is made available in a public repository.

6/9/2026

Introduced North Mini Code, an open-source agentic coding model (30B total, 3B active parameters) with a 256K context length. Detailed its performance benchmarks for agentic software engineering and terminal tasks, highlighting a 2.8x higher output throughput and 30% advantage in inter-token latency compared to Devstral Small 2. Emphasized its design for sovereign AI and local/on-premise deployment.

5/28/2026

Introduces the Model Context Protocol (MCP) as a standard for connecting AI applications to authorized enterprise systems. MCP acts as a protocol in the application and integration layer, reducing the need for separate integrations. It works via a client-server architecture (host, client, server, transport mechanism) and exposes resources, tools, and prompts. The post details how MCP differs from APIs, RAG, function calling, and agents, and outlines common enterprise use cases in customer support, sales, financial reporting, and IT incident management. It also emphasizes security and implementation considerations for MCP, including access control, authentication, authorization, server trust, tool safety, logging, and monitoring.

5/20/2026

This post introduces Command A+, a new Mixture-of-Experts (MoE) model that consolidates and enhances capabilities from previous Command generations. Key technical contributions include its sparse MoE architecture (218B total, 25B active parameters), improved performance on agentic tasks (e.g., 𝜏²-Bench Telecom, Terminal-Bench Hard, North internal evaluations), multimodal understanding (MMMU Pro, MMMU, MathVista, CharXiv), and multilingual support (48 languages). The post details significant engineering efforts in hardware efficiency through 16-bit, 8-bit, and 4-bit quantizations, leading to higher Output Tokens per Second (TOPS) and reduced Time To First Token (TTFT). It also highlights the use of speculative decoding optimized for MoE architecture and a new tokenizer that improves compression and tokenization efficiency for various languages. The model is designed for private deployment and integrates with open inference frameworks.