Cohere's North Automations Lets Employees Build Multi-Step AI Workflows Without Code

Cohere has launched North Automations, a new layer on top of its enterprise AI platform that lets employees build and run multi-step automated workflows using plain language, no engineering background required. The feature is available immediately to all existing North customers, and it represents Cohere's clearest move from model provider to owner of the full enterprise AI stack.

The coordination problem

The pitch for AI agents has always been compelling on paper: autonomous software that can reason, take action, and hand off tasks across systems. Inside most enterprises, the reality has been messier. Deployments tend to produce isolated, task-specific bots that handle one narrow job well but cannot coordinate across teams or systems.

Three structural problems drive this gap. First, one-off agents solve narrow tasks rather than end-to-end workflows that reflect how employees actually work. Second, fragmented deployments make it nearly impossible to govern AI usage at scale. Third, running every step of a workflow through the same frontier model inflates costs fast, making per-step model routing a practical necessity rather than a nice-to-have.

Gartner analysts frame this as a structural problem. "Current AI agent implementations are mostly built with individual, task-specific agents that provide incremental benefits, but create a value gap for enterprise-scale AI adoption," write research analysts Anushree Verma and Alastair Woolcock in the Agentic AI Hype Cycle 2026 report. Gartner pegs the agent orchestration opportunity at $550 billion. Cohere is positioning Automations as a direct answer.

What Automations does

Agent orchestration, coordinating multiple AI agents to complete a larger, multi-step business process, is the core concept. North Automations provides that coordination layer. Its three headline capabilities are:

• Plain-language workflow design: Describe goals in plain language to connect your tech stack, set run frequency with scheduled triggers, and produce an auditable execution path with loops and branching logic.
• Per-step model selection: Assign a different model at each workflow step to balance cost against performance. Plan mode lets you review and edit your approach before building, and versioning tracks changes over time.
• Governance and oversight: Flag key moments for human approval, analyze automation usage over time, and monitor token consumption to manage costs.

Per-step model routing deserves attention from anyone thinking about cost. Rather than running every stage of a ten-step workflow through a frontier model, you can assign a cheaper, faster model to simple extraction tasks and reserve heavy compute for reasoning-intensive steps. That shift alone can change the unit economics of running complex automations at scale.

Integration depth

Automations sit inside the broader North platform, which connects to Gmail, Slack, Salesforce, Outlook, and Linear, and integrates with any Model Context Protocol (MCP) server. MCP is an open standard that lets AI agents communicate with external tools and data sources in a standardized way, functioning as a universal adapter for plugging AI into existing software. Agents can send emails, update CRM records, trigger workflows, or reach specialized industry applications without custom development work.

For teams that need department-specific tooling, agents can be configured for HR, finance, legal, and operations. All activity produces full data traceability and audit-ready logs.

Real workflows, not demos

Cohere shared use cases from their own internal teams, which gives a grounded sense of what this looks like in production:

• The marketing analytics team connected North to BigQuery so it translates plain-language questions into SQL, runs the analysis, and returns the answer, removing the analyst bottleneck entirely.
• The customer success team built a daily tracker that pulls from HR systems, Slack, Salesforce, and other sources to generate a morning briefing for managers before 1:1 meetings.
• The sales team built a pre-call prep automation that aggregates data from Salesforce, Slack, Gmail, Notion, and external market research into a consolidated presentation.

These involve real multi-system data pulls, conditional logic, and scheduled execution, exactly the kind of workflows that previously required a data engineer or a custom integration project.

Where Cohere is headed

Cohere surpassed its $200 million ARR target in 2025, reaching $240 million with quarter-over-quarter growth exceeding 50% throughout the year. A $600 million funding round from Nvidia, Salesforce, and AMD pushed the company's valuation to $7 billion in September 2025. CEO Aidan Gomez has indicated an IPO may come soon, and a product like Automations makes the enterprise platform story considerably more compelling to public market investors.

The competitive landscape is crowded. Microsoft, Salesforce, and ServiceNow all have workflow automation plays, and the global AI workflow automation market was valued at $14.2 billion in 2025, projected to reach $184.6 billion by 2034 at a 33.7% CAGR. Cohere's differentiator is deployment flexibility: North runs privately or fully air-gapped, meaning it can operate with no internet connection, which matters enormously for regulated industries like banking, healthcare, and government.

The platform already runs inside the Royal Bank of Canada, Dell Technologies, and South Korea's LG CNS. Cohere and RBC jointly developed North for Banking, a generative and agentic AI workspace tailored to financial services firms. Automations gives those customers a faster path to deploying complex, multi-step workflows without building custom orchestration infrastructure from scratch.

For teams already on North, this is a meaningful capability unlock. For teams evaluating enterprise AI platforms, it raises the bar on what "agentic" should mean in a production environment: a governed, auditable, multi-step automation layer that non-technical employees can actually operate.