Notion Adds Custom AI Agents, Citing $500M+ ARR

Notion's custom agents, launched on February 24, 2026, are proactive, trigger-based automations, a significant evolution from the reactive, on-demand personal AI assistant. Unlike the personal agent that requires a manual prompt, these custom agents can be configured to run based on schedules, database changes, or events in connected apps like Slack, operating autonomously in the background. This shift enables the creation of an autonomous operational layer, with Notion itself running over 2,800 internal agents—more than its number of employees. The architecture behind such agentic systems often involves multi-agent patterns where complex workflows are broken down and assigned to specialized agents. These agents then interact to achieve a larger goal, a design that improves modularity and scalability over a single monolithic agent. Open-source LLM orchestration frameworks like LangChain and LlamaIndex provide the tooling to build these systems, offering modules for chaining LLM calls, managing tools, and connecting to data sources. LangChain is known for its flexibility in creating complex agentic workflows, while LlamaIndex specializes in retrieval-augmented generation (RAG) for data-intensive applications. For a Staff-level engineer, influencing without direct authority is key, a core tenet of the Principal Engineer role which focuses on setting technical direction and mentoring. This involves making strategic decisions that guide projects and maintaining high technical standards across teams. The backend architecture for these AI platforms must be designed for scale from day one, often using a microservices or event-driven approach to handle compute-intensive and asynchronous tasks. An API-first mindset is crucial, with well-documented, secure, and versioned endpoints forming the backbone for agent interactions. In insurtech, AI agents are revolutionizing core processes by automating data-heavy tasks in underwriting and claims processing. AI can analyze vast datasets to assess risk, detect fraud, and even handle initial claims documentation, reducing processing times from days to minutes. This allows human underwriters and claims adjusters to focus on more complex, high-judgment decisions rather than administrative work. Venture capital is flowing into AI-native insurtech, with investors backing companies that use AI to build more efficient underwriting and claims systems. Global InsurTech funding saw a rebound in late 2025 and early 2026, with significant rounds for companies building AI-powered "operating systems" for insurance. This trend reflects a broader market shift where AI is not just a feature but the core infrastructure, a pattern seen as venture funding for UK AI startups reached a record $7.9 billion in 2025. For developers building these systems, open-source workflow automation tools like n8n, Windmill, and Activepieces offer powerful alternatives to proprietary platforms. These tools provide visual interfaces and extensive integrations for creating complex automation rules, often with the flexibility of being self-hosted for greater security and control. Many are also incorporating AI capabilities, allowing for the creation of intelligent, code-based automation logic using languages like Python and TypeScript.