Automation & agents

Zapier: the largest connector catalogue and the gentlest learning curve, so your first automations succeed quickly. Make is the value pick once volume grows, with cheaper high-volume execution and more visual control of complex flows. The pattern that works: prove three automations on Zapier, then evaluate whether your usage profile justifies migrating the heavy ones to Make.

n8n: free to self-host with unlimited executions, which changes the economics entirely at volume and keeps data inside your own infrastructure, the deciding factor where residency or compliance rules apply. The trade is operational: you run and update it yourself. Teams without that capacity should weigh managed n8n against staying on cloud platforms.

Lindy, Relevance AI, Gumloop and Bardeen are the leading no-code agent builders: describe the workflow, connect the tools, and the agent handles multi-step work without code. The category is young and moving fast, so pilot one real workflow before standardising on any of them, and check the credit economics at your expected volume; agent runs consume credits quickly.

CrewAI and LangChain, with LangGraph as the latter's agent-orchestration layer, are the reference open-source frameworks: full control over models, tools, memory and orchestration, at the cost of building and operating it yourself. Choose frameworks when agents are product infrastructure; choose no-code builders when they are internal workflow glue. Many teams prototype no-code, then graduate.

Per-task and credit billing balloons at scale: an automation that runs every few minutes multiplies quietly into serious monthly spend. Before committing, model your real event volume, not the demo's. Batch where possible, trim polling frequencies, and revisit self-hosted n8n when volume-based pricing starts dominating the bill; unlimited executions change the calculation entirely.