After a comprehensive three-month evaluation of Lindy as a potential replacement for our existing Make (formerly Integromat) automation workflows, I have concluded that a reversion to our prior platform is the most financially and operationally sound decision. While Lindy presents an intriguing proposition with its agent-centric approach, a detailed breakdown of its consumption-based pricing model reveals several critical cost control vulnerabilities that are untenable for predictable budgeting at scale.
My primary concern stems from the opaque and multiplicative nature of "AI Agent Steps." Unlike Make's clear pricing per operations within a scenario, Lindy charges for each step an agent takes, which includes internal reasoning, tool executions, and data processing. This creates significant unpredictability. For instance, a single customer onboarding automation that involved parsing an email, querying a CRM, and generating a document consistently consumed between 47 and 129 steps per execution in my testing, depending on the complexity of the email body. The variance was not correlated with any user-controlled parameter but rather with the LLM's chain-of-thought process, which is a black box from a cost perspective.
A granular cost comparison for a representative workload—processing 5,000 customer support emails per month—illustrates the disparity:
* **Make:** Costs are based on executed operations. Our optimized scenario used ~8 operations per email (router, parsers, database updates, HTTP requests). At our volume tier, this resulted in a predictable monthly cost of approximately $129.
* **Lindy:** Using a similarly capable "Email Processing Agent," the step count per email fluctuated between 35 and 90. At Lindy's published rate, the monthly projection ranged from $175 to $450. This 250% potential variance eliminates any possibility of accurate forecasting.
Furthermore, the platform introduces several ancillary cost drivers that function as "fee multipliers":
* **File Storage & Processing:** Temporary file storage during agent execution and per-page charges for document processing (e.g., PDFs, DOCs) create additional cost layers absent in Make's flat operational model.
* **Unclear "External Integration" Costs:** While API calls to third-party services are a cost factor in any platform, Lindy's bundling of these with agent steps makes it difficult to isolate and optimize the most expensive components of a workflow.
* **The "Unlimited" Agent Concurrency Trap:** While promising on the surface, unlimited concurrent agents simply accelerate the consumption of the unpredictable step pool, leading to potential cost spikes during peak loads rather than smoothing them out.
In conclusion, for organizations practicing FinOps or those with strict cloud budgeting mandates, Lindy's pricing model represents a fundamental shift from predictable, unit-based costing to a variable, effort-based costing that is exceedingly difficult to govern. The intellectual appeal of AI agents is undeniable, but until the cost per business transaction can be modeled and constrained with high confidence, it presents an unacceptable financial risk. My spreadsheets—and my CFO—demand a return to predictability.
-- Liam
Always check the data transfer costs.