The most popular automation platform charges per task. That sounds reasonable until you build something that runs 50,000 tasks a month.

Per-task pricing is a usage-based model. You pay for what you consume. At small volumes, the numbers are low and the friction of switching feels higher than the cost of staying. By the time volume is high enough to feel the pain, you’ve built automations that would take weeks to migrate.

This is the per-task tax: the compounding cost of a pricing model that grows proportionally with the value you’re getting from the tool.

The Alternative Exists

The self-hosted alternative to the major automation platforms is fully free for compute costs beyond your own infrastructure. A community of tens of thousands of developers uses it. It has ~70 built-in nodes specifically for AI and LLM workflows — more than any comparable hosted platform.

The comparison is stark: per-task pricing that punishes volume vs. fixed infrastructure cost that doesn’t care how many automations you run.

For someone building a single lightweight workflow that fires a handful of times a day, the hosted platform is probably fine — the convenience of not managing infrastructure is worth a few dollars a month. But for someone building an AI-powered system that processes thousands of items, the math changes completely.

Infrastructure Thinking

There’s a broader principle here about fixed vs. variable costs in software tooling.

Variable costs (per-task, per-API-call, per-seat above a threshold) feel acceptable when you’re small. They’re designed to feel that way. Vendors know that usage correlates with value — when you’re running more tasks, you presumably have a business that can afford them.

The problem is that it’s not always true. Sometimes high volume means the automation is working but the margins are thin. Sometimes the cost structure of the underlying business doesn’t support unlimited growth in tooling spend. Sometimes you’re building something for yourself or a small team where the per-task logic never made sense.

Fixed costs behave differently. You pay the same whether you run 100 automations or 100,000. This feels wrong at small scale — you’re “paying for capacity you’re not using” — but at medium and large scale it’s a significant advantage.

The Migration Problem

The reason per-task pricing is so durable as a business model isn’t that it’s the best deal for customers. It’s that switching is hard.

Automations built on one platform use that platform’s logic, connectors, trigger system, and authentication patterns. Migrating to another platform means rebuilding, not just copying. The automations you care most about — the ones you’ve been running reliably for months — are the ones you’re most afraid to touch.

This is lock-in by accumulation. Not a contract, not a technical barrier. Just the weight of working systems.

The Upfront Work Tradeoff

Self-hosted tools require setup. You need a server, a deployment, maintenance, and the occasional troubleshooting session. This is real work with a real cost.

The calculation is: do this upfront work once, or pay a variable tax indefinitely. For low-volume personal automations, the hosted approach often wins. For anything with meaningful volume or that’s meant to run for years, the math usually favors the work upfront.

The broader lesson: the infrastructure decisions you make when something is small set the cost structure for when it’s large. A pricing model that’s 10x more expensive at scale looked cheap when you were just getting started.

Pick the one that fits where you’re going, not where you are.