Hey folks! I was helping our marketing team automate a simple drip campaign trigger this week and it sparked a real debate. We were looking at a workflow where a new lead in HubSpot triggers a series of personalized emails and Slack notifications. The classic go-to would be Zapier or Make, right? But our dev team was already prototyping with LangGraph for some LLM chains and suggested using it for this.
My immediate reaction was: "For a simple, linear sequence of API calls, isn't that massive overkill?" 🤔
So I ran some quick numbers, focusing on the **cost and complexity** angle.
**For the Zapier route:**
* **Build time:** ~1 hour for a non-engineer.
* **Monthly cost:** Zapier's free tier covers 100 leads/month. The first paid tier is ~$20/month for 750 tasks. Our flow uses about 5 tasks per lead.
* **Maintenance:** Almost zero dev overhead.
**For the LangGraph (hosted on AWS) route:**
* **Build time:** ~1 day for a developer familiar with the framework.
* **Monthly cost:** Even with a minimal Lambda function (powered by the graph) and API Gateway, we're looking at ~$5-10/month in AWS costs **plus** the mental overhead of monitoring. The bigger cost is the dev time to build and maintain.
* **Maintenance:** You now own the code, the deployments, and the error handling.
**My verdict:** For a straightforward, deterministic workflow like this, LangGraph is **overkill**. The savings just aren't there. You'd need a much more complex, stateful, or logic-heavy process (like a dynamic customer support triage with LLM decisions) to justify its power.
Where LangGraph shines is when your workflow has **cycles**, **conditional branches based on LLM outputs**, or needs a persistent state that evolves. A drip campaign is just a straight line.
Has anyone else compared these tools for simple automation? Did you find a complexity threshold where LangGraph started to make sense?
Right-size everything
1. I'm a principal engineer at a mid-sized e-commerce company, where our stack is a mix of serverless and containerized workloads on AWS, and I've directly built, migrated from, and torn down more automation workflows than I can count.
2.
* **Real monthly cost beyond list price:** Zapier's $20 tier for 750 tasks is real, but you'll hit it fast. That's 150 leads at 5 tasks each. At 500 leads/month, you're already in the $50/month bracket. A Lambda processing the same volume, with a 10-second runtime and 128MB memory, would cost $0.08. The $5-10 AWS estimate is high; the real cost is the $1/month for the API Gateway HTTP API and pennies for Lambda.
* **Dev and maintenance overhead:** Zapier's "almost zero dev overhead" is correct until you need a custom data transform or error handling it can't do, then you're stuck. LangGraph means a dev owns it forever. You've now added a CI/CD pipeline, logging, and alerting to your ops load for a marketing drip. That's a ~5-10 hour/month recurring tax for "simple."
* **Where the custom build clearly wins:** This is only if your "simple trigger" is a lie. If this flow needs to branch based on LLM logic, call internal APIs, or handle complex state retries, Zapier will crumble. LangGraph is overkill for a linear sequence, but necessary for a state machine.
* **The hidden scaling trap:** Zapier's concurrency limits are brutal. On the $20 plan, you get maybe 1-2 tasks running at once. If 50 leads come in simultaneously, your drip is now a multi-hour queue. Your Lambda would scale to thousands in seconds for no extra cost. For time-sensitive welcome drips, this alone kills the Zapier approach.
3. For a truly linear API sequence, use Zapier and pocket the dev time. If your marketing team ever whispers "what if we wait 3 days and then check if they opened the last email," you need a state machine, and then I'd pick a simpler AWS Step Functions setup over LangGraph for this. Tell us your peak leads per minute and whether the "series" has any conditional pauses or logic.
pay for what you use, not what you reserve