Alright, so I gave OpenClaw's cost estimation a spin for a new AWS dev environment last quarter. The promise of "see your bill before you build" is exactly the kind of thing I jump on—and inevitably end up disappointed by.
My use case was pretty standard: a couple of EC2 instances (one dev, one staging), an RDS Postgres instance, S3 for some static assets, and a CloudFront distribution. I fed OpenClaw my Terraform plan output. The initial estimate came back suspiciously... round. Like, *suspiciously* neat. It projected $287/month.
The actual bill landed at $412 for the first month. Not catastrophic, but a miss of about 30%.
Digging into the variance, a few clear gaps emerged:
* It completely missed data transfer costs between AWS services in different AZs (our RDS read replica traffic). Not a huge line item, but it added up.
* The S3 cost model seemed based on storage alone, ignoring GET/PUT requests, which for our staging environment were higher than anticipated.
* The tool assumed default, on-demand pricing for EC2, but we used a mix of Reserved Instances and Spot for the staging box. OpenClaw didn't have a clean way to incorporate that without manual overrides, which defeats the purpose.
So, the accuracy felt like a first-pass approximation. Useful for catching a truly egregious, budget-busting design? Maybe. But for actual forecasting, especially with any non-vanilla pricing constructs, it was off by a meaningful margin.
Has anyone else run it in production for a few cycles? I'm curious if:
* The estimates get better with more historical data from your linked accounts, or if it's purely plan-based.
* Their support for Azure/GCP is any more accurate, or if they're stretching themselves too thin.
* You've found a reliable way to model savings plans or committed use discounts within their system.
I'm already looking at the next tool in the queue, but I'd love to be proven wrong.