If you are on AWS and you are not using the (unfortunately named) Copilot CLI[0], you're opting in to the full complexity of AWS. Copilot fortunately makes most common scenarios waaaay easier to build, setup, and deploy including building staged environments following best practices for AWS (separate accounts; I know, in and of itself kind of a pain in the ass).
It encodes the most common types of application backend deployments into simple patterns and makes it pretty easy to build full and deploy full application stacks (Route 53 -> CloudFront -> S3 (static FE) -> ALB + SGs + TGs -> ECS cluster (backend APIs) -> DBs)
With the Copilot CLI, I find the experience on AWS significantly better and in some ways, more well-rounded than on GCP. GCP's Firebase tooling and CLI comes close, but alas, Firebase does not have the same level of extensibility that Copilot offers by allowing you to include both CDK and CloudFormation as extension points which then allows you to use Copilot to manage a good chunk of your AWS infra with a single, easy to use CLI.
For simple apps, I prefer Firebase on GCP. For more complex apps, I think Copilot on AWS is really, really good. One caveat: ECS is much slower to roll nodes over to new versions compared to Cloud Run. Best I could achieve on it was ~180s for a Blue/Green rollover whereas Cloud Run does this in seconds.
TL;DR: if you are not an enterprise and you are on AWS, your life will assuredly be better with Copilot instead of CDK, CloudFormation, or almost any other solution for deploying to AWS.
And all this to build a website that can be run on a single server, and even not a beefy one :). SQLite + Go is arguably even easier to start with and it can get you surprisingly far
complexity = money for AWS. At this point, I am actually scared I will end up misconfiguring something and stare at a 6 figure bill one morning. I'd much rather use other services or spin up my own than have very delicate discussions with pushy AWS sales guys first thing in the morning.
Not surprising in my opinion.
It is not hard to setup a VPS with whatever you want to deploy nowadays.
99% of the websites out there don't need to be able to infinity scale and it is way cheaper to host your stuff on a VPS, that doesn't have the risk of exploding in cost the way s3 does.
I think AWS is still better in terms of customer support and help that they provide. You really need to keep an eye on the billing console when starting playing around with AWS, the free tier boundary line is pretty thin. for new accounts they have started to give 6 months free plus also some credits. They are trying their best to make it simpler.
AWS is degrading on all fronts. If your a large scale business you almost certainly hit stock-outs during peak periods - the elasticity is a lie. If you are a mid scale its high prices and complexity, and finally if your just a beginner good friggin luck getting anything done now that they gutted free support and put near useless quotas on anything usable for modern AI development.
GCP had second mover advantage in that they got to build a lot of constructs to segment identity and network access in ways that make sense. It's incredibly easy to give a service account access to conduct an action in every folder and project in a GCP org, compared with having to deploy a role to each account in an AWS org. VPCs being global and subnets being regional also make a lot more sense than having them be regional and zonal, respectively, in AWS.
But the support I get from AWS is world class compared to GCP. My AWS account team is always proactively reaching out, not just for potential security risks but also with cost optimization advice.
Yeah, most people say that the best thing about AWS compared to the other 2, is that AWS works. Also AWS runs on AWS, unlike Google's and Microsoft's infra, which is apparently its own bespoke thing.
Quite sure the complexity of AWS is forcing the SME types out - even the most basic operations require reference to chatGPT to find the details buried in the UI. It's gone from being 'good enough' - I mean the UI 15 years ago was basic, but adequate to labyrinthine. We don't have complex needs, they're met by AWS, but the combo of premium pricing and a diabolical UI will force our migration eventually.
Haha, this is weird. I'm old enough to know how to do servers tha old way, setting up nginx, memcached, LAMP stacks and whatever used to be the way to do servers. I am very not much an IaC/Docker/k8s/Amazon whatever thing native. To me AWS always felt like an equally complex way of doing the same things I always used to do.
Now the next generation, who hasn't been indoctrinated into the 'cloud is the way to go, also microservices' mentality, and arriving at a time when the cloud providers aren't flooding everyone with free credits, courses and conferences to push their solutions, to them it looks like another legacy stack.
Or maybe not legacy, but definitely it has a rugged everyday feel as opposed to having the air of 'future tech' around it.
We see the same thing in day-to-day EC2 ops: the pain isn’t “features,” it’s context switching across accounts/regions.
We built a small Windows app that gives one EC2 pane + an AI command bar (plain English → guarded actions with audit).
Not pitching—curious what guardrails you’d require (role scope? dry-run? cost hint before start/resize?).
(Disclosure: I’m the founder of Cloud1.live.)
17 comments
[ 2.5 ms ] story [ 44.2 ms ] threadIt encodes the most common types of application backend deployments into simple patterns and makes it pretty easy to build full and deploy full application stacks (Route 53 -> CloudFront -> S3 (static FE) -> ALB + SGs + TGs -> ECS cluster (backend APIs) -> DBs)
With the Copilot CLI, I find the experience on AWS significantly better and in some ways, more well-rounded than on GCP. GCP's Firebase tooling and CLI comes close, but alas, Firebase does not have the same level of extensibility that Copilot offers by allowing you to include both CDK and CloudFormation as extension points which then allows you to use Copilot to manage a good chunk of your AWS infra with a single, easy to use CLI.
For simple apps, I prefer Firebase on GCP. For more complex apps, I think Copilot on AWS is really, really good. One caveat: ECS is much slower to roll nodes over to new versions compared to Cloud Run. Best I could achieve on it was ~180s for a Blue/Green rollover whereas Cloud Run does this in seconds.
TL;DR: if you are not an enterprise and you are on AWS, your life will assuredly be better with Copilot instead of CDK, CloudFormation, or almost any other solution for deploying to AWS.
[0] https://aws.github.io/copilot-cli/
I got tired of vercel's cold starts and upsell button on every feature.
But the support I get from AWS is world class compared to GCP. My AWS account team is always proactively reaching out, not just for potential security risks but also with cost optimization advice.
Now the next generation, who hasn't been indoctrinated into the 'cloud is the way to go, also microservices' mentality, and arriving at a time when the cloud providers aren't flooding everyone with free credits, courses and conferences to push their solutions, to them it looks like another legacy stack.
Or maybe not legacy, but definitely it has a rugged everyday feel as opposed to having the air of 'future tech' around it.