Why We Built Infralyst

Every infra team we've worked on has burned engineering hours on work that shouldn't still be manual. Checking dashboards, resizing instances, opening PRs to change a number in a Terraform file. It's not hard work. It's just work that never stops, never gets prioritized, and never gets done until someone panics about the bill.

We built Infralyst because we think that's absurd. In 2026, a lot of companies could run everything they need on a single dedicated server. The fact that cloud infrastructure still requires this much manual attention to avoid overpaying isn't a technical problem. It's an incentive problem.

Cloud providers are merchants of complexity

Hundreds of instance types, pricing pages that need their own documentation, savings plans you need a spreadsheet to evaluate. The more confusing it is to navigate, the more you overpay. And the more you overpay, the better it is for the provider.

Cloud providers have built-in cost recommendations. They've gotten better over time, but they're still pretty limited. Short look-back windows, narrow coverage, thresholds that miss a lot of real waste. They tell you what to change but not how to change it in your actual infrastructure code.

We don't think the providers are going to fix this. Every dollar you save is a dollar off their top line. They'll invest enough in cost tooling to check a box, but they're not going to build something that seriously cuts into what customers overpay. It's not some conspiracy, it's just how incentives work.

Third-party tools haven't really solved it either. Most of them are dashboards. They'll show you that your instance could be smaller. OK, but then someone on your team has to figure out the right Terraform change, check it won't break anything, open a PR, get it reviewed, deploy it. Do that for one resource and it's fine. Do it for fifty across a bunch of services and it's a whole project competing with everything else.

Some tools take a different approach and modify your infrastructure directly. That gets things done, but now your Terraform says one thing and reality says another. If your team treats infrastructure as code seriously, that's a problem.

The same cycle, everywhere

Here's what happens. Someone picks a size for a resource when a service launches. It works, traffic is fine, nobody revisits it. Two years later it's running at a fraction of capacity, and the company is paying for compute it never needed.

Everyone agrees cloud cost management should be an ongoing thing. Companies try. Sometimes an infra team owns it, sometimes there's a cost guild, sometimes individual dev teams are supposed to manage their own resources. It never really sticks because people are shipping features, handling incidents, doing the stuff that actually feels urgent. So costs just drift.

Then every year or two someone looks at the bill, panics, and kicks off a cost-cutting sprint. Things get better for a bit, and then the drift starts again. We've been through this cycle at multiple companies. At some point we stopped thinking "someone should fix this process" and started thinking "this shouldn't be a process at all."

The utilization data is there. The analysis is straightforward. The code changes are predictable. This is the kind of thing that should run in the background, like monitoring or alerting. Not something that depends on someone remembering to go check a dashboard.

What we built

Infralyst connects to your AWS accounts with a read-only role and looks at utilization for supported resources. When it finds something clearly oversized, it surfaces a recommendation. You get notified in Slack, and from there you can generate a Terraform PR or dismiss it. Nothing changes until your team reviews and merges the code. This only became possible recently because LLMs got good enough to reliably generate clean Terraform from well-defined inputs. Once that worked, there was no reason this should still be a manual process.

Why we care

We think companies spend way too much on the cloud, and the system is set up to keep it that way. The providers benefit from the waste. The manual processes teams build to fight it don't survive contact with real work. And the third-party tools that exist mostly just give you more dashboards to ignore.

The answer should be boring. Automated analysis, conservative recommendations, and a notification in Slack when there's something worth looking at. If it makes sense, you click a button and get a Terraform PR in a few seconds. No new dashboards to check, no spreadsheets, no FinOps hire. Just code changes you can review, in the repos you already use.

Engineers should be building things. We'd rather spend our time making sure they can.