Dylan Mikus
1.2K posts
@dbmikus
building @amikadev - infra for your software factory | YC S22 + F25, CMU CS + LTI
Your entire SDLC should be automatable with AI agents. But just like there's no one-stop dev workflow, there's no one-stop agent. That's why we're building amika.dev: infra for every company to build their own software factory, customized to them.
Agents are good at bash. Bash is not good for agents. We should cut our losses and restart now before it is too late.
Been spending a lot of time thinking about the future of SaaS. At @moment_dev, almost all our SaaS consumption has moved to little internal tools built around APIs. This video is a cost tool I built on top of the AWS + Vantage APIs, Claude completely 1-shotted it, and it's totally replaced the actual UI for us. I admit I don't really know what to make of this. A lot of SaaS UI is trying to strike a balance between being flexible and usable, which makes sense when software is expensive to make. But now it's like $1 to create a tool that does exactly what you want. What is the equilibrium state for software now? I don't know that I buy the (say) Linear argument that everyone JUST wants software that works. A lot of software that would deliver value just doesn't get written because it's not worth the effort, but if the effort is 100x lower, then it might actually suddenly be worth it. I'm sure there are lots of serious and important workflows that will still get dedicated tools, but that's not the software that's getting replaced, really. Think about what this cost tool replaces. The main thing it does is, when you use the full-text search feature, it narrows the forecast and current cost to the matching resources. As far as I know, there is no feature like this in any other tool, or at least, not one that is this fast. It does replace OUR USE of AWS Cost Explorer and Vantage. But at a larger org this tool would actually augment the use of these other tools, unless someone built a suite of similar tools for the finance department, and leadership, and so on. And it might not be worth solving delivery and authn/authz, data access issues, and so on. This is one of the main reasons we set out to solve delivery, authn, and authz for apps like this. If those things are essentially trivial, I think it's easier to see clearly which of those things are actually plausible to build for the rest of the org. I don't think anyone knows right now.
Devs love to dev so everyone and their mom is building their own version of Ramp’s Inspect. It’s a *ridiculous* amount of work though. You have to build a huge amount of features to just have a simple functional agent orchestration pipeline (that’s actually good with browser testing, cloud sandboxes that actual work, surface area across Linear and Slack, PR review, Security Review, etc, etc, etc) I know because I built several versions of this. Got sick of it and just decided to pay for Devin. Never been happier or freer to just ship a ton of features and bug fixes. You’re absolutely wasting your time if you’re not buying your agent orchestration off the shelf now. Just pick a platform and double down - stop switching - all the frontier models are equally capable now. Focus, double down and build. Don’t build the tool - use the tool. Stop playing with this stuff and just get back to work.
