I built an AI DevOps platform in 10 weeks that deploys to AWS, Google Cloud, and Azure from a single prompt. Here's everything I learned.

I've been a developer for most of my life. I've shipped products, worked on teams, and spent more hours than I want to admit staring at cloud consoles, writing YAML, and debugging CI/CD pipelines that broke at 2 AM.

At some point I stopped asking "how do I deploy this?" and started asking "why is deploying this still this hard in 2026?"

That question became NEXUS AI.

The problem I kept running into

Every time I started a new project — or helped someone else start one — the same wall appeared: you have a working app, but getting it to production takes days.

Not because the app is complex. Because the infrastructure glue between "working locally" and "running in production" is genuinely miserable:

• - Pick a cloud provider. Set up an account. Figure out IAM.

• - Write a Dockerfile (or Google one and paste it in and hope).

• - Set up a container registry.

• - Write a deployment pipeline. Choose a CI/CD tool. Write more YAML.

• - Configure environment variables. Store secrets somewhere that isn't a .env file committed to git.

• - Set up health checks, custom domains, TLS.

• - Do this again for staging. And for the next project.

The cost of this isn't just time. It's the projects that don't get started because the infra tax is too high. It's the solo founders who would ship great products but bounce off the deployment wall and give up.

I wanted to make that wall disappear.

What I built

NEXUS AI is an AI-native DevOps platform. You describe what you want to deploy, and it handles the rest — from prompt to a running container on AWS, Google Cloud, or Azure in under 5 minutes.

It's not just a deployment shortcut. It's a full deployment platform:

• - Application Generation Engine — describe your app, get a working codebase

• - Multi-cloud deployment — AWS App Runner, Google Cloud Run, Azure Container Apps, or NEXUS AI managed Docker. One CLI, all four.

• - Secrets vault — AES-256-GCM encrypted secrets injected at runtime. No plaintext anywhere.

• - Custom domains — add your domain, NEXUS AI handles DNS verification and TLS

• - Replica scaling — 1 to 10 replicas, live, no redeploy

• - One-click rollback — back to the last working state in 60 seconds

• - RBAC — roles, scoped access tokens, audit logs

• - MCP integration — 37 tools so Claude and other AI agents can deploy, scale, and rollback via the API

The pricing is $29/mo for Starter, $149/mo for Pro, and Enterprise for larger teams and regulated industries (healthcare, fintech).

How I built it in 10 weeks

I started on February 15, 2026. The first commit was literally called "push to prod" — I deployed an empty shell to make sure the deployment pipeline worked before I wrote a single feature.

Week 1–2 was the core deployment engine: Docker-based deployments, the secrets vault, the basic CLI. I didn't try to integrate all three clouds at once — I started with the managed Docker path and got it fully working.

Week 3–4: I added Google Cloud Run. This was harder than expected — Cloud Run's IAM model is opinionated and the SDK has rough edges. I burned two days on a subtle build artifact issue that turned out to be a regional artifact registry permission problem. That kind of thing.

Week 5–6: AWS App Runner, then Azure Container Apps. By this point the abstraction layer was solid enough that new providers took 2–3 days each instead of a week.

Week 7: RBAC, audit logs, the organization model. I almost shipped without proper access control "to move faster" and then thought about it for 30 seconds and didn't do that.

Week 8: Custom domains, DNS verification, TLS. Easier than expected — Cloudflare's API is well-documented.

Week 9: MCP integration. This was the most fun part. Wiring up 37 tools so an AI agent can manage your entire deployment lifecycle is a genuinely new thing and I wasn't sure if anyone would care. They cared.

Week 10: Launch prep, blog, SEO, the public-facing pages. Also 106 git commits in 10 weeks, which comes out to a commit every 16 hours on average. I did not sleep enough.

What worked

Shipping fast and in public.

Every feature I shipped, I posted about. The MCP integration got the most traction — there's a specific audience of AI builders who want their Claude agents to be able to manage deployments, and they found it immediately.

Starting with the compliance angle early.

I almost left RBAC, audit logs, and the secrets vault for "later." I'm glad I didn't. About 30% of the conversations I have with potential customers start with "do you have audit logs?" or "how do you handle secrets?" — especially anyone in healthcare or fintech. Having real answers to those questions on week 7 instead of "we'll add that" made a difference.

Multi-cloud from day one.

The instinct is to start with one provider and add more later. But "deploy to your own AWS account" is a meaningful differentiator against Railway, Render, and Fly.io, which run on their own shared infrastructure. Leading with multi-cloud opened conversations that wouldn't have happened otherwise.

The CLI as the primary interface.

I built the dashboard second. The CLI came first. Developers trust CLIs. A CLI that works well is a product that works well — you can't hide bad UX behind a polished UI.

What didn't work

Underestimating cloud provider edge cases.

Every cloud provider has at least one thing that doesn't work the way the documentation says it does. Azure Container Apps had a networking gotcha that cost me three days. AWS App Runner's build logs take 90 seconds to appear after a deployment starts. Google Cloud Run has a regional artifact registry restriction that the SDK doesn't surface clearly. None of these are blockers — but they're the kind of thing that turns a 2-day feature into a 5-day feature.

Trying to write marketing copy myself at midnight.

The product copy I wrote at midnight is uniformly worse than the copy I wrote in the morning. I now have a brand voice document (`marketing-context.md` at the root of the repo, if you're curious). Having written rules for "what NEXUS AI sounds like" has made every piece of content faster and better. I should have done this in week 2, not week 9.

Building the blog feature before publishing blog posts.

I shipped a full blog system — posts, comments, admin management, RSS feed — before I had a single post to publish. Classic developer mistake: building the tooling before the content. The blog is great. I should have published 10 posts to a static page first to validate that anyone wanted to read them.

The most important thing I learned

The hardest part of building a DevOps product isn't the infrastructure integration. It's the trust problem.

Developers are deeply skeptical of tools that touch their deployments. Rightfully so — a bad deploy can take down production. A bad secrets implementation can expose credentials. A bad access control model can let the wrong person do the wrong thing.

Every feature I built had to answer the question: "how does this make a developer trust NEXUS AI more?" The audit log isn't just a compliance feature — it's evidence that the platform does what it says. The AES-256-GCM secrets vault isn't just encryption — it's a commitment that I thought about this carefully. The scoped access tokens aren't just a permissions model — they're a signal that I understand the blast radius problem.

Trust is the product. Everything else is how you earn it.

What's next

A few things on the roadmap that I'm actively building:

- Auto-scaling — scale to zero during idle periods, scale up on load automatically

- Database attachments — provision and attach a managed Postgres or Redis alongside a deployment

- Usage-based billing — pay for actual resource consumption, not just a flat monthly seat

If you're building on NEXUS AI or thinking about it, I want to hear from you. What's the one thing that's missing? What's the one thing that made you choose a different tool?

Stack, for the curious

• - Backend: Express + TypeScript + Prisma + PostgreSQL + Redis

• - Frontend: React + Vite + TailwindCSS

• - CLI: TypeScript + Commander

• - Infra: Deployed on NEXUS AI itself (yes, it runs on itself)

• - Cloud integrations: AWS SDK v3, Google Cloud SDK, Azure SDK

• - Auth: JWT + bcrypt, OAuth (GitHub, Google)

• - Secrets: AES-256-GCM, key derived from env

• - MCP: 37 tools wired to the full REST API

If you made it this far — thank you. Building in public is weird and occasionally uncomfortable but it's the right way to do this.

If you want to try NEXUS AI: [nexusai.run](https://nexusai.run). Starter is $29/mo. Free tier available to kick the tires.

Happy to answer any questions in the comments.

— Saif