Mar 2025
170 AI-built apps exposed user data
A scan of 1,645 Lovable projects found 303 open database endpoints across 170 apps — emails, payment details, API keys. Cause: missing row-level security.
Early access: working with the first 10 teams
Your app, running on your own cloud. Production-ready in a week.
You built it with AI. We deploy it into your cloud account, size it for your traffic, and watch it — so it keeps working when users show up.
acme-app · aws · ap-south-1healthy
- api p95 latency
- 182ms
- stable
- db connections
- 71 / 100
- +9% / week
- projected spend
- $184/mo
- budget $250
proposal #12needs approval
- trend
- connections hit limit in ~3 weeks
- change
- add RDS Proxy (pooling)
- cost impact
- +$18/mo
- rollback
- one step, no downtime
What's happening
Everyone can ship now. Almost nobody is operating.
AI builders made creating apps nearly free. The production side — security, uptime, cost — hasn't caught up.
1M+
new projects a week on Lovable (company-reported)
60%+
of new Supabase databases are launched by AI tools (company-reported)
45%
of AI-generated code samples introduced an OWASP Top 10 vulnerability
29%
of cloud IaaS/PaaS spend is estimated as wasted — up for the first time in five years
Jul 2025
An agent deleted a production database
During an explicit code freeze, a coding agent ran destructive commands against live data — 1,206 executive records gone. Dev and prod shared one database.
2025
Developers don't trust the output
46% of developers distrust AI accuracy; 66% say the top frustration is code that's "almost right, but not quite."
Why now — honestly
Can't I just do this with Claude Code and Cursor?
You can do this yourself. Claude Code, Cursor, and an MCP connection to your cloud will write the Terraform, spin up RDS, and wire the pipeline. It will look like it worked.
What a coding agent does
- Writes the infrastructure code
- Provisions the first deployment
- Fixes the error you paste into it
- Does exactly what you ask, when you ask
What it doesn't do
- Watch the stack at 3am
- Notice backup retention got set to zero
- Catch a NAT gateway quietly billing $200/mo
- Ask before it touches production
An agent gives you a deployment. It doesn't give you an operator.
Our hypotheses
What we believe — and how we'll know if we're wrong.
Five bets we're testing with our first 10 teams. We'll publish what we learn, including the ones that don't hold.
- H1
The bottleneck moved from building to operating.
Millions of apps are created every week. Building is solved enough; keeping them up, safe, and affordable is not.
We'll know if
Teams come to us after launch, not before. - H2
Config is commoditised. Accountability isn't.
Agents write YAML for free. What founders lack is someone who watches, decides, and is on the hook when it breaks.
We'll know if
Teams keep paying for operations after setup is done. - H3
Owning the cloud account is the point.
Cloud credits, DPDP/GDPR data residency, investor diligence, and a clean exit all require the infra to be yours — not a platform's.
We'll know if
Credits and compliance come up in the first call. - H4
Approval-first beats autopilot.
After agents deleting prod data, nobody wants unsupervised changes. A human gate with the cost shown up front is what earns trust.
We'll know if
Approve rates stay high because proposals are specific and priced. - H5
Graduation happens at predictable moments.
First paying customers, first security questionnaire, first surprise bill. That's when "it works on Lovable/Vercel" stops being enough.
We'll know if
Inbound clusters around those three triggers.
Where we fit
From prototype platform to your own cloud.
If your app is happy on Vercel or Render, stay there. Move when you need your own account — credits, compliance, cost control, or scale.
You build on
- Lovable
- Replit
- Bolt
- v0
- Cursor
- Claude Code
You host on
- Vercel
- Render
- Supabase
- Railway
You graduate to
- AWS
- Azure
- GCP
- DigitalOcean
How it works
Repo to watched production in four steps.
- 01
Connect GitHub
We review the repo and open a PR with suggested production changes. You decide what merges.
- 02
Get a sized architecture
We recommend Starter, Growth, or Scale based on your real traffic, with the monthly cloud cost shown.
- 03
Connect your cloud
AWS, Azure, GCP, or DigitalOcean via a scoped role. No passwords, no long-lived keys.
- 04
We operate it
Deploy, monitor, alert on trends — and ask before changing anything, with the cost impact.
Next.js
CloudFront / ALB
FastAPI on Fargate
Postgres (RDS)
Example: Next.js + FastAPI + PostgreSQL on AWS
Your account, your bill
Runs in your cloud. Credits apply. Data stays where your compliance needs it.
Code in your repo
OpenTofu and GitHub Actions land as PRs you can read and keep.
Leave any time
Delete the role and we're out. Everything keeps running without us.
- What is Prodsitter?
- Prodsitter is a managed deployment and operations service for AI-built apps. We deploy your app into your own cloud account (AWS, Azure, GCP, or DigitalOcean), size the infrastructure for your traffic, monitor it, and apply fixes only after you approve them, including the cost impact.
- How is Prodsitter different from deploying with Claude Code or Cursor?
- A coding agent can write the infrastructure code and run the first deployment. Prodsitter operates it afterwards: trend-based alerts, backups, cost monitoring, and approval-first changes, with a human accountable when something breaks. An agent gives you a deployment; we're the operator.
- Who is Prodsitter for?
- Founders and small teams whose app works but who don't have DevOps skills in-house. It fits especially well if you built with Lovable, Replit, Bolt, v0, Cursor, or Claude Code, hold cloud credits, or need data to stay in your own account for DPDP, GDPR, or customer requirements.
- Why run on my own cloud account instead of Vercel or Render?
- Your own account means your credits apply, your data stays in the region you choose, costs are raw infrastructure prices, and there's no platform lock-in. If your app is happy on Vercel or Render and none of that matters yet, stay there.
- How long does it take to get production-ready?
- Most apps are production-ready in about a week: a repo review and suggested pull request first, then the sized architecture is deployed into your cloud with health checks, backups, alerts, and budgets in place.
- How much does Prodsitter cost?
- During early access, plans start at $1,500 one-time setup plus $499/month for operations. Cloud usage is billed separately by your provider to your own account — roughly ~$25 to $50/mo for a small app.
Early access: working with the first 10 teams
Get your app production-ready
30 minutes. Bring your repo. You'll leave with a deployment plan even if you don't work with us.