Cloud Migration Checklist for Startups

Cloud migrations we run from Agentic Academy Labs fail for boring reasons: nobody inventoried cron jobs, there was no rollback owner, and cost alerts went on after the first invoice. This checklist is what we walk through with startups moving to AWS, GCP, Azure, or Vercel, including teams whose product already sits on Next.js, Bun, PostgreSQL, and Flutter. Sikar-based builds and remote clients get the same four phases: assess, plan, migrate, optimize.

1. Assess

• Inventory apps, databases, cron, file stores, DNS, and the 'temporary' VPS someone SSHs into.
• Measure current spend, peak traffic, and how many minutes of downtime a founder will actually accept.
• Note data residency and encryption needs. Indian fintech and health tools often cannot shrug this off.
• Flag licenses, VPN, on-prem printers, and the Flutter app's hard-coded API host.

2. Plan

• Pick the target cloud and landing zone (accounts, VPCs, projects) for skills you have, not a conference talk.
• Per service: rehost, replatform, or rewrite. Rewriting everything is a different project; see migrate a monolith to Next.js and PostgreSQL.
• Design database cutover with replication and a validation query you can run twice.
• Write a rollback plan with a named owner and a clock. 'We will figure it out' is not a plan.
• Estimate run-rate after migration. Cutover cost is the small number.

3. Migrate

1. Stand up infrastructure as code (Terraform or the provider's native equivalent). Click-ops is how internships lose the staging env.
2. Containerize when it reduces drift. Keep a simple Bun service simple.
3. Move data with checksums. Dual-write when the product cannot freeze writes.
4. Wire CI/CD, secrets, logs, metrics, and alerts before DNS moves.
5. Dress rehearsal, then cutover, then a written smoke list: login, pay, Flutter push, cron, backup restore.

4. Optimize

• Right-size instances in the first two weeks. Demo-sized boxes either melt or waste money.
• CDN and cache for static and hot GETs. Next.js on Vercel already wants this thinking.
• Review idle disks, leftover IPs, and abandoned staging clusters weekly for a month.
• Hand over runbooks so the internal team can operate without Agentic Academy Labs on the call.

What we do in week 1

1. Dump a living inventory spreadsheet. If it is not in the sheet, it will not move.
2. Turn on (or screenshot) current hosting spend. You cannot claim savings without a baseline.
3. Restore last night's backup into a scratch database. If restore fails, migration has not started.
4. Agree the downtime window in writing with the person who owns customers, not only engineering.
5. Put budget alerts on the target cloud before any production traffic.

Risks we have already hit

Four to sixteen weeks is the band we quote after inventory, not before. A marketing site on Vercel is the short end. A monolith with years of cron and file storage is the long end. Pair this checklist with cloud migration services and data engineering when pipelines are the real blocker. Schedule a call if you want Agentic Academy Labs to run the rehearsal with you.