A modernization that requires a feature freeze is a rewrite with better slides. The freeze slips, the business keeps asking, and you merge six weeks of product work into a branch nobody remembers.
The job is to change the platform under a team that is still shipping. I have done that as a greenfield portal (Crowbar Logistics on Next.js and AWS Lightsail) and as a UI modernization on a site that already has more than 300,000 visits a month. The tactics differ. The sequencing does not.
What can move first
Move the seam, not the cathedral.
- Contracts. If the new UI talks to the old API through a documented client, product work can keep using the old screens until the new ones are boring.
- Read paths. Tracking, dashboards, and quote history are easier to dual-run than checkout. Let the new surface prove it can read production data before it writes.
- CI. A pipeline that builds both apps from the same repo is more important than a new color. If only you can deploy the new stack, you have not modernized anything.
On a greenfield portal the “old” system may be email and spreadsheets. The seam is still real: sales cannot wait for the perfect data model. I shipped quote and tracking against a database I was willing to migrate, not against a schema I was still debating.
What has to stay dual-running
Auth, writes that move money or freight, and anything with a customer-facing URL.
Dual-running means both surfaces hit the same source of truth, or the new surface is explicitly read-only. Feature flags beat a big-bang DNS cutover. If you need a flag per page, you do not have a platform yet — you have a folder of experiments.
Keep the old URL alive until the new path has error rates you would accept on a Monday. Redirects are cheaper than a war room.
What the team still ships every week
Name a stream that is not modernization: a sales request, a bug in production, a small API field. If that stream dies, modernization becomes the only story, and it will be resented.
Review stays the same. A modernization PR that bypasses tests because “it is temporary” becomes the new legacy. Temporary is a date, or it is a lie.
Ownership
Someone owns the cutover checklist who is not also the only person who understands the old system. If that is still you, write the checklist until it is not.
The lead job is to protect both clocks: the product clock and the platform clock. Stopping one to serve the other is how you get a beautiful new app that is already behind.