I have spent fourteen years doing one job under a lot of titles. Move a production system from old to new, on a schedule, without the people using it finding out it happened.
In 2012 that was a company’s documents on a dying RAID-5 array, one drive failure from total loss. I stood up a replacement on ZFS and cut over keeping every share name and address identical. Nobody noticed the storage had moved. That was the point.
The titles changed after that: sysadmin, DevOps, SRE, SRE manager, interim lead. The toolkit changed under them. Ansible in 2014, before Red Hat bought it. AWS in earnest from 2016. Terraform from 0.11 in 2018. Kubernetes in production by 2022. The job kept its shape. In 2021 it was converting a live production database to UTF-8 for a marketplace expansion, rehearsed for weeks and executed without taking the product down, zero rows lost. In 2022 it was a Fortune-class retailer whose migration of roughly 4,000 microservices had stalled for two years with fewer than ten done. The blocker was trust, not technology. We rebuilt the team, proved one repeatable cut, and reached about two-thirds in two quarters.
Different decade, same job: move the thing, don’t break it, don’t make a show of it.
What I kept noticing across all of it was how migrations die. Almost never on cutover night. They die months earlier, in the meeting where someone says the safe thing: let’s run both systems in parallel for a while, and finish when there’s a quiet quarter. There is no quiet quarter. The parallel run becomes the architecture: two systems to feed, two bills, two pager rotations, and a cutover date that slides forever because nothing forces it. The hybrid that was meant to last a month is load-bearing three years later. I have been called in to clean up a few of those.
The fix is boring. Name the scope, name the date, defend both. A migration with a date on it gets cut. One without a date rots. That is most of what fourteen years taught me, in two sentences.
So that is what I have been building. Hardcut is a solo cloud-migrations practice with a single promise: legacy infrastructure to modern cloud, in one defined window. The new system goes live and the old one goes quiet on the same date. Fixed price, fixed scope, fixed cutover. Larger estates ship as a sequence of clean cuts, never an eighteen-month parallel run. We spend the weeks so your weekend is boring.
It runs solo on purpose. A migration is a stack of judgment calls: what to lift as-is, what to redesign, what to leave alone. Those calls don’t survive a handoff to whoever is on the bench. The person who quotes your cut is the one who reads your infrastructure, writes the Terraform, drills the rollback, and pushes the DNS change at 02:00 on the night. No subcontractors, no offshore relay. That caps the practice at eight to twelve engagements a year, which is the product, not a limit.
I’m Canada-based but region-agnostic. Most startups run in us-east-1 or us-west-2 for cost, and that’s where I’ll meet you. Canadian regions and data residency are on the table when you actually need them, not a default I’ll push you toward.
AI is in the toolkit now, the same way Terraform was in 2018. It compresses the weeks so one operator delivers what used to take a team. It is the how, not the headline. The headline is still a clean cutover in weeks, not quarters.
If you are staring at a migration you have deferred for quarters, that is the conversation I want. Thirty minutes, no deck, no sales engineer. Either there is a Hardcut-shaped cut here or there isn’t, and you will know which by the end of the call.