You Inherited a Codebase From a Failed Vendor: Keep, Refactor or Rewrite?
A failed vendor leaves you with code you did not choose and a deadline you cannot move. Here is what to secure in the first 72 hours, a seven-point audit with red, amber and green ratings, and a decision matrix for keeping, refactoring in place or rewriting.
What Should You Do in the First 72 Hours?
Secure access before you form an opinion about the code. A new team can only work with what you can open, so the first three days are about control, not quality. Get the repositories transferred into an account you own, then list every system the product touches: hosting and cloud accounts, the domain and DNS, the database, CI, monitoring, payment and email providers, and any third-party API keys.
Then rotate what the old team could see. Change passwords and keys, remove the former vendor's people from every account, and confirm that backups exist and can be restored. If you do not control a system, treat the code in it as not yet yours. This is also why we ask that engineers work in your repository and your tools from day one: nothing has to be extracted from a vendor later.
For the legal and handover side of leaving a vendor, see our guide to a vendor transition plan and the software project takeover page.
What Does a Seven-Point Audit Check?
Rate each area red, amber or green after a short, hands-on look rather than a long report. The point is not a score; it is a list of what blocks you from changing the product safely. A senior engineer who did not write the code can run most of this in days.
| Check | Green | Red |
|---|---|---|
| The tests run | A clean checkout runs the tests and most pass | No tests, or they cannot be run |
| The build is reproducible | One documented command builds the same artefact on a new machine | It builds only on the old developer's laptop |
| Dependencies are supported | Languages and databases are still receiving security fixes | Core runtimes are past end of support |
| No secrets in the history | Keys live in a secrets store, not in commits | Passwords or keys appear in repository history |
| The data model is understood | A schema you can explain and a migration history | Unexplained tables, manual changes in production |
| Deployment is repeatable | A pipeline deploys, and rollback is known | Deploys are manual steps someone remembers |
| Documentation exists | A README, setup steps and the main decisions written down | Nothing beyond the code itself |
How Old Are the Dependencies, Really?
Check the runtime and database versions first, because end-of-support dates are fixed and public. A codebase on a version that no longer gets security fixes is not just old; it carries a deadline. We cover three that matter right now: the Python 3.10 end of life, the PostgreSQL 14 end of life and the .NET 8 and 9 end of support. An upgrade is a refactor, not a rewrite, so it belongs on the keep-or-refactor side of the matrix.
Adding engineers to a roadmap like this? Spec your team in a minute.
Keep, Refactor in Place or Rewrite: Which One Does the Audit Support?
Choose the cheapest option the audit supports, and write down the trigger that would make you change your mind. Mostly green means keep the code and add what is missing. A mix of green and amber means refactor in place, one area at a time, while the product keeps running. Rewrite is for the case where the audit finds red in the areas you cannot fix piece by piece, such as a data model nobody can explain or an architecture that cannot take the next feature.
| Option | Choose it when | Trigger to reconsider |
|---|---|---|
| Keep | Mostly green: it runs, it is tested, you can deploy it | Two or more areas turn amber within the first month |
| Refactor in place | Green core with amber or red edges you can replace one at a time | A fix in one area keeps breaking another |
| Rewrite | Red in the data model, architecture or security, and the product cannot move without them | The audit finds the red areas are smaller than first thought |
Why Do Full Rewrites Usually Fail?
Because old code contains years of fixes that nobody wrote down. Joel Spolsky made the case in 2000 in Things You Should Never Do, Part I, calling the decision to rewrite from scratch the single worst strategic mistake that any software company can make. His reasoning is the useful part: the old code looks messy, he wrote, because those are bug fixes. A rewrite throws that knowledge away and rediscovers it the expensive way, while the old product stands still.
The essay is a 2000 opinion about one company, not a study, and it does not say a rewrite is never right. It does explain why the bar is high. If you choose to rewrite, do it in slices that replace one part at a time behind the running product, and keep the old system working until each slice has proved itself.
Can a Small First Workstream Work as the First Step?
Yes, as a way to test the audit on real work, not as a separate audit product. A developer who is not the right fit is replaced free within the first 30 days. Give the new engineers one small, real change in the inherited code: fix a defect, add a test, ship a deploy. How easy or hard that is tells you more than a report.
Write the pass or fail criteria before day one. Engineers can typically start within one to two weeks of first contact; a full team usually takes two to four weeks, and the engineering practices that apply on every engagement, such as peer review and tests on every pull request, are the ones to hold the new team to from the first change.
What Is Your First Move Today?
Secure access, rotate the keys, then run the seven checks before you choose a direction. If you want a second pair of eyes on what the audit shows, tell us the stack and what the previous vendor left on our contact page and book a 15-minute intro call; if there is a fit, a 60-minute discovery call follows.
Building a team in Eastern Europe?
StepTo builds dedicated nearshore engineering teams in Belgrade, Serbia. Engineers can typically start within one to two weeks of first contact; a full team usually takes two to four weeks. From $4,500 per developer per month.
Related: Software project takeover
