Written by: Agata Wojtas, Chief Commercial Officer, Digital Colliers
You budget for the day rate. You forget to budget for the first four weeks when that new engineer is reading documentation, asking questions, and shipping nothing that reaches customers.
That gap is ramp-up cost. It shows up as calendar time before the augmented engineer starts closing tickets at full speed. AI coding assistants help with some of it. They do not help with most of it.
The month nobody budgets for
An external engineer joining your team needs three things before they are useful. They need to understand your domain, your codebase, and your team's decisions. The domain part takes weeks. Your billing logic, your data model, your customer workflows. No AI tool reads your internal Confluence pages and explains why the payment flow has five fallback paths.
The codebase part takes days to weeks. Where does authentication happen, how do you handle errors, which service owns which data. AI can surface examples faster than grep, but it cannot tell the engineer why you chose Postgres over Mongo in 2023 or why the mobile app calls three endpoints where the web app calls one.
The decisions part is the longest tail. Why did we not use that library, why does this endpoint return a 202 instead of a 200, why do we rebuild the search index at 3am UTC. Those answers live in Slack threads, pull request comments, and the memory of your senior engineers.
What AI assistants actually save
AI coding assistants do compress one piece of ramp-up. They write boilerplate faster. Your new engineer sees how your team structures a controller, and Copilot writes the next five controllers in that style. They see your test pattern, and the assistant generates twenty more tests that match it.
That helps. Experienced developers using AI assistants review 6.5% more code than they did before. But those same developers show a 19% drop in their own original code productivity. The assistant writes fast. The human spends more time checking whether it wrote the right thing.
More than 15% of commits from every AI coding assistant introduce at least one issue. Some of those issues are trivial. Some are not. Your augmented engineer now spends part of their ramp-up period learning which AI suggestions to accept and which to rewrite.
Where the human bottleneck stays
Code review is still a human gate. Your senior engineers review every pull request from the new hire. They check for logic errors, security gaps, and whether the solution fits your architecture. AI does not do that review. AI does not know that your team deprecated that pattern last quarter or that this change will break the mobile app.
Debugging production issues is still human work. Your new engineer hits a bug in staging. The stack trace points to a service they have never seen. They need to ask someone which logs to check, whether this is a known issue, and whether the fix belongs in the service or the caller. AI tools can search your logs faster. They cannot tell the engineer what normal looks like in your system.
Context transfer is still a meeting. Your tech lead sits down with the new engineer and explains the big decisions. Why you are rewriting the reporting pipeline, why the old vendor integration is still running in parallel, why the team is moving to event sourcing for orders but not for users. That conversation does not happen in a prompt.
The math that most teams miss
Ramp-up cost is not just the salary of the person ramping up. It is also the time your existing team spends answering questions, reviewing code, and fixing mistakes. That cost doubles when the augmented engineer is remote and timezone-shifted.
A rough model: if your new engineer takes four weeks to reach 80% productivity, and they pull two hours per day from your senior engineers during that period, you have burned 40 hours of senior time plus four weeks of reduced output from the hire. AI assistants might cut the boilerplate portion of that ramp-up by 20%. They do not cut the domain learning, the decision context, or the review overhead.
Most teams hiring augmented engineers focus on the hourly rate and the contract length. The teams that get augmentation right also model the onboarding calendar. They ask how long it takes their existing engineers to ship independently, then multiply that by 1.5 for external hires. They block senior time for pairing and review in week one. They document the decisions that new hires always ask about.
AI makes some of that faster. It does not make it free.