
You have vetted the vendor, interviewed the engineers, run a trial, and signed. The team starts Monday.
What happens over the next four weeks will predict the outcome of the engagement more reliably than everything that preceded it — and unlike the screening steps, it is almost entirely under your control.
Here is a plan that works, structured by week.
Before day one
Three things must exist before anyone starts, and every one of them is routinely missing.
A local environment that actually works. Not documentation you believe is current — documentation that someone who has never set it up followed successfully in the last month. If your setup guide has a step that says "ask Sarah for the config," you do not have a setup guide.
Access provisioned. Repository, CI, staging, error tracking, whatever internal tools they need. On your identity provider under your access controls, not a shared vendor account. Scoped to what the work requires. Doing this on day one instead of before it costs the first two days of a paid engagement.
A named person who owns questions. Not a channel. Not a rotation. A person, with the availability to respond within hours during overlap. If nobody on your side can commit to that, address it now rather than discovering it in week three.
A first task chosen — small, real, and shippable within a week, with a clear definition of done.
Week one: get a pull request merged
The goal for the first week is exactly one thing: a small change through the entire path, from local setup to production.
It sounds unambitious. It is the highest-signal milestone available, for two reasons.
First, it exercises every part of your delivery process while the stakes are trivial. Every friction point surfaces — the flaky test, the CI step that needs a permission nobody documented, the deploy that requires someone else to click something. You want all of that discovered in week one on a change that does not matter.
Second, it establishes for both sides that the loop closes. An engineer who has shipped something, however small, is oriented in a way that no amount of documentation reading achieves.
Resist the urge to make the first task meaningful. Meaningful comes in week three.
Week two: widen the surface

Now give them something with a decision point in it.
The goal for week two is to observe how they handle the parts of your system that are not written down. Every codebase has conventions that exist only in people's heads, and the second week is when an external engineer either starts surfacing them as questions or starts guessing.
Watch the shape of the questions. Good ones come early and carry a proposed answer: "There are two patterns for error handling in this service — I'm assuming the newer one is preferred for new code, is that right?" That is an engineer building a model of your system. Questions that only ask for instructions, without a hypothesis attached, suggest someone waiting to be told rather than trying to understand.
This is also the week to leave substantive code review comments, including at least one you are not sure about. You want to know whether they push back with reasoning or accept everything silently. You are buying judgement, not compliance, and an engineer who never disagrees in review will never disagree about architecture either.
Week three: give them something that matters
By now the environment works, the process is understood, and you have a read on how they think.
Week three should be a real piece of roadmap work — something with genuine ambiguity, dependencies on other parts of the system, and a stakeholder who cares about the outcome.
The thing to watch is whether they start operating with any independence. Do they notice a problem adjacent to their task and flag it? Do they ask about the why behind a requirement, not just the what? Do they have an opinion about the approach?
Engineers who never develop independence are not necessarily weak. Often the engagement was structured to prevent it — no boundary they can own, every decision escalated, no context about why the product exists. If you see no independence by the end of week three, check your structure before you check the person.
Week four: calibrate explicitly
Have an explicit conversation at the end of the first month. Not a performance review — a calibration.
Three questions, both directions:
-
What has been slower or harder than it should have been?
-
What context is missing that would help?
-
What would you change about how we are working together?
Answer them honestly yourself as well. In most first months, the majority of the friction is on the buyer's side: review turnaround, unclear requirements, decisions that take days, access that was never granted. Surfacing that at week four costs a conversation. Surfacing it at month four costs a quarter.
What good looks like at day thirty
-
Multiple changes merged and deployed, at least one of them non-trivial
-
The engineer can explain what the product does and who uses it
-
Questions have moved from "how do I" to "should we"
-
At least one instance of them finding a problem you did not point at
-
Your review turnaround is under a day, consistently
If most of those are true, the engagement is on track and the remaining risk is management discipline over time. If most are not, the problem is diagnosable now, while it is still cheap.
Full guide to the model — costs, vendor vetting, trial sprints, contract terms and measurement: Hire Dedicated Developers in 2026. We also help teams structure these engagements when getting it right the first time matters.
Frequently Asked Questions
What is the single most important onboarding milestone?
A merged pull request in week one. It exercises your entire delivery path while the stakes are low, surfaces every process friction point, and orients the engineer in a way that reading documentation cannot.
How much of my time will onboarding take?
More than you expect in the first month — realistically a fifth to a quarter of a senior person's time per three to four external engineers, weighted heavily toward the first two weeks. If nobody can commit that, the engagement will underperform regardless of who you hired.
What if the engineer is not showing independence by week three?
Check your structure before you check the person. No ownable boundary, every decision escalated, and no context about why the product exists will suppress independence in anyone. If the structure is sound and the independence still is not appearing, then it is a fit question.
Should external engineers use their own accounts or ours?
Yours — on your identity provider, under your access controls, scoped to the work. Shared vendor accounts break your audit trail and become a finding the first time a customer runs a security review on you.
How do I know if the problems are on my side?
Track review latency and blocked time. Both are usually buyer-side metrics, and in most first months the majority of friction traces to slow review, unclear requirements, or access that was never granted. Ask directly at week four and answer honestly.
When should I expect full productivity?
Four to eight weeks on a non-trivial codebase, and it is not linear — output usually jumps once the engineer has a mental model of the system rather than a map of its files. Budget the ramp explicitly rather than hoping it will be quick.