People Management
Mentor a Junior Engineer Without Becoming Their Permanent Bottleneck
A filled four-week learning record and coached keyboard-navigation case show how to build independent reasoning while retaining review and escalation.
Mentor a junior engineer by making their reasoning visible, agreeing when to ask for help, and gradually changing the support you provide. Keep normal code review and escalation available. Independence means they can frame a bounded problem, choose and explain a next step, verify the result, and recognize their limits; it does not mean working alone or avoiding questions.
This LandOffer.ai guide includes a filled four-week learning progression and a fictional before-and-after coaching case. The people, work records and outcomes are teaching examples. No employee was evaluated, production code changed, or mentoring experiment conducted for this article.

Choose one capability you can observe
“Become more independent” leaves both people guessing. Pick a repeatable activity within the engineer's assigned work: reducing a UI report to a reproducible case, interpreting a failed check, explaining a small change, or identifying the right owner for an unfamiliar subsystem. Describe what the engineer should be able to demonstrate and the review that still applies.
In GitLab's Support Engineering Mentorship guidance, last modified June 27, 2024, mentoring is a learning opportunity rather than handing the issue to the mentor. It asks mentors to explain their thinking and check shared understanding. Its example program and pairing expectations belong to that support organization; they are not a universal schedule or a measured result for your team.
Discuss the goal with the engineer instead of inferring a deficiency from repeated questions. Missing documentation, permissions, unclear requirements or an unfamiliar business rule may require an owner decision. Coaching cannot replace those inputs. An engineer who escalates an unknown requirement can be exercising better judgment than one who confidently invents it.
Choose a task with useful uncertainty and a manageable consequence of error. A new engineer does not need unrestricted production access to practice diagnosis. Use approved examples or an appropriate development environment, and retain the team's actual change controls. The learning goal must fit the permissions and work the engineer is authorized to undertake.
Agree on help before someone gets stuck
Set a help arrangement that protects both people. Name the channel, the mentor's available windows, the information that makes a question answerable, and an alternative when the mentor is unavailable. Ask the manager to account for this support in both workloads. Do not promise unlimited interruptions while maintaining the mentor's original delivery commitments.
A useful question describes the intended behavior, the observed difference, the smallest permitted example, what has been tried and what decision is needed next. This is preparation for a conversation, not an admission test. If the engineer cannot yet produce that description, help them create it; do not send them away indefinitely to “try harder.”
Separate ordinary uncertainty from urgent risk. Follow the team's existing escalation process for suspected data loss, security concerns, production incidents or changes that exceed authority. Do not require a fixed solo-investigation period before raising such a concern. A learning timebox is appropriate only for the low-risk work you have agreed to practice.
GitLab's broader mentoring guidance assigns mentees a leading role in conversations and asks participants to keep written goals and distinguish the mentor's responsibilities from the manager's. Those expectations support an explicit arrangement. They do not authorize a mentor to set a performance rating, change someone's assignment or share private discussions without the appropriate basis.
A filled four-week progression
The following plan is entirely fictional. Junior engineer Nia maintains the keyboard navigation of an internal diagram editor. Senior engineer Mateo mentors her; manager Asha owns assignments and capacity. Their goal is for Nia to prepare an evidence-based investigation plan for a small keyboard-navigation report while using ordinary review and escalating unresolved product behavior.
Asha allocates Mateo ninety minutes a week for two thirty-minute sessions and thirty minutes of preparation and follow-up, reducing his optional visualization prototype by the same amount. Nia has three hours a week of planned learning within her existing assignment, replacing optional cleanup work. Reviewer Jo retains the ordinary review allocation already in her workload and remains responsible for the code-review decision; mentorship creates no release authority.
| Week | Nia's completed learning record in the fictional case | Support used and next boundary |
|---|---|---|
| 1: observe and explain | After Mateo demonstrates how to separate an expected behavior from an observation, Nia writes a keyboard sequence and a prediction for each step. She notices that one prediction is an assumption. | Mateo supplies the first example and asks Nia to explain it back. Product owner Ren must settle the behavior question; it is not filled in by the mentor. |
| 2: lead a bounded investigation | Nia chooses two small diagrams that differ in one selection condition, records the visible focus destination, and prepares a proposed comparison. | Mateo asks which difference the comparison can isolate. He helps narrow it but does not take over the investigation or declare the cause confirmed. |
| 3: propose and verify | Nia writes an expected-behavior table, a proposed change boundary and checks to request in review. She independently includes a single-selection case but needs a prompt to include a no-selection case. | Jo reviews the proposal through the ordinary process. The missing case remains evidence of support needed, not proof of a personality trait. |
| 4: transfer the reasoning | For a different navigation report, Nia produces the problem statement and alternative explanations before the planned session. She identifies an unresolved product rule and routes it to Ren. | Mateo reviews the reasoning afterward. Ren's answer is pending; Nia has not completed a fix or established independence across the editor. |
The record shows a change in who frames the work, with support still visible. It does not claim that four weeks is enough for every engineer or that Nia became faster. If the agreed work had been repeatedly interrupted, Asha would need to revisit the plan rather than use missed opportunities as evidence of inability.
At each review, choose the next support level from the evidence. You might demonstrate once, ask the engineer to lead while you observe, review their plan before action, or discuss the completed reasoning afterward. These are options for the selected task, not permanent ranks assigned to the person. Agreeing the support level tells Nia when to proceed and when a review is still needed. A task with greater consequences can appropriately require more supervision again.
Before and after: coach a decision rather than supply a patch
Here is the completed teaching case behind the second and third weeks. The fictional report says that pressing the right-arrow key moves focus unexpectedly after selecting two nodes in the diagram editor. The expected destination for that condition has not yet been confirmed by product owner Ren. No application was run for this case, and the observations below are invented case inputs rather than executed test results.
Nia's original request is: “The selection code is wrong. Can you tell me which function to change?” Her initial note contains a screenshot and a proposed shortcut: move focus to the next node whenever the arrow key is pressed. It does not distinguish selection from keyboard focus or specify how the next node is chosen.
Mateo first asks Nia to walk through the user action and point out what she knows versus assumes. He explains the distinction between selected nodes and the node receiving keyboard input, using a small diagram. Nia then writes this revised record:
| Case input | Recorded observation supplied by the fictional report | Expected behavior and next decision |
|---|---|---|
| One node selected; focus on that node | Right arrow moves focus to the adjacent node | Ren confirms that this is intended for the example; retain it as a comparison case. |
| Two nodes selected; focus on the first | Right arrow leaves focus on the first node | Intended behavior is unresolved. Ask Ren whether the key should move focus, change selection or leave both unchanged. |
| No nodes selected; canvas focused | No observation was supplied | Nia adds this missing case after Mateo's prompt. Obtain the permitted example before proposing behavior or a fix. |

Nia's revised proposal reads:
I cannot yet call this a selection defect. The report establishes a difference between the one-node and two-node cases, but we have not agreed what the two-node case should do. I will ask Ren to confirm the intended focus and selection behavior. Then I will propose the smallest change, if one is needed, and include the no-selection case in the review checks. I will not apply the always-move shortcut because it could change behavior outside the report.
Ren's fictional response confirms the one-node behavior and asks for a clearer two-node interaction example before deciding that rule. The completed coaching session therefore ends with a better problem statement and an owned requirement question, not a resolved defect. Mateo records that Nia distinguished observation from expectation after instruction and added one missing case with a prompt.
That limited result matters. It identifies what to practice next without taking credit for a fix nobody made. On the next report, Mateo can ask Nia to draft the comparison and identify missing expectations before the session. He still supplies direct explanation if a concept is unfamiliar; withholding useful information would not make the exercise more independent.
Explain review comments and let the engineer choose
Coaching continues through review, but the review needs a clear purpose. Mark a required correctness change as required, explain the concern and let the author propose an appropriate solution. Label optional style preferences accordingly. Otherwise a junior engineer may interpret every suggestion as a rule they must obtain permission to depart from.
Google's code-review comments guidance advises explaining reasoning and balancing direct guidance with allowing the developer to decide. It also recognizes that explicit suggestions or code can sometimes help. That is code-review guidance, not a controlled study proving a mentoring technique or a reason to avoid necessary corrections.
For Nia's proposal, a useful comment is: “Required before choosing a fix: confirm the expected two-node behavior. The current comparison shows a difference, but it does not establish which result is incorrect.” An optional comment might suggest a clearer diagram label. The first protects the engineering decision; the second should not obscure it.
Ask the engineer to explain the revised reasoning in their own words and preserve the useful explanation where future reviewers can find it. Do not require them to imitate your exact implementation merely to demonstrate agreement. Assess the proposed behavior and evidence against the team's standards, including cases where their approach differs from yours.
Reduce dependence without withdrawing support
Review what help was needed across several appropriate opportunities. Can the engineer formulate the question before you arrive? Do they identify missing evidence? Can they describe why a check matters and recognize when another owner must decide? These observations give you more useful information than counting questions or timing every conversation.
Record the support as carefully as the achievement. “Identified the missing requirement independently; needed a prompt for the no-selection case” is actionable. A label such as “not proactive enough” can leave the engineer unsure what to change. Describe the missing action and the support available instead. Do not use the mentoring notes as a substitute for the manager's formal process or a secret performance dossier.
If every question still comes to one mentor, inspect the system as well as the learning. Add an approved decision example to shared documentation, identify a backup reviewer, or schedule a broader help session. Asha must confirm that any backup has capacity; putting another name in a document does not create it. If support cannot be staffed, narrow the work or change the plan.
Finish the cycle by agreeing which capability Nia can now practice with less immediate help, which still needs review, and what happens when Mateo is absent. Save the filled progression and corrected reasoning example. At the next agreed check-in, name one capability, the support it still needs and who will answer when the mentor is away. For related workplace-development guidance, visit LandOffer.ai.


