The mentoring session runs forty minutes. The mentor explains how they handled something similar on another team and recommends three things to read. Then come the "you should"s: speak up more in reviews, ask for context first, don't accept scope without asking questions. It ends with "if anything comes up, just ping me." The mentee says thanks, enjoyed the conversation, and goes back to a week that looks exactly like it did before.
Two weeks later there's another 1:1, and it starts from scratch. The same advice comes back, now with more conviction.
The habit here is a small one. A mentoring 1:1 ends with a one-line experiment: one small, observable thing the mentee tries before the next conversation, with a date for reporting back. The next 1:1 opens with that report, before any new advice. By the end of this post you'll be able to check three things: the line exists, it has a date, and it got reopened.
Why advice-only conversations don't change the week
Advice is a hypothesis that somebody still has to test. If nobody agreed on who tests it, when, and where, the hypothesis goes into the mentee's memory and sits there with all the others.
Google re:Work's guide to effective 1:1s was written for manager-and-report 1:1s, not mentoring. Still, the agenda it recommends closes with "Next steps to confirm actions and agreements," and it asks you to review action items in the following 1:1. If the conversation confirmed no actions, the next one has nothing to review. That's why it starts from scratch.
In The Leader as Coach (HBR, 2019), Herminia Ibarra and Anne Scoular put classic mentoring on the directive side, the "telling" side. That style, they write, "unleashes little energy" and can even depress motivation, because it assumes the person talking already has the answer. In a role-play they describe, executives who had already settled on the answer slid from open questions to leading questions until they were just telling. The result: "At the end of the exercise, no one has learned anything about the situation or themselves."
Atlassian's guide to action items describes the same problem from the meeting side: when people leave without a clear sense of what they're responsible for, work falls through the cracks.
There's a useful parallel from outside mentoring, too. Gollwitzer and Sheeran studied the gap between meaning to do something and actually doing it. The NCI page summarizing that work lists forgetting to act and missing the moment as typical failures. That research is about personal goals and health behavior, not mentoring 1:1s, so take it as an analogy and not a number about mentoring. It's a good analogy, though. "I'll speak up more in reviews" is an intention. It doesn't say which review.
Close with a one-line experiment
"Experiment" is DevDojo's name for one small, observable try. It isn't anyone's method. The sources talk about actions, next steps, or action items. What they share is the close.
Structuring Mentoring Conversations for Success, a job aid published in 2025 by the Canada School of Public Service (CSPS), structures the mentoring conversation using the GROW model. The last stage comes down to "what will you do, and when." Two of its suggested questions are "What actions will you take to achieve your goal, and when?" and "How would you like to be held accountable?" The guide also makes it clear that the mentor can give advice mid-conversation, but the mentee is the one who sets the closing actions, working with the mentor.
The GitLab handbook's suggested 1:1 agenda format does the same thing with a label: an item tagged TODO gets a completion date in the shared doc, and that doc keeps getting updated from one meeting to the next. Nobody starts each meeting on a blank page.
The rest is our own editorial take. The experiment fits on one line and has three parts:
| Part | What goes in | What doesn't |
|---|---|---|
| What I'll try | Something another person could actually see | "Be more proactive," "communicate better" |
| Where and when | A real occasion that happens before the next 1:1 | "When the opportunity comes up" |
| Report date | The next 1:1, or a day on the calendar | "If anything comes up, just ping me" |
For example, written in the shared 1:1 notes:
> "By the October 25 1:1: in the next checkout review, I'll state the risk in one sentence before the diff; I'll bring back what happened."
That's one experiment. Atlassian suggests keeping the action list short, somewhere around three to five items before the next checkpoint. For mentoring, our number is one. Five experiments in a normal working week turn into five intentions, and we just saw what happens to intentions.
Advice is still allowed. The mentor can share what they did, suggest three paths, and argue against one of them. The conversation just doesn't end there. The mentee picks one path and shrinks it until it fits on a line they can try before the next conversation.
At the next 1:1: tried, saw, didn't try
Google re:Work and Atlassian say the same thing: items from the previous meeting get reviewed in the next one. Atlassian adds that anything left open carries forward to the next agenda instead of disappearing. So the first thing in the next 1:1 is last week's line.
The report format is ours, and it has three parts:
- Tried: what was done, and on which occasion.
- Saw: what observably happened, with no guessing at anyone's mood or intent.
- Didn't try: if it didn't happen, why, in one sentence.
Continuing the example:
> "I tried it on Tuesday's review call: I opened with the risk in one sentence. I saw the reviewer go straight to the retry section instead of reading the diff top to bottom. I didn't try it on Thursday, because that review went async."
"Didn't try" is information, not a confession. Maybe the occasion never came up, maybe the experiment was too big, maybe the line lived in a note nobody opened. Each of those reasons leads to a different line in the next conversation. None of them calls for a lecture.
Only after the report does new advice come in, and now it starts from something that actually happened instead of a hypothetical.
What this conversation is not
- It isn't pulling the promotion bar out of your manager. That's a written bar for a cycle. This is one try inside one week.
- It isn't peer SBI feedback. Nobody is rewriting a "great job, but…" here. The mentee is the one acting.
- It isn't the async help request. A "ping me anytime" at the end of a mentoring session doesn't teach anyone how to ask for help. It's just a mentoring close that didn't land.
- It isn't the solo experiment of re-reading your own diff. That one has no mentor and no report at the next conversation.
This week's check: the line exists before you leave the call
At your next mentoring 1:1, whether you're the mentor or the mentee, run a single check before anyone clicks "leave": is there a line written down, in the notes or in the calendar invite, with the experiment and the report date?
Pass if the line is there, with an observable action, a concrete occasion, and a date. Fail if the conversation ended in "just ping me," in a list of "you should"s, or in a line with no date. The "ping me" is friendly, and it's sincere. It just sits exactly where the date should be.
If it failed, there's still time: two minutes to pick one item off the list and shrink it until it fits in the week.
DevDojo's recommendation: a mentoring 1:1 that ended in advice alone is unfinished until it has a dated try in writing. If the next 1:1 starts from scratch, open with last week's line before offering any new advice. If the report is "didn't try" twice in a row, the problem is the size of the experiment, not the person. Shrink the line before you think about dropping the habit.
For this week, all you need is one line with a date, written before you leave the call.