Stephanie Davis was Google’s director of large-company sales, and one of the winners of its Great Manager Award, picked from employees’ nominations. She still had something to learn from her team.
Project Oxygen had found that managers mattered and had boiled good managing down to eight behaviors. Google turned them into a survey in which employees rated their own managers, twice a year. One line in Davis’s first report surprised her.
One person on her team didn’t think she held regularly scheduled one-on-one meetings.
“I saw this person every day,” she told Harvard Business Review. What the survey made her realize was that seeing someone every day “was different from having regularly scheduled individual meetings.”
Andy Grove had drawn that line decades earlier.
Seeing someone every day isn’t a one-on-one. Give them a set time and their own agenda.
Eight times or once
Grove, then Intel’s president and later its chief executive, wrote High Output Management in 1983. He called the one-on-one the principal way a manager and a subordinate keep up their working relationship, and said its main purpose is “mutual teaching.”
Then he turned the usual arrangement around. The one-on-one “should be regarded as the subordinate’s meeting, with its agenda and tone set by him.” His reason was practical. Somebody has to prepare. A supervisor with eight subordinates would have to prepare eight times, the subordinate only once, so the subordinate should bring an outline.
Talk at someone’s desk tends to follow whatever is urgent that day, and whoever starts it. A slot that belongs to the other person gives their questions a place, including the ones that never feel urgent enough to interrupt anyone with.
Bumping into your doctor at the supermarket is not a check-up. You may see each other every week. You talk about the weather, because nobody brought the test results.
How often, how long
Grove did not prescribe one rhythm for everyone. He tied frequency to how much experience a person has with the specific task, what he called task-relevant maturity: one-on-ones often, for example once a week, with someone new to the work; every few weeks with an experienced veteran. How quickly things change in the job mattered too. On length, he was firm: a one-on-one “should last an hour at a minimum.”
Researchers have only lately caught up. In a 2022 paper in Organizational Psychology Review, Jonathan Flinchum, Liana Kreamer, Steven Rogelberg and Janaki Gooty wrote that one-on-ones “have not been studied empirically as a focal topic despite making up nearly half of all workplace meetings.”
Flinchum’s doctoral study the same year, supervised by Rogelberg, surveyed 303 people twice. In that sample, what happened in the meetings, especially the manager’s relationship-oriented behavior, mattered more for engagement than how often they met, with one caveat: the meetings had to happen at least monthly. The one-on-ones were also linked to engagement over and above other meetings and contact with the manager, roughly the distinction Davis’s survey drew. It is one dissertation, not a settled finding.
The question at the desk
Say you sit two desks from an analyst you talk to every day. Your weekly one-on-one keeps getting moved, because you already know what she’s working on. Meanwhile another team has asked her to help on its project, and she can’t tell whether you’d see that as a chance or a distraction. At the desk, the talk is about today’s reports, and her question never seems to fit.
If the hour is hers and she arrives with a list, that question can be the first item. If it isn’t, she might turn the other team down without asking you. You might first hear about it months later, in someone else’s update.
When the agent asks first
Working with an AI coding agent has its own version of this meeting, and Grove’s inversion is the part that carries over.
Anthropic’s guide to Claude Code suggests that for larger features you have Claude interview you before it writes anything. You ask it to dig into “technical implementation, UI/UX, edge cases, concerns, and tradeoffs,” keep going until the questions run out, then write the answers to a spec. The one doing the work brings the questions, and you answer them.
The same guide tells you to “course-correct early and often”: correct Claude as soon as it goes off track. And it notes that each Claude Code session “begins with a fresh context window.” Whatever you settled last time reaches the next session only if it was written down, for example in a CLAUDE.md file.
Much of Grove’s chapter doesn’t carry over. An agent has no career to discuss, no morale to read, and no reason to need an hour. What does carry over is the order. Stephanie Davis saw her report every day, and her report still didn’t count it as a one-on-one. With an agent, the check-in worth scheduling may be the one where it asks first.