--- title: "The pressure to look competent with Copilot when you're still figuring it out live" date: 2026-08-14T00:00:00Z updated: 2026-09-11T11:47:29Z tags: ["Career", "Copilot", "Visibility", "Judgment", "Adoption"] canonical: https://bianca.codes/blog/the-pressure-to-look-competent-with-copilot/ --- # The pressure to look competent with Copilot when you're still figuring it out live _Six people watching your cursor, a tool you've barely opened, and a room that assumes you've already mastered it. The pressure to look fluent with Copilot is manufactured - here's what to do instead._ It's a Tuesday afternoon screen-share. Someone asks whether the summary is ready, someone else says "Copilot can do that in about ten seconds", and now there are six people watching your cursor while you type a prompt into a pane you've opened maybe four times in your life. Whatever comes back, you'll need to react to it like you expected it. The assumption in the room is that you've already got this. Copilot has been in the tenant for months, it's in every ribbon, leadership has mentioned it in two consecutive all-hands. Admitting you're still working out what it's actually good at feels less like a status update and more like a confession. The standard advice is to close the gap quietly - do a course, watch the videos, catch up at home until the fluency is real. That advice misdiagnoses the problem. The gap it wants you to close was never a skills gap. It's an expectations gap, and you weren't the one who set the expectation. ## The learning curve didn't change, the audience did Every tool you're currently good at, you got good at in private. Nobody watched you learn Excel. You fumbled INDEX and MATCH alone at your desk, swore at a circular reference, eventually produced a workbook that worked, and the workbook is what people saw. The learning happened offstage. The competence went on stage. That has been the deal with office software forever, and it's why a normal learning curve has never been a career problem: it was invisible. Copilot broke the deal. The interaction is the performance. Prompting happens live, inside the document, often with the room watching, and the output lands in front of everyone at the same moment it lands in front of you. There is no offstage. The same learning curve you had with pivot tables now runs under stage lighting, and a normal learning curve watched live looks a lot like incompetence. It isn't. It's learning with the lights on, which is a different thing, and almost nobody has done it for office software before because office software never demanded it. That's half the diagnosis. The other half: the fluency being assumed of you mostly doesn't exist anywhere in the room. [Copilot arrived as an expectation before it arrived as a practice](/blog/copilot-arrived-as-an-expectation/) - the licences landed top-down, the expectation landed with them, and the practice time never landed at all. The colleague who said "ten seconds" is usually quoting a demo, not describing their week. You're not behind the room's baseline. The baseline is manufactured - assembled from launch keynotes, a demo clip someone saw, and everyone else also deciding not to be the first to say they're still figuring it out. ## What to do instead of faking it Not five things. Three, and they all fit inside a normal work week. **Narrate, don't perform.** When you have to prompt live, frame it as a test both you and the room are watching: "I'll ask it to pull the action items out of this thread - let's see what it does with it." That's one sentence, and it changes the physics of the moment. If the output is good, you get the win. If it's mediocre, you've demonstrated the tool's limit rather than yours. A performance can fail. A test can only produce a result. **Get your reps on one recurring task, in private.** Fluency comes from repetition against work you already understand, not from courses about work you don't. Pick one task you do every week - the status summary, the forty-email thread catch-up, the first-draft agenda - and run it through Copilot every time, before doing it your usual way. You learn its failure modes on familiar ground, where you can spot a wrong answer instantly, and within a few weeks you hold the one thing the room doesn't have: firsthand comparative knowledge. And [most of what looks like prompting talent is a learnable structure](/blog/copilot-doesnt-work-is-usually-a-prompting-problem/), not a knack you either have or don't. **Let judgment be the visible skill.** The scarce skill in the Copilot loop is not typing the request. It's evaluating the response. When a summary comes back and you're the person who says "that's missed the decision from Thursday's thread", you have displayed more expertise than the person who wrote the prompt, because evaluation runs on job knowledge and you already have years of that. Spend your visible moments on the judgment, not the incantation. ## Frequently Asked Questions ### **Isn't admitting I haven't mastered Copilot yet bad for my career?** Usually the opposite, provided it's paired with a test. "I haven't used it for that - I'll find out and report back" followed by an actual result reads as rigour. What damages credibility is the fluent nod that gets contradicted three weeks later, when you can't reproduce the thing you implied you'd already done. ### **My manager keeps asking what we're doing with AI. What do I say?** Give a small, specific, true answer instead of a broad aspirational one: name the one task you've been running through Copilot and what you found, including where it fell short. A manager asking that question is usually feeling their own version of the same pressure from above, and a concrete finding is worth more to them than confident vagueness. ### **Everyone around me already seems fluent with Copilot. Am I actually behind?** Watch what people produce, not what they say. Genuine fluency leaves artifacts: reworked drafts, summaries that held up, a task that visibly stopped taking an hour. If nobody can show you a before-and-after from their own work, what you're seeing is atmosphere, mostly assembled from demos and other people also not wanting to look behind. ### **Does this change if I'm senior, or supposed to be leading the rollout?** Seniority raises the stakes of the move without changing the move. A lead who says "here's what it did well on our documents, and here's where it fell over" builds more real adoption than one who projects fluency. It also makes learning in public safe for everyone junior to you, which is the part of a rollout no licence covers. ### **What if my organisation tracks Copilot usage numbers?** Usage dashboards count prompts, not outcomes, so the weekly recurring-task habit already covers you: real usage registers exactly the same as performative usage, and only one of them leaves you with something to say when the number gets discussed. Don't invent activity to feed a metric. Run the real task and let the count take care of itself. ### **How long does it take for Copilot fluency to be real rather than performed?** Running one weekly task through it, expect a few weeks before you can predict its failure modes on your own work, and a couple of months before your instincts about what to hand it are reliable. That timeline sits a long way from "ten seconds", which is exactly why holding firsthand knowledge beats holding the demo reel. ## This week The next time Copilot comes up and the room assumes you've already done the thing, say the accurate sentence instead of nodding: "I haven't used it for that yet. I'll try it this week and tell you what it actually does." Then try it once and report back, whatever the result. The output matters less than the position it puts you in - the one person in the room whose claims about the tool have been load-tested.