The Plateau Playbook: how to get better at something you are already good at

Getting from bad to competent is easy. You do the thing, you get obviously worse outcomes than you wanted, the gap tells you what to fix, and you fix it. Feedback is loud and free.

Getting from competent to genuinely good is a different problem, and most people never solve it. At competent, the feedback goes quiet. Your work is fine. Nobody complains. You practise by doing the job, which is exactly the thing you are already able to do. Ten more years of that produces ten more years of competent.

This is the plateau. It is not a motivation problem and it is not a talent ceiling. It is a feedback problem wearing a motivation costume. Here is the system for breaking it.

Step 1: Find the sub-skill, not the skill

“Get better at writing” is unactionable. “Get better at first paragraphs” is a project.

Take your skill and break it into the smallest units that could be practised in isolation. For most work skills there are between six and twelve.

• Writing: openings, transitions, cutting, structure before drafting, headlines, endings, dialogue, the specific verb.
• Sales: the opening 30 seconds, discovery questions, handling one specific objection, silence after the price, the follow-up email, the close, the re-engagement of a dead lead.
• Coding: naming, decomposition, test design, debugging strategy, reading unfamiliar code, API design, performance profiling.
• Managing: one-to-ones, delivering bad news, delegating with a spec, running a decision meeting, written updates, hiring interviews.

Now score each one honestly, 1–10. Not on how much you like it — on how good you are. The plateau lives in the sub-skills you rated 5 or 6 and quietly route around every day.

Worked example. A designer four years in scores: visual polish 8, layout 8, typography 7, user research 4, presenting work to executives 4, saying no to scope 3. She has been “practising design” for four years by doing more of the 8s. Her ceiling is not visual. It is the 3.

Step 2: Build a rep that isolates it

A rep is a repetition of the sub-skill that is (a) short, (b) separable from real work, and (c) produces a result you can judge within minutes.

The test: could you do twenty of these in an hour? If not, you have designed a project, not a rep.

• Openings: take twenty published articles. Write a new first paragraph for each. Twenty reps, one hour.
• Objection handling: record yourself answering “it’s too expensive” twenty different ways. Play them back. Two are good.
• Executive presentation: take an old project. Compress it into three sentences. Do it for ten old projects.
• Naming: open unfamiliar code. Rename every variable in one file before reading the logic. Then read the logic and see how wrong you were.
• Saying no: write the refusal email for ten requests you actually said yes to last quarter.

The point of isolation is that real work bundles ten sub-skills together, so a good outcome tells you nothing about which one carried it. Reps unbundle.

Step 3: Get the error signal inside 24 hours

Practice without a fast error signal is just repetition, and repetition past competence entrenches whatever you already do. You need to know you were wrong while you still remember what you were thinking.

Three sources, in order of usefulness:

1. A ground truth you can check yourself. Best available. Predict the answer, then look it up. Write your version, then read the published one. Estimate the number, then measure it. This is free and instant and almost nobody does it.
2. A person better than you at that specific sub-skill. Not a generally impressive person — the one who is an 8 at the thing you are a 4 at. Ask a narrow question: “Here are three openings. Which is worst and what specifically is wrong with it?” Narrow questions get real answers; “what do you think?” gets politeness. (Our Feedback Extraction Playbook is the long version of this.)
3. Your own work at a delay. Weakest but always available. Put the reps in a folder. Read them cold in two weeks. You will see the pattern you could not see live.

Step 4: The 30-minute block, four times a week

Two hours a week, protected, on one sub-skill at a time. Not two hours of “working on my craft” — two hours on the single lowest-scored sub-skill you can actually practise.

The structure of one block:

• 0–3 min: re-read last session’s note on what went wrong.
• 3–22 min: reps. Volume over polish. Fifteen mediocre attempts beat two careful ones, because you are hunting for the failure mode, not producing an artefact.
• 22–28 min: mark your own work against the ground truth or the model example. Be specific about the failure: not “weak,” but “I front-loaded context before giving the reader a reason to care.”
• 28–30 min: write one sentence about what to attack next session.

Run one sub-skill for four weeks — roughly 32 sessions’ worth of reps — then re-score and move on. Four weeks is long enough for a habit to change and short enough that you will actually finish.

Step 5: Raise the difficulty until you fail again

The moment reps get comfortable, they stop working. You are no longer practising; you are performing. The fix is to make the same rep harder along one axis at a time:

• Time: same task, half the time.
• Constraint: the opening, but under fifteen words. The pitch, without a slide.
• Audience: the same explanation for a hostile CFO instead of a friendly peer.
• Cold start: no notes, no warm-up, no second attempt.
• Stakes: do the rep in front of someone whose opinion you care about.

You want a failure rate around 20–30%. Zero failures means the difficulty is too low to teach you anything. Constant failure means you skipped a level and should back off one notch.

Four ways this goes wrong

1. You practise the 8s. The most common failure by a distance, because the 8s are pleasant and the 4s are humiliating. Every hour on a strength is an hour not spent on the thing that is actually capping you. If a practice session felt good throughout, you probably picked the wrong sub-skill.

2. You substitute consumption for reps. Reading about negotiation is not negotiating. Watching a teardown is not doing a teardown. Input feels like progress and produces none. Rule of thumb: at least three units of production for every one of consumption.

3. You never close the loop. Reps with no error signal are the most common form of wasted effort at this level. If you cannot say what specifically was wrong with rep number seven, the session did not count. Build the marking step in before the reps, not after.

4. You quit at week three. Deliberate practice on a weak sub-skill gets worse before it gets better, because you have started noticing errors you used to be blind to. Feeling less competent at week three is the expected result, not a signal to stop. It means the error signal finally arrived.

The one-page version

1. List 6–12 sub-skills. Score each 1–10. Pick the lowest one you could practise this week.
2. Design a rep you could do twenty times in an hour.
3. Decide the error signal before you start — ground truth, a specific person, or a two-week delay.
4. Four 30-minute blocks a week. Reps, then marking, then one sentence for next time.
5. Every week, raise difficulty on one axis until you are failing about a quarter of the time.
6. Re-score at four weeks. Move to the next lowest. Repeat.

Two hours a week, thirteen weeks a quarter, four sub-skills a year. That is the whole method. It is slower than a course and it is the only thing that reliably works.