The Practice That Feels Best Teaches You the Least
By Derek Neighbors on November 15, 2026
You sit down to get better at something. Maybe it is writing specs an agent can build from without guessing, or giving hard feedback, or judging whether an agent’s change is safe to ship. You pick one kind of problem and do it ten times in a row. The first few are slow. By the sixth you have the pattern. By the tenth you are fast and clean, and you close the laptop feeling sharp.
A week later the real thing shows up, and it does not arrive in a neat row. The spec you need covers a different kind of problem. The feedback conversation goes sideways in the first minute. The change the agents bring back mixes three problems you practiced separately. You are slower than you expected, and you wonder where the practice went.
The practice went into making that session feel good. The kind of practice that feels best while you are doing it teaches you the least, and researchers comparing blocked and random practice have been showing this for close to fifty years.
Why People Who Practice a Lot Still Freeze
Practice produces two things: the session and the skill. You can watch the session as it happens. You get faster, make fewer mistakes, and it feels like learning. The skill only shows up later, under different conditions, when nobody is handing you the same problem ten times in a row.
Grading practice by the session means judging the skill by how it looks in the moment, when what counts is whatever is still there a week later. Most people do it anyway, because the session is right in front of them. That would be fine if a good session reliably meant a good skill. It doesn’t. The two can point in opposite directions, and the research on practice order makes the point better than anything else. You will never control the order the real situation arrives in, so the part you control is how you practice for it and how honestly you grade that practice.
Blocked vs Random Practice: What the Research Found
Researchers split practice into two kinds. Blocked practice repeats one skill many times in a row before moving to the next. Random practice mixes several skills in an order you can’t predict. In 1979, John Shea and Robert Morgan ran a simple experiment comparing them. People learned three quick arm-movement patterns. One group practiced them in blocks: all the reps of the first pattern, then all of the second, then all of the third. The other group practiced the same patterns in a random order. During practice, the blocked group looked better. They were faster and more accurate. On the tests that came afterward, the random group won. They remembered the patterns better and handled a new pattern better. Researchers call this the contextual interference effect, and it has been repeated in many forms since.
The version I like best comes from baseball. In 1994, Kellie Hall, a kinesiology researcher at Cal Poly San Luis Obispo, and two colleagues worked with a junior college baseball team in California. The players took extra batting practice twice a week for six weeks against three pitches: fastballs, curveballs, and changeups. One group saw the pitches in blocks, all of one type and then the next. The other group saw them mixed, so they never knew what was coming. At the end, the mixed group hit better. They also hit better when the final test threw them pitches in blocks, the exact format the other group had trained on. The blocked hitters lost in their own format.
The third study should bother anyone who trains people, including themselves. In 2008, Nate Kornell and Robert Bjork had people learn to recognize painters’ styles from examples. Some saw each painter’s work grouped together. Others saw the paintings mixed. The mixed group did better at identifying the painter of paintings they had never seen. Afterward, most participants said the grouped method had helped them more. They held that view after the test showed the opposite.
That last finding explains why blocked practice survives. People pick practice by how it feels, the feeling favors blocks, and even evidence against the feeling does not change most people’s minds.
There is one honest exception. The effect is smaller for a complete beginner who cannot yet produce the basic move at all. Someone swinging a bat for the first time may need a short stretch of the same pitch before mixing helps. The mistake is staying in that stage once the move exists, and most people who practice something at work passed that stage long ago.
| Blocked practice | Random practice | |
|---|---|---|
| Order | One skill repeated, then the next | Several skills mixed, order unpredictable |
| During the session | Looks better: faster, fewer misses | Looks worse: more misses, no rhythm |
| On later tests | Remembered less, transfers less | Remembered more, transfers better |
| How it feels to the learner | Feels more helpful | Feels less helpful, even after it wins |
| What each rep trains | Running a move someone already named | Reading the situation, then choosing the move |
Why the Ugly Reps Work
Blocked practice lets you skip a step. When the next rep is the same as the last one, you never have to figure out what you are facing. You already know. You just run the move again, a little cleaner each time.
In the game, figuring out what you are facing is most of the job. A hitter gets a fraction of a second to read the pitch before choosing a swing. An engineer looking at a production failure has to recognize which of five familiar problems it actually is before pointing the agents at it, and a manager in a tense one-on-one has to notice when a performance conversation turns personal. Blocked practice never trains that recognition. Random practice trains it on every single rep, because each one forces you to read the situation, pull up the right response, and then execute it. That is why the session feels worse. You are doing the hard part every time.
Aristotle saw the general version of this a long time ago. In the Nicomachean Ethics he says we become builders by building and lyre players by playing the lyre, and then adds that both good and bad lyre players come from playing. Aristotle was talking about playing well or badly, and the order you practice in is one of the ways you practice well or badly. Reps alone guarantee nothing. They build a settled habit, which Aristotle called a hexis, whatever kind of practice you do. The kind of practice decides which habit you end up with: ten clean reps in a row teach you to execute once someone has named the problem for you, while mixed reps make you name it yourself every time.
I run in the desert most days, and the trail forces this on me whether I want it or not. Sand, rock, a dry wash, a loose switchback, and the footing changes every few steps. A treadmill would give me smoother miles, and years of uneven trail have taught my feet to read the ground in a way smooth miles never would have.
Choosing the Ugly Rep Is a Character Choice
If the research is this clear, why do so many people still practice in blocks? Because the alternative asks you to look worse, to yourself and to anyone watching, for longer. Random practice means more misses during the session. It means ending a practice block without the satisfying run of clean reps that makes you feel competent. For someone who is used to being good, that is uncomfortable, and most people drift back to whatever lets them feel competent again.
In Book 3, chapter 12 of the Discourses, Epictetus talks about training, askesis, the word the Stoics used for deliberate practice of character. He mocks training that is really a show, like walking a tightrope, which earns admiration and builds nothing anyone needs. He tells his students to aim their training at what they personally get wrong. If you lean toward something that harms you, practice against that lean, because that is where the work is.
Blocked practice works like that tightrope: it looks good and feels good, and it trains the part of the skill that was never the problem. Choosing the mixed set, the rep you will probably miss, is choosing to train the part that will actually fail you later. That choice is about honesty. You give up feeling good about the session so the skill holds up when it matters.
I have written before about designing a handicap on purpose and about rehearsing the specific moment you dread. Practice order is a cheaper move than either one. You do the same reps, in a different order, and accept that the session will look worse.
The Model Makes Random Practice Cheap
People also say mixing practice is too much work. Somebody has to build the varied problem set, shuffle it, and make sure the hard cases show up as often as the easy ones. That was always possible by hand, and the people who skipped it chose the easy order. The tedium gave them a reason that sounded better than “I wanted the session to feel good.”
A model removes that excuse, and I would leave it in the loop permanently. It can shuffle the reps for you, and doing them is still your call. Ask it for twenty problems that mix the five types you are trying to learn, in an order you can’t predict, and tell it not to group them. Have it play the person you dread talking to, and tell it to change tactics partway through so you can’t settle into a script. If you build agents, the same rule applies to your evals: an eval set that runs the same five clean prompts in the same order tells you as little as a hitter who only ever sees fastballs. Mix the cases, include the ugly ones, and let the model generate the variations you would never think to write.
A model produces variety fast and has no interest in making you feel good about the session, which removes both of the comforts blocked practice was giving you.
What to Do This Week
-
Mix the order. Take whatever you are practicing and stop grouping it. If you are learning three kinds of problems, do them shuffled. If you are working on three conversations, rehearse them out of order. Same reps, different sequence.
-
Expect the session to feel worse, and keep going. You will miss more and feel less fluent. Write that down before you start, so that when it happens you read it as the practice working.
-
Grade yourself a week later. The end of the session is the worst moment to judge what you learned. Test yourself a week later on a mixed set you have not seen. That score is the one that predicts the game.
-
Let a model shuffle the reps. Have it build the mixed problem set, play the hard conversation with changing tactics, or generate the eval cases you would skip. Keep it in your practice from now on, so that nobody, including you, can slide back to the easy order without noticing.
-
Practice the choice as well as the move. Before each rep, say what kind of problem you think it is before you start solving it. Recognition is the skill the game asks for first, and saying it out loud makes you train it on every rep.
FAQ
What is the difference between blocked and random practice?
Blocked practice repeats one skill or problem type many times in a row before moving to the next. Random practice mixes several skills or problem types in an unpredictable order. Blocked practice produces better performance during the session. Random practice produces better retention and better performance when the real situation arrives in a mixed order, because every rep forces you to recognize what you are facing before you respond.
Why does random practice feel worse if it works better?
Random practice feels worse because you make more mistakes and never settle into a rhythm. Each rep asks you to figure out what kind of problem you are facing and then pull up the right response, which is harder than repeating the same response. That effort is what builds the skill, but in the moment it registers as struggle. Studies have found that learners rate grouped practice as more helpful even after a test shows mixed practice taught them more.
What is askesis in Stoic philosophy?
askesis is the Greek word for training or exercise, and the Stoics used it for the deliberate practice of character. Epictetus told his students to aim their training at what they personally get wrong instead of at feats that impress other people. The word carries the idea that character is built the way an athlete builds a body, through reps that are hard on purpose.
Final Thoughts
The best practice session you had this month may have taught you the least. It felt good because nothing in it asked you to figure out what you were facing, and that is the part the real situation demands first. The junior college hitters who never knew what pitch was coming looked worse for six weeks and then beat the other group at its own test. Aristotle’s warning still holds: both good and bad lyre players come from playing. The reps are going to build a habit either way. Choose the ones that feel ugly, and let the session look worse so the skill holds up when someone is actually pitching to you.
Practice that holds up under pressure is easier to build with people who will mix the reps for you and tell you when you are hiding in the easy ones. That is what MasteryLab is for.