Autonomy, Competence, Relatedness: The Motivation Theory That Kept Working

I have started more habit trackers than I have finished habits.

The pattern is always the same. Something I genuinely wanted to do — read more, sit for meditation, learn a piece of mathematics properly — gets a system built around it. Streaks. A spreadsheet. A number that goes up. For a while the number going up feels great. Then one week the number doesn't go up, and instead of the mild disappointment you would expect, the whole thing collapses. Not just the streak. The reading itself.

That collapse is not a willpower failure, and it is not a mystery. It has a name and a literature, and the literature is older and steadier than almost anything else in motivation science.

The Three Needs, in Plain Language

Edward Deci and Richard Ryan, working at the University of Rochester, spent decades assembling what became self-determination theory. Its modern statement landed in a 2000 paper in American Psychologist. Twenty-six years on, the core claim is still standing while a striking amount of what surrounded it in the same era has quietly been withdrawn.

The claim is that human beings have three basic psychological needs, and that motivation which satisfies them lasts while motivation that starves them does not.

Autonomy is the sense that your actions come from you. This is the one most people get wrong, so it is worth being precise. Autonomy is not independence, and it is not having lots of options. It is volition — the feeling that you endorse what you are doing rather than being pushed into it. You can be autonomous while following an instruction, if you agree with the instruction. You can be crushingly non-autonomous while choosing freely from a menu you never wanted.

Competence is the sense that you are effective, that your effort connects to an outcome, that you are getting better at something. Not that you are good — that you are improving, and that the improvement is visible to you. This is why a difficulty level that is slightly beyond you is more motivating than one that is easy, and why a task where feedback never arrives is demoralising regardless of how well you are doing it.

Relatedness is the sense that you matter to other people and that they matter to you. It is the quietest of the three and the one that gets designed out of things most often. A goal pursued in complete isolation is not necessarily doomed, but it is carrying a weight the other two cannot lift.

The framework's power is that it changed the question. Most of what gets written about motivation asks how much of it you have, as though it were a fuel level. Self-determination theory asks what kind you have, and treats the kind as the thing that predicts whether you will still be doing this in a year.

Why Paying Someone Can Stop Them

In 1971 Deci ran an experiment that looks almost too simple to matter. Students worked on Soma cube puzzles, the sort of spatial problem people do voluntarily for fun. One group was paid per puzzle solved. The other was not. Then the experimenter left the room for a "free choice" period, with the puzzles still on the table and magazines available as an alternative.

The unpaid students kept playing with the puzzles. The paid students, once the money stopped, put them down.

Paying people to do something they already enjoyed made them enjoy it less. This became known as the overjustification effect, and it was controversial for a long time. A meta-analysis by Deci, Koestner and Ryan in Psychological Bulletin in 1999 pulled together well over a hundred experiments and found the undermining effect held up, particularly for tangible rewards that were expected in advance and contingent on doing the task.

Those qualifiers do a lot of work, and honest writing about this has to keep them. Not all rewards corrode. The corrosive ones are tangible, anticipated, and tied to performance — the ones that shift your reason for acting from inside to outside. Unexpected rewards do not do this. Neither does verbal feedback that carries information about how you are doing, provided it isn't delivered as pressure. Praise that says "you did that well" supports competence. Praise that says "good, keep it up, I expect this every week" is a leash with a compliment on it.

It is also worth saying that the undermining effect has real critics. Judy Cameron and David Pierce published meta-analyses arguing the effect is narrower and more conditional than the SDT camp claims. The debate has not fully closed. What survived the argument is the conditional version: rewards that feel controlling reliably damage intrinsic motivation, and rewards that feel informational generally don't.

Motivation Is Not a Switch

The most useful part of the theory, and the part that rarely makes it into summaries, is that motivation is not a binary between intrinsic and extrinsic. It sits on a continuum of how thoroughly a reason has been absorbed into who you are.

At the bottom is external regulation — you do it because someone will reward or punish you. Just above that is introjected regulation, where the pressure has moved inside but is still pressure: you do it to avoid guilt, shame, or the feeling of being a person who breaks their promises. Then identified regulation — you do it because you genuinely believe it matters, even though the doing is not enjoyable. Then integrated, where the reason has merged with the rest of your values. And at the far end, intrinsic: you do it because doing it is its own reward.

Notice what this rescues you from. The advice to "find what you love and you will never work a day in your life" is not just useless, it is actively misleading, because most things worth doing are never going to be intrinsically fun the whole way through. Filing taxes. Physical therapy. The third hour of debugging. Learning the mathematics under a method rather than importing the library.

The realistic target is not intrinsic motivation. It is identified regulation: I do not enjoy this, and I am doing it because it is mine to do. That is a durable place to stand, and it is reachable for almost any worthwhile task. Intrinsic is a gift when it shows up. Identified is something you can build.

Streaks, Points, and What Gamification Actually Does

Now look at the app on your phone with this vocabulary in hand.

A streak is introjected regulation, manufactured at scale. It works by making you dread the loss. That is genuinely effective in the short term — loss aversion is a strong lever, which is exactly why it gets pulled. But it also relocates the goal. After a few months the thing you are protecting is the number, not the practice, and the two have come apart without anyone announcing it.

Which explains the failure mode everybody has lived through. You miss one day. The streak resets. And instead of shrugging and continuing, you stop altogether — because the thing that was actually holding the behaviour up just got deleted, and the behaviour was resting on it rather than on anything of your own.

Points and badges are more ambiguous, and their effect depends entirely on framing. A number that tells you something true about your progress supports competence: it is feedback. The same number, presented as a target you owe someone, becomes control. Leaderboards are the sharpest case — genuinely energising for the handful of people near the top, quietly corrosive for everyone else, because being ranked twelfth is information about your standing rather than about your improvement.

The design question worth asking of any app that wants your daily attention: does this tell me something about my own progress that I could not otherwise see, or does it manufacture a small debt that I now have to service?

What AI-Assisted Goal-Setting Does to the Three Needs

This is where the framework earns its keep on something it was never designed for.

Hand a language model your vague ambition and it will hand back a structured plan: milestones, a weekly cadence, a definition of done. The plan is usually good. That is the problem worth naming.

The autonomy risk is straightforward. If the system picks the goal, decomposes it, sets the pace, and prompts you when you drift, then the volition has been outsourced along with the planning. You end up executing a plan you did not author, which is precisely the configuration that produces compliance and then quitting. The fix is not to refuse the help. It is to keep the choosing. Let the model generate three plans and pick one yourself, or better, tell it what you have decided and ask it to find the holes.

The competence risk is subtler. Competence is a feeling that comes from having been effective, and being effective requires that some of the difficulty was actually yours. If the hard part is always handled elsewhere, the plan finishes and the feeling of having grown never arrives. I notice this most in my own work: the code I understand is the code I struggled with, and the code that arrived complete is code I will have to learn twice — once when it was written, and again when it breaks.

Relatedness is the one no model supplies. It cannot be your accountability partner in any sense that matters, because nothing is at stake for it. This is not a criticism of the tools; it is a description of what they are. If a goal needs relatedness — and most sustained ones do — that part still has to come from a person.

A Short Audit: Score a Goal or a Job

Pick one thing. A current goal, a job, a course you are halfway through. Answer each statement from 1 (strongly disagree) to 5 (strongly agree). Be honest rather than aspirational; the point is diagnosis.

Autonomy

1. I would choose to do this even if nobody were watching or counting.
2. I have real say over how and when I do it, not only over whether I do it.
3. When I explain why I am doing this, the reason is mine rather than someone else's.

Competence

4. I can tell whether I am getting better at this.
5. The difficulty is roughly at the edge of what I can do — not trivial, not crushing.
6. I get feedback that tells me something useful, from the work itself or from someone who knows.

Relatedness

7. At least one person I respect knows what I am working on and why.
8. This connects me to people rather than isolating me from them.
9. If I disappeared from this tomorrow, someone would genuinely notice.

Add each pair of three. You now have three scores out of 15. The total matters less than the shape — a 13, 12, 4 is a completely different problem from a 7, 7, 8, and they need opposite interventions.

What to Do With a Low Score

Autonomy low. Find the part of this you would still choose and make that part explicit. If a job is the subject, autonomy over method is usually more available than autonomy over goal, and it is nearly as protective. If you cannot locate any version of the reason that is genuinely yours, that is real information about whether to keep going.

Competence low. Almost always a feedback problem before it is a skill problem. Shorten the loop. Find the smallest unit where you can tell whether you did it well, and make that visible weekly. If the difficulty is the issue, adjust the task rather than your effort — grinding at something far above or far below your level burns motivation in both directions.

Relatedness low. The cheapest fix in the whole framework, and the most neglected. Tell one person what you are doing. Not a public commitment, not an accountability app — one person who will ask you about it in a month. The mechanism is not surveillance. It is that the thing stops being purely private, and things that live only in private are easy to quietly abandon.

There is a caution about universality worth carrying. Research on choice across cultures — notably work by Sheena Iyengar and Mark Lepper — found that children from different cultural backgrounds responded very differently to personal choice, with some doing better when a trusted adult chose for them. SDT's answer is that autonomy means volitional endorsement, not personal selection: endorsing your mother's choice is autonomous if you genuinely endorse it. That answer is coherent, and the debate about how far it stretches is still live.

What I keep coming back to is the sentence underneath all of it. The question is not whether you are motivated enough. It is whether the reason you are doing this is one you would still give if nobody asked.

Questions People Actually Ask

Does this mean I should never use rewards?

No. Rewards work fine for dull, mechanical tasks where there was no intrinsic motivation to damage in the first place. The caution applies to activities someone already finds worthwhile. Even there, unexpected rewards and informational feedback are largely safe. It is the expected, contingent, controlling ones that do the harm.

Is autonomy the same as doing whatever I want?

No, and this is the most common misreading. Autonomy is endorsing your own action. A person who follows a demanding routine they chose is autonomous. A person who drifts through the day answering whatever pings loudest is not, despite technically doing as they please.

What if my job scores badly on all three?

Then the score is telling you something you probably already knew. Before quitting, check what is actually movable — method autonomy and feedback quality can often be negotiated in ways that goal autonomy cannot, and relatedness can be built sideways with colleagues rather than granted from above. If none of it moves after a real attempt, that is your answer.

How is this different from intrinsic versus extrinsic motivation?

That binary is the simplified version, and it is the reason the theory gets misapplied. The continuum from external to introjected to identified to integrated is where the practical value sits, because identified regulation — doing something unfun because it is genuinely yours — is where most durable adult effort actually lives.

Has any of this failed to replicate?

The core has held better than most of its contemporaries, but it is not untouched. The magnitude of the undermining effect is contested, and the cross-cultural universality of the autonomy need has been seriously challenged. Treat the three needs as a well-supported working framework, not as settled physics.