Caffeine is the most widely used cognitive drug in the world and the discussion of it tends to be either promotional or prohibitionist. Neither is useful. It is a real drug with real effects, some of which are genuinely valuable and some of which are borrowed.
What it does
It restores alertness in someone who is depleted. This is its main effect and it is real. It blocks the signalling that accumulates through waking hours and produces sleepiness.
It improves vigilance and reaction time, particularly when tired.
It helps sustained attention on dull tasks.
It has a mild mood effect for many people, which is part of the appeal.
What it does less well than it feels like
It restores the feeling of alertness better than the function. This is the crucial asymmetry. After insufficient sleep, caffeine returns the subjective sense of being fine while leaving a substantial part of the performance deficit in place. You feel restored and you are partially restored — and you are no longer receiving the signal that would have told you otherwise.
It does little for complex cognition in someone already well rested. Most of the measurable benefit is in reversing depletion rather than in enhancement.
Tolerance develops. Regular use means a large part of what a morning coffee is doing is preventing the withdrawal that has accumulated overnight. That is not an argument against it — it works — but it changes what "it helps me focus" means.
The specific problem
Caffeine is the main reason people do not notice a sleep problem.
The sequence is familiar. Sleep is inadequate. Mornings are difficult. Caffeine fixes the morning. The afternoon is hard, so more caffeine. That caffeine is still present at bedtime, so sleep onset is later or sleep quality is lower. The next morning is harder.
Each step is reasonable. The loop is stable, self-reinforcing, and it can run for years, with the fog attributed to age, work or something medical.
The reason it matters clinically: someone in this loop presenting with cognitive symptoms has a treatable contributor that no test will find, because the deficit is masked well enough that nobody suspects sleep.
The half-life, which is longer than people think
Caffeine takes several hours to reduce by half in most adults, and the variation between individuals is substantial — some people clear it much more slowly, and the difference is largely genetic rather than a matter of tolerance.
The practical consequence: a mid-afternoon coffee leaves a meaningful amount in your system at bedtime. It may not stop you falling asleep. It can still reduce the quality of the sleep you get, which is the part you do not observe.
People who say caffeine does not affect their sleep are usually reporting correctly on sleep onset and have no information about sleep architecture.
A reasonable arrangement
Not abstinence. A few things that make it work rather than work against you:
A cut-off, earlier than feels necessary. Early afternoon is a common recommendation. If you clear caffeine slowly, earlier.
Not immediately on waking. There is a common suggestion to delay the first one, and the evidence for the specific mechanism often cited is thinner than the confidence with which it is asserted. The defensible version: if you need caffeine to function within minutes of waking, that is worth noticing as information about your sleep rather than solving with better timing.
A steady amount rather than an escalating one. If the quantity has been climbing over months, that is a signal about something upstream.
Know your actual intake. Including tea, cola, energy drinks, pre-workout supplements and chocolate. People routinely underestimate by a factor of two.
The test
If you want to know what caffeine is doing for you rather than to you, the experiment is two weeks.
Reduce gradually — abrupt cessation produces headache, fatigue and irritability for several days, which will dominate the first week and tell you nothing.
Then two weeks at a low or zero level, with one line a day on how the day went and how you slept.
What people typically find: worse for four or five days, then sleep improves, then daytime alertness is better than expected — sometimes better than it was with caffeine. Some people find the opposite, and that is a real answer too.
Either way the result is worth having, and it is a rare case of a self-experiment that can actually be run.
What this is not
Not a claim that caffeine is bad. For most people it is a reasonable, effective, inexpensive drug with a well-characterised profile.
The point is narrower: if you have cognitive symptoms and you are using caffeine to get through the day, you have an unexamined variable in the middle of the picture, and it is the one variable that is specifically capable of hiding the answer.
Worth mentioning your actual intake in an appointment, and worth mentioning if it has been increasing. Neither is a confession. Both are information.