Closed loop is not automatic
Hybrid closed loop automates the background and not meals. What it genuinely handles, what it cannot know, and why expectations decide how it feels.
Hybrid closed loop is the closest thing to automation that exists, and the word people hear is automatic. The name it is given in the guidance is more honest: hybrid, meaning it does one half by itself and you still do the other. Knowing which half is which is the difference between it working brilliantly and feeling like a disappointment.
- It automates the background, adjusting a trickle of insulin minute by minute against your sensor readings.
- You still tell it about food. That is the hybrid part, and it is the part people underestimate.
- It reacts to where you are now; it cannot know about the meal you are about to eat or the match you are about to play.
- Sensor and set problems still stop everything, and the system is only as good as the readings it is being given.
- It is genuinely very good, and expecting no involvement is the thing that makes it feel disappointing.
The half it genuinely handles
The background is where it shines, and the reason is the mismatch it fixes. A fixed background rate is a guess made weeks ago about what you would need at three o’clock on an unspecified Tuesday. A loop looks at your actual glucose every few minutes and adjusts, nudging up when you are drifting high and easing off or stopping when you are heading down.
The thing that follows from that is overnight, which is where most people notice the difference first. Nights are the long stretch where nothing but background is acting, so continuous small corrections have hours to work, uninterrupted by meals. Steadier nights and fewer overnight lows are what people report, and that alone is a substantial change in how much diabetes intrudes.
It also handles the slow drifts you would otherwise be chasing: an unusually lazy day, the tail of an illness, a background that is slightly wrong this week. Those get absorbed rather than becoming a job.
Why it still needs you at meals
The limitation is simple and unavoidable: a loop can only react to what has already happened. It sees your glucose starting to rise after a meal, but by then the food is already in and rapid insulin takes time to work, so catching up is much harder than getting ahead. That is why you still announce meals rather than letting it notice them.
The same applies to anything predictable. It cannot know you are about to play football, about to sit an exam, or about to go to bed after a big session. Most systems have a way of being told, an activity or exercise setting, and using it as designed is far better than letting the loop discover the situation halfway through.
And it is downstream of the sensor. If the sensor is wrong, the loop is confidently wrong, because it is acting on those numbers. A sensor that is failing, compressed, or in its unreliable first hours produces decisions built on bad information, which is a different failure from a pump simply stopping.
Why expectations decide how it feels
People who arrive expecting an artificial pancreas often feel let down, and people who arrive expecting help with the background are usually delighted. The technology is identical. What differs is what they were told to expect, and the marketing language around this is a good deal more confident than the guidance is.
There is also a settling-in period that catches people out. Loops learn from your data, and the first weeks are frequently unremarkable or worse than what you had before, which is the point at which people conclude it is not for them. Sticking with it through that, with your team involved, is usually the right call.
The other thing worth naming is that it changes what you do rather than removing it. Less firefighting and fewer overnight corrections, but still carbohydrate counting, still set changes, still sensor management, and now a system with its own settings and behaviours to learn. Different work rather than no work is the honest description.
What affects how well it works
Announcing meals consistently makes more difference than any setting, because the meal is the half it cannot see coming. Absorption matters more than people expect: a loop giving insulin into a site that is not absorbing gets no feedback that its decisions are landing. And time helps, since most systems improve as they gather more of your data.
Where the line falls
Examples, not instructions or doses.
A steady overnight line after a day that was anything but. Hours of small adjustments with no meals interrupting is exactly the situation loops are best at.
A spike after eating without announcing it, because the loop only reacted once the rise had started and insulin takes time. Not a fault, just the half it cannot see coming.
A failing sensor reading low, so the loop reduces insulin, and you climb for hours. The loop did what the numbers told it, which is why sensor quality matters more here.