The Auto Mode experience: 670G two weeks in


Checking back in, here at the end of my second week on auto. Thought it might be interesting to give a more experiential version of what it’s like.

First off, It looks the like a regular pump, but it ain’t. I’m just starting to get my head around that. At the macro level, the biggest difference is that you can’t wing it, the way you can with a regular pump. If you’re used to a lot of hands-on, stick ‘n’ rudder control over what your pump is doing, as I have been, this is a big adjustment. F’rinstance:

Bolusing for meals or corrections.

  • This is a Wizard-only proposition. The system always needs a “why” for whatever you’re doing, in the form of carbs or meter BG or sensor values, so it can adjust itself accordingly, therefore there is NO Manual bolus. You *can’t just decide, “Y’know, I think I need an extra unit” for whatever reason and go plug that in." This has been especially difficult these early days where settings aren’t dialed in yet and I’m running higher than I want. It sets its own correction values and you can’t just hit the up arrow to make it give you more. The algorithm has to learn you need more, and what it is allowed to do to get you back on track is limited by your settings. This is especially frustrating at the beginning, because they want the algorithm to do its thing for a while before deciding, ok, you can change one of those settings and see if that does it. Means kind of a long haul getting it dialed in.

Basal doesn’t mean basal

  • This is subtle and a bit elusive to get my head around. Those big square wave basals have been replaced by these little continuous blips, right? And at first you’re inclined to think of that as your basal rate, but on your old pump that was just a continuous increment based on an informed guess about what it’s going to take, meals notwithstanding, to keep you in a certain range. The pump doesn’t have any way to know if you’re in that range. In Auto, though, it does know where you are, and it’s continuously trying to pull you down to, or pause so you float up to, that magic 120. This is a different logic and impinges on a lot of things. Such as…

Sensitivity factor isn’t a factor.

  • This one still perplexes me. You do have a setting for it, but it’s only operative in Manual Mode. In Auto, it doesn’t do anything. Drives me a little crazy because I want to adjust the ratio of how many BG points a unit insulin of insulin brings me down so I can force the thing to give me MUCH bigger corrections than I’m getting now. But instead, they want the algorithm to figure out what it has to do to reach the magic 120 (actually 150, since that’s what it targets when figuring correction boluses, then the microboluses are supposed to get you the rest of the way to 120). Well, ok fine, sez I, but when I do a BG check and I’m at 240, and the pump says “You need a correction!” and I sez “Yes, hit me with a big one!” and it blips out ~1 paltry unit when my experience tells me I need more like 8 - 10, it’s a bit maddening. I’m still trying to get more information on this one but the partial answer I do have is that the Algorithm (oh mighty Algorithm!) just hasn’t learned me yet. Which leads into…

Don’t exercise, just sit there!

  • This is apparently a mistake I made, my first week-plus in Auto: I’ve been exercising down my highs, since the pump isn’t doing it on its own. The problem is that that prevents the pump from learning that IT needs to work harder, so some of the learning I was hoping it would do… it wasn’t doing. Therefore I’ve been deliberately not exercising the last few days, because I want the baseline to reflect what I need when I’m not going for a strenuous walk after lunch every day. Been also holding off biking for the same reason. Not for much longer though, dammit.

Bottom line is that I’m still curious enough to stick with the program and see if I can get out the other side to having the results I’m seeing from some of the more experienced people in the 670 user groups. So hopefully I’ll come back to report things going more the way I’d like next week.

BTW, if interested, my post about my first weeks (auto and pre-auto) are here:

I’ve also posted these on FUD, where there are some different questions/responses that may be of interest:

Awful customer support: 670g frustration
Anyone have a 670G?
New pump tomorrow

I’m not a pump user – I’ve been on the MDI/pump fence for many years – but I sure do appreciate your comments here and elsewhere about your experience with this system. Having to endure higher-than-desired bg’s for weeks to generate the data and allow “the algorithm” to complete its calibration (that sounds a little ominous) is way above and beyond the normal D routine. For those of us who are always interested in the latest and greatest, here’s a bow and a hat tip to you!


Thank you, @DrBB, for updating us. Since the 670G was introduced, there’s been a thirst for user reviews and they are scarce! This is really generous of you to take the time to go into so much detail to share with us.

Caleb is looping and we went through weeks of what I thought was insanity of bg results trying to adapt to a new way of pumping and using correction factors, in particular. We made it worse on ourselves by attempting this through illness and other variables that only complicated things. It took longer than I would have liked, but we are on auto mode now - auto mode in this case meaning not giving much thought at all to settings or bg for that matter. It was tough to get here, but we’re happy with the results now that we’re on the other side. What you describe sounds a little like what we went through. At what point do you expect (do you have any sense) to see a little more auto-mode-getting-you-in-your-desired-range mode take action? :slight_smile:


Yeah, with this system too it seems like a list of “DON’TS” while still training the thing would be helpful. A big one for me, belatedly, was that “Don’t exercise” one, but there’s only so long I’m willing to put my ordinary life on hold. The “learning” week before going into auto I had also avoided a couple of the higher-carb lunches I usually have, and was told I shouldn’t have done that, so you can’t win. I think the whole program is oriented toward non-LCHF practitioners, but then that’s true of T1 treatment generally.

YES! This is the better news. I evidently passed some kind of threshold over the weekend, because suddenly I started seeing “Max Delivery” alerts, indicating the pump was trying to be more aggressive and running up against some kind of limitation. I went in and raised the Max Basal setting… and this is where The First Rule Of Algorithm Fight Club stepped in again. What is TFROAFC, you may well ask? We can only answer by mixing our movie metaphors, to whit: “That word doesn’t mean what you think it means.” Apparently the Max Delivery setting that I tweaked only applies–yet again–to Manual Mode operation. Whereas when Auto Mode tells you it has hit Max Delivery, it is referring to a limit set by the algorithm itself; it doesn’t give a fig for the one in Options: Delivery Settings. Such matters are beneath its notice. It eschews them. For haughty and strange are the ways of Holy Algorithm, and yet we must put all our trust therein. Or something.

Meanwhile, Dawn Phenomenon, its mortal enemy, continues to sit in its fastness, maintaining its iron grip on the 3:00 - 10:00 a.m. countryside, and exacting a harsh tax from the inhabitants thereof. Better numbers this a.m., but it’s frustrating to see my overnights ticking along very nicely in the low 100s/upper 90s, and then start ratcheting up and up from 3am on, while the algorithm is apparently catching a few z’s itself.


Activity was definitely one of the variables that complicated getting our settings straight. Variability in food as well. Those things I just couldn’t change. Caleb’s got to live his life and although it made it more challenging, oh well. I wish I opened the loop and waited things out during his illness though - that’s another story including medications that drove things wonky (a hell I wouldn’t wish on anyone).

So then what happens when it hits max delivery in auto mode? It that just an FYI to let you know it’s going to rethink things bc it sees it’s currently not keeping up?


If the logic seems a bit tail-swallowing that’s because it is! But that’s my understanding, yes.


I share your questioning here. So the pump helpline and various 670 Facebook user groups seem to be counseling holding back on exercising with the thinking that the system needs to learn your baseline non-exercising state first. Seems like the algorithm should be able to learn this in 3-4 days but your writing suggests this has been going on for a few weeks. If you had just kept up your exercise routine, wouldn’t the system have learned that your baseline includes exercise?

With Loop and exercise, the tactic I use is to raise my target blood glucose range from 80-90 mg/dL to 90-100 mg/dL. It seems to help if I do this about 30 minutes before I start to exercise. My exercise is likely not as taxing as yours. I spend 30-40 minutes on an exer-cycle. My preferred activity of walking has been set aside due to fasciitis and/or tendonitis issues. I’m working to rehab my feet and hope to return to my former walking soon.

I appreciate your reporting on this and you are likely gaining insights as writing forces you to a certain level of coherence. I sympathize with your not being able to see and understand fully the system you’re using. I’ve been using Loop for over nine months now and I feel I’ve not learned all the levers available to me, partially due to me finding and using some good workarounds. My workarounds however mean more manual attention and action, one of the things users and designers likely want to avoid.


Well, the thing is that in that first week on Auto I WAS using exercise to kick things down, so the pump wasn’t learning that it needed to work harder. It came up with my trainer at my first follow-up session after going into Auto (a little less than a week ago as I’m writing) so it hasn’t been that long.

I thought about that but the thing is my exercise practice isn’t that regular. Sometimes I bike commute, sometimes not, depending on weather, do I need to drive to an appointment, etc. Similarly I don’t want to ALWAYS have to go for a walk after lunch or else have my BG go crazy. But there is a way in which the more regular in your habits you are, the easier a system like this will be able to manage. In the extreme case it starts getting to be like back in the old R/NPH straightjacket: gotta organize your whole life to the insulin, not the other way around. Which would be pretty ironic.

Got that right!

The closed user group has a list of “Questions Experienced 670G Users believe Medtronic Needs to Answer” that I just downloaded. A lot of 'em are already familiar as questions I’m trying to answer here.


I think we are in the very early days of closed loop systems. It’s a fairly complex and sophisticated thing that designers have tried to accomplish.

  • They’ve designed a system around their understanding of blood glucose metabolism while trying to listen to feedback from users.

  • As a commercial system that needs to satisfy the FDA gatekeeper at the outset, it means a system that prioritizes a “batch” approach that prizes getting it right with the first formal attempt rather than the luxury of an iterative approach that the DIY community enjoys.

  • Feedback, both commercial and DIY, is subject to the frailties of human communication as well as the “your diabetes may vary” truth that we know exists.

  • I suspect that few on the Medtronic technical team have a comprehensive understanding of how the whole system works. I’m sure that even the small subset of people with an excellent understanding disagree with one another. Oh, to be a fly on the wall of that room!

  • I suspect that we will look back on these times, ten years hence, with a clearer picture of what’s really going on. At this point, I think that even the experts feel they’re in a bit of a fog.

  • We are indeed guinea pigs, but it is for a good cause!


As someone who is currently using Humulin N/R, and is about to start their new 670g on saline … today! … and has their insulin pump insulin-start next week, I want to say THANKS! for the information!!! It should be very useful.


@DrBB - Obviously you have interest and experience far beyond the average person who would use a pump.

In your opinion, is this solution feasible for somebody who is not going to get this involved and have this level of insight?



If I could offer my .02 (been in auto for about 2 months now)-

First, your comment of wanting to add a unit or two to a bolus, but the wizard wont let you…my workaround is simply adjust the carbs until I get the bolus I want. Its simple, really :slight_smile:
As far as corrections go…it is my understanding that basals given in auto do NOT show up in IOB (thats a fact), BUT! When youre correcting highs, they ARE. The 670G knows how much it has already been working on that high via auto mode. I have made the mistake several times in adding an additional bolus correction on top of what it suggests, only to find myself low several hours later. My instructions to myself are, LET THE TECHNOLOGY WORK! I have found it WILL correct you down to normal if you leave it alone. Does it take time? Yep it does. But it will.

Not sure if you know this but on your main screen if you hit the chart key to see the chart of your BG’s, anywhere along the top you see a pink dot, that means it has given you a “basal” amount. If you use the keypad and move it left to each one, it will tell you how much and what time it gave that amount. I have seen it give as little as .050u and as much as .400. Remember, it can give a basal amount every 5 mins (when its updated from the Guardian sensor).

Hope that helps!


Which is what a lot of people do, but I’m trying to stay within the guidelines to, as you say, let the technology work. I was specifically told not to do this, so as to force the algorithm to truly learn what’s going on, but it’s a strong temptation, especially in the a.m.

Well, yeah. But if it keeps taking as long as it did today, letting me get up to a post-prandial 250 and only getting me down to 210 by dinner time, I’m gonna be looking at going from a 6.0 A1C to a 7.5 or worse–not the kind of result I’m hoping for with this thing. My MDT trainer advised me to keep requesting corrections when high as a legit way of training it to be more aggressive. Adjusting Active Insulin Time is also supposed to be a big help, but I’m still at the stage of wanting my trainer to ok it first. Not sure how much longer I’m going to hold back on taking those matters into my own hands though.

Yup. Or zero out entirely. I’ve hit that limit a few times when I was using exercise to knock down the highs that it didn’t want to treat aggressively enough, but my trainer advised me to lay off doing that so it would get a baseline not dependent on me going for a strenuous walk or bike ride.


Interesting question. I think the intention is that you end up spending a lot less time thinking about the big D because it’s doing such a great job handling things, and that does seem to be a reality for people who’ve been using it for a while. One thing is that it might be easier to learn if you don’t have to unlearn the logic of a standard pump, which is part of what I’m struggling with. @Yve65 might be a good person to ask in two or three weeks. Humulin N/R to 670G–that’s a heck of a leap!


@DrBB agree with everything. BTW regarding active insulin time…with every other pump Ive had Ive always set it at 3.5 hours, as that is what my testing has shown; however, on the 670G, Ive got it set to 2 hours 15 mins. Reading what few user experiences there are, this is actually pretty common. 2 hours seems to be popular.

YMMV of course.


Yikes, that is a super conservative algorithm, I think I would go crazy with that. To be fair, I recently had that type of persistent 200+ situation after a big Thai meal with friends – I just couldn’t resist – but for a relatively normal day I’d be looking at what was wrong with a sensor, with the insulin, or something…


Same here. We’re just starting to tweak things. I tightened up the AIT setting 24 hrs ago and it has helped. There are a few things other users have recommended that I haven’t tried yet b/c I’m keeping to my trainers’ recommendations, but I’m at the point where I feel like I know it well enough that I’m going to take matters into my own hands soon.


I’ve been on the system since the first of July and have had many of the same problems mentioned in most of these posts. I still don’t like the fact that it wants me at 150 instead of 100, but maybe that will happen with the next upgrade. Please, please, please contact your trainer. I have spoken to mine multiple times via phone, email and in person. Every time I see her, we tweek my carb ratio and alter my basal rates based on my pump downloads via carelink. This has made a huge difference in my averages. Despite more elevations after certain meals, my A1c is now 6.4% with <1% listed as low (<50). This is a great improvement for me. Hope this helps. Good Luck!


Good advice to all. I’ve been uploading my data and emailing them (I have two trainers) status updates every day. Over the last week we’ve changed one of my ICR settings, and more significantly, tweaked my AIT twice, moving it down from 3hrs to 2. This does seem to have gotten the thing’s attention. The problem I’m having right now is that the sensor keeps undershooting my BG by a non-trivial amount, which means the pump thinks it’s doing a lot better than it is doing, and as a result it’s still not putting out what it needs to get me down to that 120 area during the times when my BG tends to run high already. You’re not supposed to confuse it with too many calibrations, so I’m trying not to totally overdo it, but dammit…


Hi all. Checking in here, sorry for being away but I’ve been kinda busy the last week.

I think the constructive advice on this thread is very useful. Definitely work with your Medtronic trainer/CDE on your setting if you can, and can reach them. Unfortunately for me my Medtronic trainer and CDE where my endo works have been “out to lunch” so to speak, and rather than offer suggestions/advice have offered nothing.

Ok, so much for the ranting on that. I am, after 1 month in Auto Mode, having great success thanks to a lot of advice and guidance from 670G Facebook groups.

The biggest challenge with transitioning to Auto Mode, in my opinion, is that what we understood AIT (active insulin time) and CR (carb ratio) to be before is NOT the same in Auto Mode. Essentially, in Auto Mode, those settings influence the aggressiveness of how the Auto Mode algorithm behaves.

There is only one AIT setting so that’s where I began tweaking things (lowering my AIT from 3 to 2 hours) since it was easiest. Currently my AIT is 2 hours. Making that change improved things quite a bit, but it was still not optimal since I only had one CR (carb ratio) setting and it was 10 and not doing the trick.

So I began (myself) tweaking down my CR, setting up multiple CRs for different time ranges of the day - and things improved tremendously. I made these changes using my Carelink 670G data (especially the assessment report), and continuously monitoring my pump’s 24 hour trend graph to see what patterns were happening at different times of the day.

That all took awhile, and a lot of patience…so work with whatever help you can get from Medtronic, your CDE, and endo to improve things gradually. It will take time, at least a month or so before you get satisfied with your results.

Also - make sure you are inserting your sensor in body locations where you are getting the best most accurate results. My abdomen did not work well for me. I currently use back of my arm (the flubby part) and sides of my calf (lower leg) and am getting fantastically accurate SG results as compared to blood sugar testing. Can’t compare this with the Dex since I never had an opportunity to try it (although I tried before going on the 630G/670G). But I am completely satisfied with SG accuracy with the 670G.