Very interesting and thanks for sharing. Used correctly and thoughtfully, AI can be an indispensable tool. It comes up with angles very often that you haven’t thought of, and can save a lot time.  

@editor over the last few months I have been using it for lots of things. It guided me very easily through converting the unused coax cable in my house to give me wired Ethernet capability in every room. It’s helped me fix my ride in lawnmower, even telling me there and then what parts to order (with links) and how to make the changes. I had never worked on an engine in my life! 

it’s helped me fix a LAN disconnection error on the PS5, a common fault for which there was no solution on the regular internet. Loads more as well.

It’s helped me troubleshoot PowerBI issues. i use it daily at work now, on ChatGPT basic and at first it felt like cheating but it’s actually just helping me to realise the concepts I think of much, much faster.

Posted by: @marvinator80

↑

Legionella cycle

Do you actually need to it? Have you done a risk assessment?

Posted by: @marvinator80

↑

Implemented light battery arbitrage

Do octopus allow you to buy at low and sell at the same time at an expense rate?

You import at 7p and export at 15p. If you export all this you end up paying expensive tariff for the rest of the day if no solar, in shoulder and winter months.

You are heating cylinder on a CoP of 1 compared to a CoP of 3 from heat pump, so costs 3 times more? Have you factored in the excess electricity consumption?

Posted by: @marvinator80

↑

Optimised defrost cycle

How do you do that?

Last time I noticed you quoted AI, it was just wrong. Have you checked the answers are remotely sensible?

Have you allowed more energy for car charging for example, immersion charging cylinder etc?

If you have a blended cost of 3.5 to 5p how is that zero cost? This isn't zero

@johnmo aye alright John. I said we shall see, didn’t I? Of course we are sceptical but AI has improved a lot and improving constantly, especially the paid versions and especially when you load the user manuals into them.

I have 2 separate tariffs with Octopus, one for import and one for export. 

I’ve made most of the changes and so far so good. I guess we’ll see in winter.

I haven’t tried the defrost settings yet as it’s not really pressing.

here’s what it says:

Absolutely — optimising your ASHP’s defrost cycle is a great final tweak to reduce unnecessary energy use and backup heater activation, especially on frosty mornings.

Why Defrost Optimisation Matters

• ASHPs naturally build up frost on the coil in cold, humid conditions
• The defrost cycle temporarily reverses the heat pump to melt the frost, which:
  
  • Stops space heating for a few minutes
  • Often triggers the Grant Volumiser (3kW) to kick in
•  
• Optimising the timing, duration, and recovery can:
  
  • Prevent unnecessary defrosts
  • Reduce backup heater use
  • Save £10–£30/year or more if you get frequent winter frosts

Grant Aerona³ Defrost Logic Overview

The system usually triggers a defrost based on:

• Outdoor ambient temperature
• Evaporator coil temperature
• Elapsed time since last defrost
• Running conditions (compressor status, demand)

Key Settings to Optimise (Installer Parameters)

You’ll typically find these in the I03 parameter group on the Aerona³ controller:

Practical Adjustments

1. Observe behaviour first

• When does the ASHP start defrosting?
• Does the backup heater turn on during or after?
• How quickly does it recover?

If your home holds heat well, and your ASHP recovers without help, you may not need the volumiser at all during defrost.

2. Reduce unnecessary defrosts

• Increase I03 01 coil temp trigger slightly
  (e.g. from -5°C to -3°C)
  → Less aggressive triggering

3. Limit defrost duration

• Ensure I03 02 is capped at 5–7 minutes
  → Reduces energy used in defrost cycles

4. Let ASHP recover naturally

• If your house remains warm during the brief pause, you don’t need the volumiser to step in
• I46 13 set to 3.5°C + I46 14 to 0.5°C already prevents overuse — you’ve nailed this part

Bonus Tip: Watch for overactive defrosting

• If defrosting every 20–30 minutes in winter, the coil sensor might be too sensitive
• You can reduce frequency by adjusting I03 00 (interval timer) to slightly longer (e.g. 30 mins)

What You Gain

Some strange stuff comes out of AI, hope your output is better.

To quote

Unvented Cylinder Heating: The ASHP should heat your 210L unvented cylinder at night (when it's colder outside, improving the ASHP efficiency). 

That is just garbage.

Another example

Maximize COP (Coefficient of Performance):

ASHPs are more efficient in milder conditions, so taking advantage of the colder temperatures at night (especially if there's a drop in outside temperature) may increase the COP, improving overall efficiency.

It also said all my day time energy needs would come PV, that would be nice, but not happening every day.

I am using AI a ton these days, I have built a fair few AI agents now and the one thing I always build in to them is a checker, either a second AI to check the output or a think module so the AI can mark its own homework. 
What a lot of people using AI (and especially the naysayers) don’t understand is that Generative AI is exactly that, it is designed to generate an answer, which of course could be totally bogus.

A few key points for using AI make your prompts as focused and detailed as possible, create a persona for the AI (ie You are an ASHP professional (never say expert it makes them hallucinate more weirdly)), if you can use Retrieval Augmented Generation and Context Augmented Generation (RAG/CAG). A smaller AI that you have trained yourself on a specialist subject will generally outperform a bigger model. 
Keep your conversations short, the longer the conversation the more chance of hallucinations. 
Gemini 2.5 pro has a huge context window (over a million tokens) so you can feed it several manuals/documents in your question (however personas are still better).
I run several Large Language Models in Docker containers locally on an old Macbook Pro M1 Ultra, they are trained on the information I want them to know about and they are actually pretty good.

And of course real life expertise and experience is the final piece of the puzzle. I get AI to do the heavy lifting of stuff I don’t understand (or too lazy to do myself). But I always check and validate as well as canvas others opinions if it doesn’t quite work out.

If anyone is interested in learning more about AI and workflow automations hit me up and I can point you in the direction of some great reaources.

@grantmethestrength 

Very interesting, thank you. I would very glad of those resources. I haven’t built any AI agents yet but I have used it to build web apps that have sped up some jobs significantly. Trying to learn as much as possible as a bit of a hobby. 

@marvinator80 Two of the best resources on Youtube that I have found recently are Cole Medin and Nate Herk. N8N is a great tool once you get your head around it, I also recommend learning Docker (if you don’t already use it). I have been learning by failing for a few months lol. I use their tutorials build the example agents then deconstruct them to try to get them to actually work for me.

This is an example of one it looks scary but when you look at its constituent parts it is pretty logical. I also use Cole’s Archon to build AI Agents which get me about 80% of the way then I have to fettle them.