MAICE Blog

Over the past few months I have been working hard to build an AI coach that will help everyone – particularly those whose jobs are being impacted by AI – deal with the current turmoil. MAICE is the AI coach that guides you every step of the way.

The one AI theme that is constantly in the news lately is that AI will take all our jobs. Most of the influential thinkers in the space disagree with this outcome, however they struggle to explain where all the new jobs will come from.

A 2024 article from Kogan, Papanikolaou, Schmidt and Seegmiller argues that there are two groups in every technology transition: those that benefit from the new technology, and those that are replaced by it. They found that where technology complements the skills of a worker, it generally leads to higher pay and employment. Where technology replaces the worker, the situation is more dire. The transition to AI will be no different. There will be winners and losers. For most of us, however:

“Your job won't be taken by AI. It will be taken by someone who can use AI.”

That is exactly why I built Maice. Maice is the AI coach that helps you make sure that someone who can use AI is you. Maice helps you learn faster, work smarter and keep pace with AI. Maice is a coaching product. Maice works with you – asking the right questions, teaching you proven techniques, and helping you apply new skills directly to your work. It is not a system that automates things for you. Think of it as having an expert sitting beside you, guiding you through every step of the way.

Over the coming months I will share my journey to creating Maice. How I learnt to vibe code like a pro, the mistakes I made and the successes that I had. My recommendations on how you can create your own AI tool, and if you learn from my mistakes, how you will save a lot of time and a great deal of heartache. It was a fun journey and I learnt a lot. I look forward to sharing it here.

Read the Kellogg Insight article this references

Like many of you, I listen to a ton of podcasts about AI. On each and every one of them there is an expert who proclaims that AI will change the way that everyone works. No one will be able to escape the impact.

Unlike the tabloids, not everyone is full of doom and gloom. Many experts also say a version of “Your job won't be taken by AI. It will be taken by someone who can use AI.” This is a common theme from the optimists. I wholeheartedly agree.

Driving home from the gym one morning, I was stuck in traffic and thinking about how most of my friends keep saying that they have no idea how to start with AI. Most people use AI as a search engine. Few really try to learn new skills or use AI to improve the way that they work. I hit upon the idea that I could use AI to teach people how to use AI. I could develop an AI coach – and the idea for Maice was born.

If you use Maice or any other AI system, make sure that you use it to learn new skills, to improve the way that you work, and to stay in touch with how AI is affecting your profession. The world is changing, and quickly. AI, properly used, is the best way to keep pace with AI.

Once I decided to create an AI coaching tool I had to work out how on earth I was going to create such a tool. I had a very clear idea of what I wanted to create. I had invested time thinking deeply about what had worked for me in the past when I had to learn new skills, adapt to new technology and deal with rapid change.

At the time Claude had the best reputation for writing code so I signed up for a Claude Pro account. I sat down with my new friend Claude and told them that I was going to build an AI coaching product that would teach users how to use AI. Claude told me that was a magnificent idea. Claude then produced a training program for how to teach users to use AI. It was quite a well structured training program however it was nothing like what I wanted to create.

I told Claude that I am someone that thinks by talking (I am actually). I said “I will describe exactly what I wanted to create and that I want to use them (Claude) as a sounding board or expert advisor”. Claude obliged however it took some time to convince Claude not to jump to solution but to listen and understand what it was that I wanted to create.

If you ask an AI to build or create something for you, it will. Really quickly. However it will build what it thinks you want. It is unlikely that it will be what you actually want. That is not a fault with the AI, it is a fault with how you prompt the AI. You do not need to take an AI prompting course. Just take the time to think deeply about what you want to create, you are the subject expert.

Talk with your AI partner and exchange ideas and hold off from development for as long as you need. I spent a month of 4 plus hours a day working through exactly what I wanted to develop and the sort of features that I would like to create. Over time Claude became a wonderful development partner. It not only helps you develop a better product, it helps with development once you start vibe coding.

The most important thing that you need prior to getting started on building your vibe coded product is knowing what you want to build. You know your subject area, probably too well. You can see the forest and you know the name of every tree. It is likely, particularly if you are anything like me, that you will try and boil the ocean. You will want to build an everything app that can remove all the pain from the work that you have undertaken over the years.

It can be overwhelming to have so much choice in how you approach this task. When I was thinking of building an app of any type based upon my experience and knowledge it was difficult to decide where I could add value that others couldn't. Where I could find a niche that others had either overlooked or was too small for the major tech companies to care much about?

Focus is the only cure. You must relentlessly focus. Start as small as possible. You can expand your thinking later if it makes sense. Even with a narrow focus area there is a lot more to create than initially meets the eye.

Start with one work flow or idea. Spend time working on fleshing out that workflow or idea. Write down every step. Do as I did and use an AI as a sounding board. Be careful with their suggestions though. Many suggestions will be good but it will take the AI time to learn your true intent with your product. You will understand the idea better than the AI and the AI will generally reply with what it has learnt in training. This will change as the AI gains experience and knowledge via working with you. When the AI truly understands the intent behind what you are trying to create, they will become much more than an expert coder, they will be an invaluable development partner.

Build it and they will come is not a great strategy when the build costs are high and the lead times are long. Consulting companies have made their fortunes advising on market research and knowing who your customer. AI is changing that quickly.

I decided that whatever I built would be a learning experience. It did not matter if it was never released. Like the first book of many successful authors. Created and written but never published and that is ok. A friend who is very successful in business said to me “but where is your moat”. I was going to explain that there are no more moats now that AI has reduced the barriers to entry. However I took a different tack and explained that Maice was a learning experience. I learnt how to vibe code, I learnt how to develop and structure a product that people could use and I learnt how to get it out into the world in this new AI era.

Products are now more rapidly moving into their sweet spot for use and they will remain there for a much shorter period. What is hot and useful today will soon be yesterday's hero. The skill for all of us will be to be able to adapt and change more quickly than ever before. Product development has been democratized. We can all develop our ideas into product.

Take your best idea and turn it into a product today. The first product will likely take you a few months of iteration and testing. Don't worry about it taking off and making you your first billion. Conceptualize it, built it and release it. Work out how to make it successful and then start your next product. Maice has already evolved way beyond my first build. Your future ideas will be better than your best ideas of today.

Naming products is difficult. Names need to be short, memorable, relevant and have URL's available at a reasonable price. An impossible ask? Difficult but not impossible.

We have all seen product names which inspire thoughts like “WTF is that”, “how will I ever remember that” and “how do you spell that word?”. There is a huge art to product naming. I am definitely not an expert however names that help in remembering, understanding and finding the product seem to be the key.

Soon after i had the idea for Maice, in fact about 500 meters after I had the idea (regular readers will remember that I was driving in traffic when the original idea came to me) I decided to name the product. I like acronyms which gives a memorable name with a meaning. Many will know that YAHOO stands for Yet Another Hierarchical Officious Oracle. Lego is short for Leg godt which is Danish for “play well” and WD40 stands for Water Displacement 40th Attempt. I briefly thought of trying to fit an acronym around Coach but quickly gave up. I thought to myself “this is My AI Coach….” I wanted the product name to sound more like a human name so I added Everyday. From that moment MAICE was born. Pronounced May-See I felt it was easy enough to remember (although some have trouble with the pronunciation), it relates to the coaching product that I was developing and I was able to purchase the .ai URL. A 5 letter URL with the .ai extension was perfect.

Take some time to think about your product name. No name will be perfect but the more that people use it, the more natural it will sound. Ask a few people close to you that will tell you the truth (I asked my kids, no greater truth tellers). Play with the name for a while and see how it sits with you. Research possible conflicts and trademarks. You can always change the name later but that can be very expensive and time consuming.

I signed up and paid for my Claude Pro account and then told Claude that I wanted to develop an AI Coaching tool. Within a second Claude had suggested a program that was far more like a training program than a coach. It was actually very well structured and looked like it would interesting and useful to attend but I was not interested in developing a training program.

I politely thanked Claude (if the AI's do take over I want them to at least believe me to be polite) and said that I was trying to create something different to what was suggested. I had a strong idea in my mind but the connections between my brain and my typing fingers can sometimes produce waffle. Meaningful waffle that I completely understand but nonetheless waffle that an LLM will have to guess at meaning.

I told Claude that I am a person that thinks by talking (I do and I usually talk to myself, I know I am not alone here). I said to Claude that I will explain what I want to create and ask for ideas and input where I need it. Once Claude stopped trying to solve my problem but listen to what I wanted we quickly started to put together a more coherent description of what Maice would eventually become.

My advice is to talk to your LLM and make that they understand what you are trying to create or do prior to them rushing to solution. Take your time explaining and discussing. I still had to stop Claude many times along the way though. Claude took quite some time to truly understand what I was trying to create. Claude did adapt over time and eventually understood my thinking. This is when Claude became a great working partner.

I had been working on Maice for about 6 weeks when disaster struck. Claude and I had developed my ideas significantly. We had worked out how to turn them into code. How to structure each piece and the overall project. We had created the first code, possibly 10% of what eventually became Maice. I was in the middle of cutting and pasting something and Claude went blank. Everything disappeared. I was suddenly at the sign up screen. I could not sign in, my account as not being recognized. A terrible feeling washed over me. I was in a public library at the time and and inadvertently said “oh shit” much more loudly than I should have. A couple of people looked up and I said “sorry first world problem”.

I politely thanked Claude (if the AI's do take over I want them to at least believe me to be polite) and said that I was trying to create something different to what was suggested. I had a strong idea in my mind but the connections between my brain and my typing fingers can sometimes produce waffle. Meaningful waffle thMy entire Claude account had been deleted. There was nothing. The system did not recognize me when I tried to sign in. I could not access all the work that I had done over the past 6 weeks. It was all gone. I tried talking to the help chatbot but it kept telling me that I had not paid my account invoice. It refused to believe me when I tried to upload receipts showing that I had paid. I frantically checked my bank account to see if a payment had been rejected and that had caused the problem but everything was how it should have been. Anthropic had been paid but my account had been deleted.

I had no choice but to sign up again with the same email address in the hope that the system recognized my account and reinstated it. The system did not recognize my account. It did not restore anything however my credit card details were prefilled when I signed up for the second time. I was back talking to a new version of Claude however it had no memory of what we had done. No idea of what we were trying to create. I had no choice but to start over again.

The main thing that I learned from this experience is that you should not trust that the system will work. Regularly ask your AI to provide you with a summary of everything that has been done. Claude provides great summary documents that can be downloaded (and saved in multiple safe places just in case). One advantage of asking Claude for a summary document is that it will provide all the information that it needs to pick up from where it left off. More on this later but I very strongly recommend that you get into this habit. Every day if necessary.

After my account was deleted and I signed up again, I was surprised to find that Anthropic kept charging me monthly for the original account and now for the new account.

Despite the double charge I kept working with Claude. Who knew if Anthropic would stop charging me even if it did delete the account. I have tried talking to the Anthropic chat bot (what a waste of time that is, avoid this experience at all costs). I have tried asking for human intervention with the problem (it takes ages and at the time of writing I am still waiting for a response). Unfortunately there has been no resolution to date. I still hold out hope though. One day.

I have no problem paying for a great product but nobody should be charging you twice for the one service. I have no polite suggestion for how to stop Anthropic double charging you. Maybe someone knows somebody at Anthropic and we can finally get some human intervention. It is not a large amount of money and Anthropic appears to be a principled company.

After learning an important lesson with my account being deleted I was surprised to learn another important lesson about messages in chat. They can disappear for no apparent reason. Despite having spent 30 minutes talking to Claude about a concept and having Claude respond and make suggestions about that concept when the messages disappear Claude will have no idea what you are talking about.

There are a number of important conceptual structures built into Maice. I won’t go into them however for Claude the concepts were new and Claude did not understand what I was trying to achieve prior to my explanations. I took meticulous care to explain them. We had great discussions on the concepts themselves and on how to implement them in code. When the messages disappeared I once again had to start again.

This was now the third time that I had explained one particular concept to Claude (the deleted account, the deleted messages and now again). One advantage of having to do this is that it forces you to clarify your thinking each time. Whilst that is actually useful repeated work is terribly painful. I have now taken to copying long detailed explanations of concepts into word documents. If I need to explain something for a fourth time I can now cut and paste (and review before posting).

I have found this is a good discipline with all important pieces of work, thinking and explanation. I have now taken to drafting longer concept explanations in a word document which I then cut and paste into Claude. This allows me to review our discussions when I am off line. It helps to clarify my thinking.

I had been working with Claude for a few months and our conversations had become long and detailed. Once conversations become too long, the system starts to freeze, messages appear to be posted however Claude has not been able to read the message despite it being in chat. Messages take longer and longer to load and the whole thing becomes unworkable.

If you pay close attention, a system warning message appears in yellow at the top of the screen saying the conversation is getting too long. It only lasts for a second or so and you are unlikely to spot it if you are focused on the latest message or typing.

The solution is to start a new conversation. This new conversation will be with a new Claude. This new Claude will have no idea about your conversations with the previous Claude. (Projects has helped fix this issue however it is still not perfect)

One advantage of using my suggestions of regularly asking for summary documents and keeping off line copies of longer explanations is that when you have to start a new conversation you can bring the new Claude up to date very quickly. Just upload the summary document and the important explanations. I do however tend to upload explanations when they are needed during our work. Claude can forget things that were said much earlier in conversations.

Early in my vibe coding journey I cut and paste a large amount of code that Claude had created. I saved and tested the code on localhost and it did not work as intended. Something was wrong with the code. A line that Maice was supposed to print out once a user had input their response, was only partially printing.

My initial assumption was that I had done something wrong (99% of the time that is the case). Claude and I spent most of the next hour going through the code checking this bit and that bit. We tried about a dozen things that might be causing the problem and nothing worked.

I was reading though the code just in case I spotted something. I am not totally foreign to coding, I did learn some basic python a few years ago just so that I could better understand how software worked but I am a long long way from being able to develop code that worked. Low and behold I found the error. Claude had used a single apostrophe in the word “what’s”. This single apostrophe was cutting off what the code was supposed to be printing out.

My suggestion is to have a look at the code that Claude produces. You will gradually become more familiar with structure and learn something about how the software works. You don’t need to write the code but it helps to understand what it is trying to do.

Ok this is a really basic coding error but I found it! Claude spent an hour and could not find the mistake! I was so proud of myself. I can tell you that a few beers were had that night and I retold the story many times. I must admit that the fish got bigger with every retelling.

There have been multiple times, particularly early in my journey with Claude where Claude forgot what I had asked it to do. This is most pronounced in when we are dealing with code. I was very clear with Claude that I am not a coder and that step by step instructions need to be given every time. Under no circumstances should Claude think that I know what I am doing.

The problem with Claude is that after a few successful attempts at cutting and pasting and getting code to work, Claude will slowly forget just how inept I actually am. Claude knows where it wants code to be placed or updated and forgets to provide me with context on how to find the piece of code that needs to be updated. If I carry out instructions correctly a few times Claude makes the big assumption that I now know what to do (I don’t).

The only way that I found around this problem is to continually ask for step by step instructions. Claude never gets upset or loses patience. As humans we are not used to this level of tolerance for error. Claude is happy to comply every time.

It does take a little time to get used to this changed way of interacting with something that appears to have human qualities. Don’t hesitate to ask Claude to explain again, give step by step instructions or to “please fix this mess that I just made of your code”. Claude never complains (almost but that is a story for another day).

As a non coder I need my advisor to give me instructions one at a time. I need to complete one instruction and then start to worry about the next instruction. There have been many times where Claude forgets this request. As soon as I seem to show some competence Claude starts to provide multiple instructions in the one chat message. Instructions about different parts of the code on different pages and in the front end and the back end.

I get lost fairly quickly. The chat messages are long and I need to continually scroll up and down and switch between applications to try and fix what ever it is that needs to be fixed or created. My brain just does not work like a coder. I therefore make predicable errors (so predictable that Claude now asks me to check the areas where I make my most common errors first when code does not work).

The only way that I have found around this problem is to continually ask for step by step instructions, one block of code at a time (regular readers will be noticing a pattern here). Claude will do as asked for a while but be prepared for Claude to forget again. Remember, Claude has infinite patience and you can ask as many times as you need.

There are more sophisticated ways of getting Claude to remember and if asking and asking again and again becomes a problem for you, research those methods. Once I get Maice more fully developed and I have a deeper understanding of how to get the most out of Claude I will be doing exactly that. The best way to become more skilled at using Claude is to use Claude and find solutions or better ways when you need them. You will have context and the instructions will have greater meaning.

I have sat through more training sessions than I can count and walked away forgetting almost everything that was said. In my experience that happens to most of us — and it's not because we weren't paying attention.

The problem is the method, not the person.

Most of us review what we've learned by reading it again. It feels productive. It really isn't. Re-reading creates the illusion that we know something because the words feel familiar. But familiarity isn't the same as understanding.

Active recall flips this around. Instead of reading your notes again, you close them and try to remember. You test yourself before you look anything up. The effort of pulling something from memory is what makes it stick — and the gaps you find are exactly where you need to focus.

After your next training session or coaching conversation, give yourself five minutes before opening anything. Write down everything you can remember. What comes easily is already yours. What doesn't is your real learning agenda.

If you're using Maice, ask your coach to test you on something from your last session before moving forward. Active recall built into the conversation, every time.

I used to think that if I paid attention during a training session I had learned something. I was wrong. A week later most of it was gone. Not because I wasn't trying — because I never went back to it.

That's the problem with how most workplace learning works. We learn something once and move on. But memory doesn't work that way.

Spaced repetition means revisiting what you've learned at increasing intervals — not cramming, but returning to material after a day, then a week, then a month. Each time you return, the memory gets stronger and lasts longer. It works with the brain's natural forgetting curve rather than against it.

It doesn't require much time. A ten minute review a week after learning something, and again a month later, makes an enormous difference to what actually stays with you.

Pick one skill you've learned in the last month. Schedule a short review for next week. Test yourself on it rather than just rereading notes.

If you're using Maice, ask your coach to revisit something from a previous session before you move on to something new. That one habit will change what you actually retain.

I have spent a lot of time in rooms where people nodded along to things they didn't really understand. I have been that person. You follow the logic in the moment, it makes sense while someone is explaining it, and then you try to apply it and realise you only had the surface words.

Surface learning breaks down the moment the situation changes slightly.

Elaboration means connecting new knowledge to what you already know. Self-explanation means asking yourself why something works, not just what it is. Together they turn passive understanding into something you can actually use.

After learning something new, try explaining it in your own words — not to show that you know it, but to find out where you don't. The places where the explanation gets vague or reaches for jargon are exactly the places where the understanding is shallow.

This week, after you learn something new, take five minutes to write a brief explanation as if you were telling a colleague who has never heard of it. The gaps you find are your real learning agenda.

IIf you're using Maice, describe something you've been learning to your coach and ask them to push back with questions. The questions will show you quickly where the understanding is solid and where it isn't.

There is a particular kind of confidence that comes from being familiar with something without truly understanding it. I have felt it many times. You've heard the term, you can use it in conversation, you nod when others mention it. And then someone asks you to explain it simply and you realise you can't.

Richard Feynman, the Nobel Prize-winning physicist, had a simple test for real understanding. Explain it as if to someone with no background in the area. Where you reach for jargon, where the explanation gets complicated, where you go vague — that's exactly where the gap is.

It's a remarkably honest technique. You can't fake your way through it.

Pick a concept you've been learning about recently — an AI tool, a process, a skill. Try to explain it out loud in plain language, as if to someone who has never encountered it. Notice where you hesitate. That hesitation is the learning.

If you're using Maice, ask your coach to ask you to explain something back in simple terms. The stumbles aren't embarrassing — they're the point.

One of the most common reasons people put off learning something is that it feels too big. I did this with building Maice — I looked at the vibe coding skills that I needed and decided it was too complex to tackle right now, as if the whole thing had to be learned at once or not at all.

It doesn't. That's the point of chunking.

Chunking means breaking a complex skill into small, discrete pieces and tackling one at a time. Each piece is learnable on its own. Mastery builds progressively. The overwhelming thing becomes a series of manageable things.

The cognitive science behind this is straightforward — our brains handle focused, bounded tasks far better than open-ended complexity. Give yourself a clear starting point and a clear first step, and the barrier to beginning drops dramatically.

Think of a skill you've been avoiding because it feels too big. Break it into five specific sub-skills. Rank them. Start with just the first one this week — nothing else.

If you're using Maice, tell your coach what you want to learn and ask them to map it into chunks with you. A clear path changes everything.

Most of us practice by doing what we're already comfortable with. I have been guilty of this more times than I'd like to admit. What we prefer, we practice. It feels like progress. However it produces very little progress.

Comfortable practice reinforces what you already know. It doesn't build new capability. The research of Anders Ericsson, who spent decades studying expert performance across fields from chess to surgery to music, found something consistent — what separates good from exceptional isn't the volume of practice, it's the nature of it.

Deliberate practice means identifying the specific thing you find hardest and focusing your effort there — just beyond your comfort zone — with immediate feedback. Not the parts you enjoy. The parts that are difficult.

The cognitive science behind this is straightforward — our brains handle focused, bounded tasks far better than open-ended complexity. Give yourself a clear starting point and a clear first step, and the barrier to beginning drops dramatically.

Think about a skill you're developing right now. What's the part you consistently avoid because it's hard? That's exactly where to focus this week. Seek feedback on it specifically. Resist the pull back to the comfortable parts.

If you're using Maice, ask your coach to focus on your weakest area in a skill you're working on rather than covering new ground. Targeted is almost always better than broad.

I spent a long time in my career thinking that the most helpful thing I could do for someone was give them the answer. It was only after speaking to a General Manager that I admired about this that I realised that answers tell people what to think. Questions teach them how to think.

Socratic questioning is one of the oldest learning techniques in existence for a reason. It works. By guiding someone to reason through a problem themselves — through a series of probing questions rather than direct instruction — you produce understanding that is deeper, more flexible, and far more likely to transfer to new situations.

The difference in experience is significant. Being given an answer feels efficient. Working out an answer through good questions feels harder in the moment and produces something that actually lasts.

Next time you face a problem at work, resist the urge to immediately ask someone for the answer. Ask yourself three questions first: what do I already know about this, what have I tried, and what would I suggest to someone else in this situation.

If you're using Maice, ask your coach to guide you through a problem with questions rather than answers. It's a different experience — and a more powerful one.

I have watched people leave training sessions that I have run, genuinely motivated to change something. Two weeks later, nothing has changed. Not because they weren't sincere — because there was no bridge between the session and the work.

That bridge is immediate application.

Learning without application fades quickly. The transfer from a session to real work is not automatic — it has to be deliberate. One real application of a new skill, within 24 to 48 hours of learning it, is worth more than another hour of instruction. The skill gets tested, refined, and embedded in a way that classroom learning alone never achieves.

Before your next learning session ends, identify one specific thing you will apply to real work before the end of the week. Name the task. Put it in your calendar. The commitment itself changes the likelihood that it happens.

If you're using Maice, your coach will ask you to commit to an immediate application task at the end of each skill session. That commitment is part of the learning — not an afterthought.