Where to start?

Introduction

In this series of articles I'm going to be investigating and documenting my experience with AI.

At the time of writing, mid-October 2026, public access to AI has been available for less than 4 years.  At this point in time there seems to be three major AI 'applications':

• Open AI - Chat GPT (public release 30 November 2022)
• Microsoft - Copilot  (public release 7 February 2023)
• Anthropic - Claude (public release 2023)

To be precise on November 30 2022 at 11:38am, Sam Altman tweeted:

He made 3 further posts that day, which, just for interest I reproduce here:

Language interfaces are going to be a big deal, I think. Talk to the computer (voice or text) and get what you want, for increasingly complex definitions of "want"!
This is an early demo of what's possible (still a lots of limitations -- it's very much a research release).

Soon you will be able to have helpful assistants that talk to you, answer questions, and give advice. Later you can have something that goes off and does tasks for you. Eventually you can have something that goes off and discovers new knowledge for you.

But this same interface works for all of that. This is something that sci-fi really got right; until we get neural interfaces, language interfaces are probably the next best thing.

(Sam Altman's keyboard doesn't appear to have had a shift key, but I have capitalised where necessary. I've also corrected one spelling mistake, 'got' for 'god' - also corrected in a later post by Sam Altman/)

In addition to Open AI, Microsoft and Anthropic there are two other major AI companies worth mentioning for general use:

• Google Gemini
• Perplexity

Most people have already probably used Gemini without realizing it. When carrying out a Google search there is an AI option and, if selected, this will use Google Gemini to produce results.

And for completeness, or at least partial completeness, there are two others I ought to mention:

• GitHub Copilot
• Cursor

These two are aimed specifically at coding. In GitHub Copilot's case it is an 'add-in' to Visual Studio, and is not to be confused with Microsoft 365 Copilot.

Cursor is an AI-first programming environment that 'understands' an entire codebase and is popular with professional developers.

And how do I know this?

I know because I had a conversation with Microsoft Copilot! The questions I asked were simple but the answers I got back were fairly comprehensive (but not complete). Just to give you a flavour of the conversation I show below the questions I asked and the order in which I asked them.

• When was Copilot launched, and how does that compare with Chat GPT?
• Could you also explain where Claude fits into this picture?
• Are there any other AI 'applications' that I should consider when reviewing AI?

But wait, where is Meta? Well, Meta does have an AI presence, but, it seems to me, to be partially hidden in Meta's products: Facebook, WhatsApp, Instagram and Messenger. In theory this gives Meta the biggest user base but in practice I doubt it means very much at all. However this might be about to change. In September 2026 Meta announced its AI product called muse.at. Currently (mid-October 2026), this is only available in the United States and Canada but looking at YouTube videos it appears to be similar to products like Gemini and Copilot.

Lastly I ought to mention Amazon. Amazon has two (main?) products: Amazon Quick, targeted at the business user and Alexa+ for the personal user.

Amazon Quick has a free tier and, on initial investigation, looks very similar to Copilot. Alexa+, on the other hand, although it's aimed at the personal user, is charged at £19.99 per month or is included free with an Amazon Prime subscription. Considering an Amazon Prime subscription is roughly £15 a month, this looks like very odd pricing to me. Alexa+, of course, integrates with Amazon products such as Amazon Echo and Show tablets and can be voice-controlled.

My experience, so far

Apart from the occasional Google search where I have chosen to use the AI option I have almost exclusively used Copilot, either Microsoft 365 Copilot or, when coding, GitHub Copilot.

To add to the Copilot 'confusion' I've noticed I have two Copilot shortcuts in my task bar. Investigating these it appears I have one Copilot associated with my personal Microsoft account (only used for my outlook.com email address) and the other with my work Microsoft account (associated with my Microsoft 365 Business Standard subscription).  In theory there are differences between the two but in practice I have found nothing really to distinguish them.

General

For 'general' AI use I have tended to gravitate to Microsoft 365 Copilot, the one associated with my 'work' account - in the probably mistaken belief that it was more powerful than the one associated with my 'personal' account. I cannot be specific as to when I started using Copilot but I do know that it would have been before September 2025. About that time I was planning a trip to Japan for the cherry blossom season in 2026. My first questions to AI were around the best time to go so we would be able to make the most of the cherry blossom season. Having established that the best time to go was probably late March to the beginning of April, I then asked Copilot to plan an itinerary. It did this very successfully and we ended up going to Japan in March/April 2026 for a most enjoyable holiday following almost exactly the itinerary suggested by Copilot.

So far my interaction with Copilot has been exclusively in chat mode. In this mode I've asked a question, got the result, looked through the result, and perhaps added or asked subsidiary questions until I got the information I was looking for. There are other options in Copilot that I haven't explored, such as Agents and Notebooks.

Copilot, in chat mode, has basically two models: a 'quick; response and a 'think deeper' response. Generally I have gone with the 'quick' response for non-coding questions and found this perfectly adequate. Over time I think I've detected that Copilot is getting less sycophantic than it was when I first started using it; less of the "Great question Chris," and more just giving me the answer. Some of the answers can be quite verbose but by reading through them one does get the whole picture. In summary I've been more than satisfied with the answers I've been getting from Copilot for general chats.

Coding

My experience of using Copilot for coding has been less successful.

But I need to split this between GitHub Copilot and Microsoft 365 Copilot. The first and obvious difference between the two is that GitHub Copilot is effectively an add-in for Visual Studio and comes with usage limits. Whereas using Microsoft 365 Copilot for coding has no such limits (that I have found, anyway).

Looking at GitHub Copilot first, perhaps I should preface this section by saying that, firstly, I'm a self-taught programmer with not a lot of experience and not a lot of talent either!

Having said that my experience with GitHub Copilot has been largely positive. Its autocomplete of statements seems pretty good. This is called 'code completion'.  Where necessary it seems to be able to ripple through changes where one word or statement has changed which meant that successive statements have also needed to be changed. I have also found that it coped quite well in 'chat mode'. For example I gave it a chunk of code and asked it to re-work the code so that its output would look good on a mobile device. It did this successfully even though it may have taken a couple of iterations. The disadvantage of GitHub Copilot, however, is that one quickly uses up one's free monthly limit. I think I managed to use up the monthly limit in a couple of days of intensive use. I understand that the 'code completion' usage limit is 2,000 code completions per month and the chat mode is limited to 50 chats per month.

Turning to Microsoft 365 Copilot, the first thing to say is that I didn't hit any usage limits. As far as I can see Microsoft does not publish any usage limits although I also believe they are lurking in the background somewhere.

The second thing to say is that as far as coding is concerned I did have some problems. Firstly, some code was just plain wrong, for example, missing the beginning of HTML tags. Secondly, asking for code corrections I frequently had code returned to me that didn't work for one reason or another. In the end I normally ended up with working code but this was at the expense of many backwards-and-forwards questions and answers before I actually had working code.

A general criticism of using AI to create or fix code is that one can quite often end up with code one doesn't truly understand. I fell for this and spent quite a lot of time trying to understand how Copilot generated code worked. I think it would be very easy to end up with software that would be very difficult to maintain in the future.

Where to next?

Perhaps it would be easier to say what I'm not going to do. Whilst it might be tempting to try and review all the AI products available, I think that task would be overwhelming and would also fall into the trap that as the market is changing so quickly at the moment, one would be forever chasing one's tail.

Instead, I'm going to choose one product and explore the capabilities of that one product. As I have already used Microsoft Copilot in one form or another, I'm going to choose a different product. Purely on the toss of a coin, I'm going to select Anthropic's Claude. I aim to look at Claude in its two modes: the conversational or chat mode and its agentic or agent mode. I'm also going to try and put myself in the shoes of someone working for a small business to see how AI might benefit such a business.

 I have no idea where this will take me, but let's just wait and explore...