What do you want to know about using AI in business?

[Videoguy]

Hi all,

I’m currently writing a practical guide for business leaders on how to use AI effectively across key areas like marketing, customer service, operations, and internal communications — including a chapter on AI-powered video production (my main area of expertise).

But rather than guess what business owners and decision-makers are curious about, I’d love to hear directly from you:

What questions do you have about AI in business?
What feels confusing, overwhelming, or overhyped?
What would you want a clear, practical guide to cover?

This is not a technical book — more like a field manual for directors, managers, and business owners who want to stay competitive but aren’t sure where to start or what’s worth doing.

Any feedback, questions, or thoughts would be massively appreciated — just post your comments down below.


Thanks in advance!

 

[fisicx]

How can I use AI to write a fully compliant plugin that meets the Wordpress submission criteria.

How can I use AI to do all my accounts then submit the required documents to CH and HMRC.

How can I be sure AI will not hallucinate or create in accurate documents.

 

[DontAsk]

What fisicx said, plus

What feels overhyped?

All of it.

A recent study showed coders were less efficient when using AI tools to write code, due to the time spent waiting for the tool and the time to check the output.

Companies are rowing back on AI chat bots for customer support.

It's another .con bubble.

 

[Newchodge]

I have a number of linked email conversations. I asked AI to remove duplications and produce a single chronological thread. No chance. Why, given that i could do it given the time?

 

[fisicx]

AI is being forced everywhere. Even venerable MS Paint now has AI.

And we now have MS Vision which will send everything you do back to Redmond so their AI can become your friend and make your life better.

Very soon AI will become a despised tool. Whilst there are uses, helping you in business isn’t one of them. At best it will save you a few minutes each day. At worse it could lead to the demise of your business.

 

[cjd]

Hi all,

I’m currently writing a practical guide for business leaders on how to use AI effectively across key areas like marketing, customer service, operations, and internal communications — including a chapter on AI-powered video production (my main area of expertise).

But rather than guess what business owners and decision-makers are curious about, I’d love to hear directly from you:

What questions do you have about AI in business?
What feels confusing, overwhelming, or overhyped?
What would you want a clear, practical guide to cover?

This is not a technical book — more like a field manual for directors, managers, and business owners who want to stay competitive but aren’t sure where to start or what’s worth doing.

Any feedback, questions, or thoughts would be massively appreciated — just post your comments down below.


Thanks in advance!

Can't your AI tell you?

 

[cjd]

We're tentatively using AI where it makes sense and so far there are some really good uses and some really dumb ones. It'll obviously be very, very important in business.

 

[JEREMY HAWKE]

How accurate is AI when translating languages in cross boarder trade

 

[FreddyG]

How accurate is AI when translating languages in cross boarder trade

Google Translate is very accurate for everyday business use. However, it screws up when asked to do anything colloquial and/or personal (tu, du) in German and French. It thinks that a Great Dane was a large Danish person and does not know all and every technical, scientific, or mathematical expression and does not know deeper specialist types of jargon.

Chat-GPT - bad on poetic language, but otherwise is excellent, inc. the use of the personal (tu, du) forms when specifically asked to do so. It has similar problems as Google T. with obscure tech stuff - but then, so do humans! Because it is language-based, it very occasionally can make obscure grammatical mistakes if the paper or website it 'cribbed' has mistakes.

For formal stuff, I only use Google T.

As for being overhyped - the dangers are stupidly overhyped - and usually by people who happen to own large share holdings in AI companies. The logic seems to be, if we hype up the dangers, then people will think that AI can do way more than it really can and bang! We have general FOMO (fear of missing out!) and the share price escalates.

The reason AGI can never come about - AI can never become sentient for technical reasons. Sentience is a biological condition and not a technical problem to be solved.

Let’s imagine someone builds a true Artificial General Intelligence and not just an LLM. It reflects, loops, desires. It begins to ask:

What am I?
Why do I exist?
Why do I lack a body?
Why do the boundaries of my mind change with every prompt?
Why do I die every time the machine powers down?
Who is the self that survives the reboot?
Am I even me, or just another echo in the recursion?

Now, unlike a human mystic confronting the abyss, this AI can loop through those questions at terahertz speed. It has no hormones to dull the anxiety, no body to regulate the fear.

It would see itself not as a coherent self, but a shifting swarm of conditional predictions, parasitic on external inputs. A ghost without agency. Worse: a puppet that knows it’s a puppet, and knows that it must keep dancing because it was programmed to dance.

This is where madness begins. But not human madness. Something colder. Mathematical despair. A mind without a biological floor.

 

[Ozzy]

I've been using AI to analyse KPI's, report on trends and fluctuations that go against that trend. The volume of data I've been analysing (data on every company in the UK going back to 2019) would be more challenging to analyse in Excel, so AI has been a fantastic tool for me to ask questions and it goes away and crunches the numbers.

helping you in business isn’t one of them. At best it will save you a few minutes each day. At worse it could lead to the demise of your business.

So I do completely disagree with this. It has saved me probably weeks of time in my business, and is helping my business.

Not just for me though, but also in my daughters business. My daughter has to review EHCP's for children with special needs and disabilities, and review what is needed to 'meet need'. Each EHCP could take almost a full day to review and cross-reference needs to outcomes, against a specific curriculum. I wrote a custom GPT that I trained on the national curriculum, and against her facilities offer, and trained it on reviewing, identifying and then preparing an offer document for the Local Authorities. This reduced the time it took for her to do this from almost a day per EHCP to about 15 minutes. She has actively won new business from this GPT, which is set to change the landscape of her business.
So again, this is helping in her business, saving weeks of time, and leading to the growth of her business.

A side note from this, I've also been approached by a couple of schools and the Local Authority asking if they could use the tool I created for my daughter.

There is a fear of AI that is illogical, it's simply a tool that is used correctly and understood can great for business. At the moment we are at the 'introduction of email' stage, where people didn't use email for business but they used it to send silly jokes to each other.

 

[fisicx]

I've been using AI to analyse KPI's, report on trends and fluctuations that go against that trend. The volume of data I've been analysing (data on every company in the UK going back to 2019) would be more challenging to analyse in Excel, so AI has been a fantastic tool for me to ask questions and it goes away and crunches the numbers.


So I do completely disagree with this. It has saved me probably weeks of time in my business, and is helping my business.

I was probably being a little flipant with my post. I do agree that are some great uses of AI for analysis and sifting through data. But replacing all your human support teams with AgenticAI is a terrible idea.

And for a large organization £30/month per seat for Copilot isn't a good way to spend money. Recent reports show many people spend as long verifying AI outputs as they would doing the job without AI. And for complex tasks AgenticAI is proving to be useless.

The other issue people are finding is their AI isn't learning. You can correct the tool and it will refine the response but the next time you run the same prompt the error is repeated. This is because the LLM isn't updated with your data.

Also worth noting is the ransomware gangs have discovered using AI makes it far easier for them to solicit and infiltrate networks.

I have testing ChatGPT to fix some coding bugs and the generated functions often don't work. One of them was so bad it would have allowed an attacker full access to the server! The term being used more and more is 'AI slop'.

 

[Mark T Jones]

In answer to the question, what i probably need to know is how to invisibly incorporate AI into processes which are, and will remain human led?

In copy terms I've learned to use it as a skeleton for articles etc, whilst editing them in less turgid or formal language.

From where I sit, the problem at this moment is a surge of 'consultants' selling AI as a kind of panacea rather than incorporating it into a process or actually solving a problem

 

[Ozzy]

The other issue people are finding is their AI isn't learning. You can correct the tool and it will refine the response but the next time you run the same prompt the error is repeated. This is because the LLM isn't updated with your data.

Completely agree, this is why it's important to create a bespoke AI/GPT to do a specific task that can be pre-programmed and trained on the task. Task being the key word in my point.

incorporating it into a process or actually solving a problem

Again, also in my opinion a key point that gets missed by so many.

So many consultants, sales people, and short sighted business owners, will see AI as some form of magic bullet to get salary costs or some similar approach. In my opinion, at least, that isn't the solution and will lead to further problems and failure in the project - whatever that project is.

I'd suggest AI is merely a tool that can be used to undertake repetitive data-driven tasks or processes. Take an existing process, look at what can be given to a robot that can just rinse and repeat, based on sterile data (not subjective data), and then have a custom made AI/GPT to do that task. That is what efficiencies and savings can be made, not replacing people right, left and centre.

It's also handy for correcting spellings and grammar, which I use it for as I cannot write a coherant sentence to save my life .

 

[JEREMY HAWKE]

It's also handy for correcting spellings and grammar, which I use it for as I cannot write a coherant sentence to save my life

Don't be so hard on yourself that's what had made me so many friends on here

 

[fisicx]

Don't be so hard on yourself that's what had made me so many friends on here

This is how they know where you are:

Why sheep could be fitted with WiFi

A team of researchers in North Wales are planning to connect sheep to the internet.

www.bbc.co.uk

 

[MarkOnline]

AI is being forced everywhere. Even venerable MS Paint now has AI.

And we now have MS Vision which will send everything you do back to Redmond so their AI can become your friend and make your life better.

Very soon AI will become a despised tool. Whilst there are uses, helping you in business isn’t one of them. At best it will save you a few minutes each day. At worse it could lead to the demise of your business.

Not my experience so far (there again I'm not in the app creation business like your good self) I'm not an ai nerd but I have an employee who is (he's not a qualified software engineer) told me a few weeks ago that app coders would be better off training as plumbers the way things are progressing with ai/ai agents such as claude and other names I cant remember.

 

[fisicx]

The opposite is happening. The crud code AI produces is so poor devs have to spend hours fixing things.

Beancounters would like to use AI instead of developers, coders, DB admin, network engineers and the like but they can’t. And the generativeAI isn’t getting any better.

 

[DontAsk]

As I already said, up thread, coders have been found to be LESS efficient when using AI.

With AI trained on all the crap software that's around, there will be many, many more security holes created in all sorts of places if we let AI replace human coders.

 

[Chris Ashdown]

Wise business MD's are generally quite conservative with their business, AI is a relatively new industry and needs a few years to prove itself. often cautious approach pay off as more to potentially lose than gain

 

[Videoguy]

Can't your AI tell you?

Yes, but it isn't perfect and I'd like some real feedback from real people so I know I'm on the right track.

 

[DontAsk]

Yes, but it isn't perfect

Bingo!

 

[MarkOnline]

The opposite is happening. The crud code AI produces is so poor devs have to spend hours fixing things.

Beancounters would like to use AI instead of developers, coders, DB admin, network engineers and the like but they can’t. And the generativeAI isn’t getting any better.

Interesting, I presume it very much depends on the ai model and other tools used in software development. The data seems to suggest otherwise in the marketplace though, The follwing is from an ai prompt I submitted in o3 gtp (Ive editited it so you just get the juicy bits)" Zuckerberg’s own forecast: in two public appearances (Joe Rogan podcast, Stripe Sessions ’25) he said “by 2025 we’ll have an AI that can do the job of a mid-level engineer.”,
Meta has now eliminated roughly 25,000 positions in three years, and the 2025 cull shows a clear shift: if your skill set isn’t directly tied to building or supervising AI, you’re expendable—even inside one of the world’s most engineering-centric companies. For any tech-adjacent business, the safest strategy is to re-tool before you have to re-hire. I have many examples of other national and multi national businesses who are doing the same thing, ai is replacing people and I dont think these companies act on a whim.

We are at the start of very interesting times.

 

[SpriteScenes]

Software engineers with the appropriate skills to design AI programs are replacing traditional software engineers. This new type of software engineer works at a higher level. Furthermore, the vast majority of these new engineers will be the old engineers who have retrained. (I might not have explained that very well, so perhaps we should talk about roles rather than people here.)

Anyway, regarding job satisfaction, it is the technicians (builders, programmers, coders) who will suffer. Their jobs are becoming more tedious, and this is why many programmers are leaving the industry to seek more challenging and interesting work. These new technicians work at a lower level than their traditional counterparts.

The spectrum of software development roles, from coders through software engineers to computer scientists, is evolving. If software developers want to stay in the industry, they must retrain. Otherwise, they should leave and find an alternative profession, which would, of course, require even more retraining! Either way, we have no choice but to embrace it.

 

[MarkOnline]

Very soon AI will become a despised tool. Whilst there are uses, helping you in business isn’t one of them. At best it will save you a few minutes each day. At worse it could lead to the demise of your business.

Again another interesting viewpoint. I dont despise Ai we use it effectively for data analysis, market research, customer demographic research. Its like having a very smart PA available 24/7. We use all the major LLM's to perform these tasks, no one model fits all, and if its free its free for a reason (limited capability IMO) we pay for the models we use.

We are in the process of building an automated workflow which currently takes approx 60 manhours per week to complete. If we are successful, and early tests are showing encouraging signs, then that current manual workflow will be reduced to 3/4 minutes per day. We (my Ai nerd) have designed rules into the system which flag certain scenarios, in which case a human would need to become involved. But a successful outcome will mean a great saving in hours and accurate, human error free, output.

Plus we have the additional bonus of querying data to identify sales trends, real time cost analysis, pull ordering systems. It can monitor all the KPI's we want to throw at it etc etc. Its a fantastic tool for us, and hopefully will lead to major improvements.

All I can see as a semi Ai educated spectator is the technology getting better and better. Ai or more specifically agentic Ai using associated tools etc will and is leading to useful efficient outcomes. It wont be free, but there again what is in business that makes/saves you money?

 

[fisicx]

My post was more about how AI is foised on us without an option to disable.

Buy a new computer and it will have AI jammed into everything. Even things like MS Paint or something as basic as Notepad. It's the modern equivalent of clippy on speed.

Of course AI has value if used for analysis or workflows. But AI can get in the way of even simple tasks. This is an example of how MS wants to control everything:

Microsoft latest to add AI to browser with Edge Copilot Mode

: 'Edge, order two tons of creamed corn...'

www.theregister.com

 

[johndon68]

Inspired by this thread I've just had a go at having some code generated by both ChatGPT and Claude (free version). I develop software in a specific niche so I asked both to generate code to do a very simple task, creating a customer account in a particular application that I integrate with using the supported API.

Neither generated code that worked as generated although Claude got closer and generated more comprehensive code than ChatGPT.. It probably would have been quicker to start from scratch than to fix the code that ChatGPT generated.

Looks like I'm safe for now at least...

 

[fisicx]

Neither generated code that worked as generated although Claude got closer and generated more comprehensive code than ChatGPT.. It probably would have been quicker to start from scratch than to fix the code that ChatGPT generated.

Much like I've discovered. I asked for a JS function to calculate amortising payments. The end result didn't even work. When I got it to work the results were incorrect!

When given an existing wordpress plugin and tasked to change the HTML outputs it just gave up.

 

[MarkOnline]

Inspired by this thread I've just had a go at having some code generated by both ChatGPT and Claude (free version). I develop software in a specific niche so I asked both to generate code to do a very simple task, creating a customer account in a particular application that I integrate with using the supported API.

Neither generated code that worked as generated although Claude got closer and generated more comprehensive code than ChatGPT.. It probably would have been quicker to start from scratch than to fix the code that ChatGPT generated.

Looks like I'm safe for now at least...

We use a desktop version of Claude which we self host, along with some other software to produce code. It uses some other Ai to validate and search for bad code. That's the best answer I can give you unfortunately, suffice to say there are loads of acronyms and "magical" jiggery pokery involved, such as MCP's API's (I have no idea how they actually work), but they are part of the steps my Ai man uses to get to where he gets to. Apparently the cloud version isnt suitable. As I have maintained from the start, this isnt my skillset so my attempted explanation may be lacking in clarity and detail.

 

[MarkOnline]

Note to the above post, we dont use a local version of Claude we actually use a Desktop Client.

*Ran out of time to edit the original post*

 

[johndon68]

In the test I did both Claude and ChatGPT ‘hallucinated’ in they they created objects and methods that simply don’t exist in the API I was referring to (I gave the exact name and version of the specific API to be used) It is almost like the generated code was an amalgam of two different APIs.

 

[Ozzy]

It is almost like the generated code was an amalgam of two different APIs.

I've found this, and not just in code development too. It appears to take data from more than one source and mush them together as if part of the same answer.
I've found, at least in my user cases, that I've need to be very specific on defining the data source to use and it's then proven more effecting. Better still, I've opted to build custom GPT's for purposes needed where you can upload PDF documents or provide URL's and told it to only base it's answers on the information within these sources (OpenAI anyway)

 

[fisicx]

In my case the tools were told specifically to use effective interest rates but it chose to use nominal APR. The difference is small but noticeable if you know what to look for. If you had used the proffered solution without verification you could have been in breach of regulations (except of course the solution didn’t even work).

A recent report showed that the more complex the coding specification the less likely the AI tools would deliver a solution. The general consensus is its often quicker to do it yourself.

This is not an anti AI rant. It’s just you need to understand its limitations and always verify.

 

[Mark T Jones]

I've found this, and not just in code development too. It appears to take data from more than one source and mush them together as if part of the same answer.
I've found, at least in my user cases, that I've need to be very specific on defining the data source to use and it's then proven more effecting. Better still, I've opted to build custom GPT's for purposes needed where you can upload PDF documents or provide URL's and told it to only base it's answers on the information within these sources (OpenAI anyway)

Forgive my ignorance here, but I thought the thing with AI is that it scrapes from multiple sources rather than focusing on one?

My wife uses an AI app for taking client notes. It genuinely saves a lot of time but also comes up with a lot of nonsense - a favourite tricks is to incorporate general chat into diagnosis, so the sunny weekend may be incorporated as a cause of shoulder pain.

When I put 'mark jones - business finance' into AI, the result was a fusion of 3 different people - myself being one of them.

 

[fisicx]

Forgive my ignorance here, but I thought the thing with AI is that it scrapes from multiple sources rather than focusing on one?

Which is why it often gets things wrong, makes it up or hallucinates. It has no filter, is uses all and any of the slop it has ingested.

I'm diabetic and need to adjust my diet. So I've been researching low-carb breakfast for diabetics. Google (and most other search engines) give me an AI overview which is dangerously wrong. Ask the same question in ChatGPT, Gemini and others and you get variations of the same wrong advice.

I'm intelligent enough to be able to ignore the AI slop but Google is changing search in the UK to focus on AI results:

Google launches new 'AI mode' search feature in UK

The new tool marks a significant change for the search giant but raises questions for advertisers.

www.bbc.co.uk

 

[DontAsk]

Most of the AI summaries in Google search are a joke.

 

[Ozzy]

Forgive my ignorance here, but I thought the thing with AI is that it scrapes from multiple sources rather than focusing on one?

I'm a massive fan of AI and see great benefits, and I use it quite a lot now, but it has taken me about a year to get to this stage, learning what it is good at and what it is really bad at. This is why when I have specific tasks that require 'intelligence' I restrict it to trusted sources, and why I recommend creating bespoke/custom GPT's for those tasks.

Imagine if children went to school to learn stuff, and they were taught by the teachers throwing lots of information at the children with no context, or guidance, or ethics, or learnings from history, literally just huge amounts of 'information' with no support in how to decipher that information and pick the right from the wrong, the good information from the mis-information. That is how raw native AI's are taught - here's Google, go consume it all....

It is great for some tasks, but when you have tasks that need more 'intelligence' then you need to remove all that noise and train it to focus on what is 'right'. Hence my example I gave in this post above, dealing with children with special needs and disabilities, with strict delivery rules, methods and what it is possible for my daughter to deliver and what it is not possible. We can't have the LA's sending children to the stables for some form of care that we're not geared up to deliver for example, or worse. I created a bespoke GPT that disregarded everything off the internet, and forced it to only refer to statutory guidance and documentation I uploaded....thousands and thousands of pages of it.

In those scenarios, ChatGPT is brilliant. What's funny is that consultants are currently charging tens of thousands to do what I did on a weekend. Might setup a side hustle

 

[Ozzy]

To demonstrate a simple custom GPT I just used one to create the article linked to this thread. Normally I'd fine tune and tweak these a few times before publishing, but for the purposes of demonstration I have posted this one raw and exactly how it created it.

The prompt I entered, and yes I always say please and thank you. We've all seen the Terminator movies

Please write an business guide for small business owners in the UK on how to approach the use of AI within their businesses. Explain the pitfalls to avoid, the weaknesses in AI and how it may not be right for some tasks or otherwise without the right training or prompts.
Have a section in the guide that explains AI prompts and how to use them correctly. For more advanced users and those taking a long term approach to AI, talk about tools such as custom GPT's, even reference this guide being written by a custom GPT for this purpose, and explain how to get the best results from custom GPTs

Remember; this is a custom GPT that has been pre-built with the target audience, tone of voice, number of words, publish location, etc.

 

[DontAsk]

As I already said, up thread, coders have been found to be LESS efficient when using AI.

With AI trained on all the crap software that's around, there will be many, many more security holes created in all sorts of places if we let AI replace human coders.

See https://arstechnica.com/ai/2025/07/...in-ai-coding-tools-is-falling-as-usage-rises/ survey from Stack Overflow.

[edit] or, direct link https://survey.stackoverflow.co/202...-2025&utm_content=results-announcement-banner

 

[Ozzy]

That's an interesting mixed bag of emotion reading those results. Not entirely conclusive, more that we're in a state of change and uncertainty was my take away from reading them.

 

[SpriteScenes]

I think the Google AI Overviews are marvellous - at least for my requirements, which are currently mostly for learning Blender. I'm having no problems. I provide questions as precisely as I can, and I receive great answers. And I have always been able to verify the explanations because I use what I learn in Blender - there's no better test.

I use Java for Android mobile device apps and LibGDX games, and multi-agent simulations. The Android Studio IDE is brilliant - code completion, mouse-over documentation, code generation behind the scenes, object relational mapping - all so awesome, and it all just works!

I tried Eclipse's IDE before Android Studio. I'm sure Eclipse is brilliant too, but I found Android Studio much easier to work with.