AI does have its uses

fisicx · Feb 19, 2026

I've been using various AI tools for a while now - mainly to get code and the do odd bit of CSS when I've felt lazy. I've also used it to reword my somewhat terse prose into something more friendly.

But last weekend I wondered if AI tools could create me a plugin to manage risk assessments for a charity. And it did.

Writing the prompt took a while as there were a number of requirements and conditions needed incorporating but once I had something reasonable I plugged into Claude and set the machine running.

This was the result:

Risk Assessment Manager | Testblog Testblog Testblog

[risk_assessment_form]

testblog.co.uk

It's been tweaked a bit and I've changed the CSS but it works and has made life for the charity a lot simpler.

The code isn't brilliant and there are some security issues that need fixing but it's not a bad effort.

Frank the Insurance guy · Feb 19, 2026

Nice! That's the thing about AI - don't expect a perfect finished solution. For me it is never perfect, but can provide a great base to push on from. One of the key things is having a decent prompt - ask it for slop and you will get slop! The more specific the prompt the better result.

antropy · Feb 19, 2026

Frank the Insurance guy said:
Nice! That's the thing about AI - don't expect a perfect finished solution.

Yeah, I've often had to keep asking it to try again, and eventually you can get to a good enough result. It's not yet at the point where non-developers can use it for anything important in my opinion. That day is coming though of course.

Paul.

YasmeenLondon · Feb 19, 2026

Frank the Insurance guy said:
One of the key things is having a decent prompt - ask it for slop and you will get slop! The more specific the prompt the better result.

Absolutely this. 99% of people complaining about AI capabilities will stop complaining if they understand prompting and improve their prompts.

NickGrogan · Feb 19, 2026

If you ask it to fix security issues, it will do that too.
And if you screenshot it, it will improve the CSS as well.

fisicx · Feb 19, 2026

I stripped out a lot of the css. Made it a lot simpler. And fixed a lot of the spaghetti code - load speed dropped by half.

The security was a minor bug to fix XSS that Claude couldn’t get right.

fisicx · Feb 19, 2026

antropy said:
Yeah, I've often had to keep asking it to try again, and eventually you can get to a good enough result.

It comes back to the prompt. For this project the prompt was around 300 words. Didn’t need any update.

Helps that I was a technical writer in a previous life.

NickGrogan · Feb 19, 2026

Drop the code into Gemini, tell it that Claude made it, but you think Gemini can do a much better job.

14Steve14 · Mar 3, 2026

antropy said:
Yeah, I've often had to keep asking it to try again, and eventually you can get to a good enough result. It's not yet at the point where non-developers can use it for anything important in my opinion. That day is coming though of course.

Paul.

We have had ChatGPT going round in circles trying to update a small website module and it was getting it wrong every time. Even trying to tell it it is getting it wrong, just makes it worse. In the end we copied the initial question into Gemini and it changed the module first time. We then went back in to chatgpt and posted what we had, which we knew worked, and it tried to change it, which again didn't work.

I suppose with most things, it will get better in time.

Newchodge · Mar 3, 2026

14Steve14 said:
We have had ChatGPT going round in circles trying to update a small website module and it was getting it wrong every time. Even trying to tell it it is getting it wrong, just makes it worse. In the end we copied the initial question into Gemini and it changed the module first time. We then went back in to chatgpt and posted what we had, which we knew worked, and it tried to change it, which again didn't work.

I suppose with most things, it will get better in time.

Wishful thinking!

fisicx · Mar 3, 2026

14Steve14 said:
We have had ChatGPT going round in circles trying to update a small website module and it was getting it wrong every time. Even trying to tell it it is getting it wrong, just makes it worse. In the end we copied the initial question into Gemini and it changed the module first time. We then went back in to chatgpt and posted what we had, which we knew worked, and it tried to change it, which again didn't work.

I suppose with most things, it will get better in time.

Try using Claude.

Are you revising the prompt each time or starting over. Sometimes revising the prompt is worse than starting over each time.

UKSBD · Mar 3, 2026

I use an AI content generator plugin to create little blog posts.

I don't trust what it automatically creates but edit it, add images, links to pages it is pushing then repost.

I dare say there are better (paid for) plugins that can automatically automate everything, but I prefer doing it the way I do.

It creates posts like this - https://swrds.co.uk/aerator-hire/
2 or 3 minutes of editing I repost them like this https://www.locallandscaper.co.uk/hire-a-lawn-aerator/

johndon68 · Mar 3, 2026

fisicx said:
Try using Claude.

Are you revising the prompt each time or starting over. Sometimes revising the prompt is worse than starting over each time.

For generating code, PHP in my case, I’ve found Claude infinitely better than ChatGPT.

Frank the Insurance guy · Mar 4, 2026

14Steve14 said:
We have had ChatGPT going round in circles trying to update a small website module and it was getting it wrong every time. Even trying to tell it it is getting it wrong, just makes it worse. In the end we copied the initial question into Gemini and it changed the module first time. We then went back in to chatgpt and posted what we had, which we knew worked, and it tried to change it, which again didn't work.

I suppose with most things, it will get better in time.

For general (personal) use I stopped using ChatGPT weeks ago. I find Gemini infinitely better, especially if you want something that is up to date!

Nathanto · Mar 4, 2026

johndon68 said:
For generating code, PHP in my case, I’ve found Claude infinitely better than ChatGPT.

Agreed, Claude is head and shoulders better than ChatGPT for coding (I'm old-school Perl myself.)

The biggest problem I find with ChatGPT is amnesAI (AI amnesia.) You'll be working through a number of problems, it'll present you the latest patch but it's quietly forgotten a bug-fix or addition from half an hour ago. So if you're not very careful you end up eventually fixing one problem only to find that along the way something else is inexplicably missing or broken. You can tell it to only patch the latest version until you are blue in the face and it will still use an older version anyway...

With ChatGPT I ended up having to check literally every single patch hadn't missed out fixes from the previous patch. To this end I highly recommend Beyond Compare, an absolutely indispensable tool for comparing differences between two different files; I use it primarily for code but it also does folders, text and images - just a brilliant piece of software.

MarkOnline · Mar 5, 2026

fisicx said:
I've been using various AI tools for a while now - mainly to get code and the do odd bit of CSS when I've felt lazy. I've also used it to reword my somewhat terse prose into something more friendly.

But last weekend I wondered if AI tools could create me a plugin to manage risk assessments for a charity. And it did.

Writing the prompt took a while as there were a number of requirements and conditions needed incorporating but once I had something reasonable I plugged into Claude and set the machine running.

This was the result:

Risk Assessment Manager | Testblog Testblog Testblog

[risk_assessment_form]

testblog.co.uk

It's been tweaked a bit and I've changed the CSS but it works and has made life for the charity a lot simpler.

The code isn't brilliant and there are some security issues that need fixing but it's not a bad effort.

Thats a big jump from your "Ai is a waste of time" type posts from not too long ago. Train it and prompt it correctly and all sorts of things become possible. It will only get better IMO.

fisicx · Mar 5, 2026

MarkOnline said:
Thats a big jump from your "Ai is a waste of time" type posts from not too long ago. Train it and prompt it correctly and all sorts of things become possible. It will only get better IMO.

Much of AI usage still is a waste of time.

Many organizations have discovered replacing people with agentic AI has come back to bite them.

And the huge resources thrown at AI its unlikely to ever recoup the costs. Then there is are environmental costs - a recent report suggests data centre additional power demand will exceed the total energy production in the UK. And the water consumption will be greater than the replenishment rates.

Yes AI will get better. But the cost in employment and resources may make it unsustainable.

Ozzy · Mar 5, 2026

Frank the Insurance guy said:
For general (personal) use I stopped using ChatGPT weeks ago. I find Gemini infinitely better, especially if you want something that is up to date!

Interestingly, I found ChatGPT better than Gemini but for policitical reasons not worth getting into I would like to find another solution instead of OpenAI. My colleagues are recommending Claude, but that is more for programming than content and brainstorming.

Frank the Insurance guy · Mar 5, 2026

Ozzy said:
Interestingly, I found ChatGPT better than Gemini but for policitical reasons not worth getting into I would like to find another solution instead of OpenAI. My colleagues are recommending Claude, but that is more for programming than content and brainstorming.

Maybe the difference between Paid and Free version - I went from using the Free ChatGPT to the free Gemini and found Gemini superior. Perhaps ChatGPT's paid version is better and can provide more "current" and up to date answers than the free version.

Ozzy · Mar 5, 2026

Following both my CTO's advice and a conversation with @Data Swami I'm looking into Claude at the moment, and am using a pilot project I'm working on as a test case for it.

Paul Kelly ICHYB · Mar 5, 2026

I have just spun up Openclaw and am looking at building my own LLM/system!

fisicx · Mar 5, 2026

Paul Kelly ICHYB said:
I have just spun up Openclaw and am looking at building my own LLM/system!

Excellent! Gives hackers lots of opportunities to kill off your business. It’s open season for anyone with an internet facing network.

Ozzy · Mar 5, 2026

Paul Kelly ICHYB said:
I have just spun up Openclaw

@Data Swami is all over Openclaw at the moment. Looks like it does a lot, but I'm already just impressed with Openclaw Light (Claude)

Paul Kelly ICHYB · Mar 5, 2026

fisicx said:
Excellent! Gives hackers lots of opportunities to kill off your business.

How?

It is on it's own hardware and incoming traffic is not routed to it

Data Swami · Mar 5, 2026

Ozzy said:
@Data Swami is all over Openclaw at the moment. Looks like it does a lot, but I'm already just impressed with Openclaw Light (Claude)

Ye...... Its taken me down a garden path but so much to dev with it now. Own private discord setup with it for agents with specific roles and delegating to execution agents. It is soooo powerful but defo need to make sure what you put on is locked down like using env files etc etc.

A few tips in terms of using Local LLMs use nomic emded v2 for your embedding model and use memory skills and a skill called QMD to help with knowledge bases too to reduce token usage.

Also Playwright for Browser automation as the inbuilt version is a bit flakey and times out. With playwright ive automated so much of the PR stuff i have with all approvals going to the "CMO" in discord

Data Swami · Mar 5, 2026

Paul Kelly ICHYB said:
How?

It is on it's own hardware and incoming traffic is not routed to it

Depends on how youve set it up and also making sure you know exactly what skills are installed form clawhub as some of the top skills installed were malware. And alot of people have set it up with root access so can install whatever. Its just about being sensible with it

Data Swami · Mar 5, 2026

fisicx said:
I've been using various AI tools for a while now - mainly to get code and the do odd bit of CSS when I've felt lazy. I've also used it to reword my somewhat terse prose into something more friendly.

But last weekend I wondered if AI tools could create me a plugin to manage risk assessments for a charity. And it did.

Writing the prompt took a while as there were a number of requirements and conditions needed incorporating but once I had something reasonable I plugged into Claude and set the machine running.

This was the result:

Risk Assessment Manager | Testblog Testblog Testblog

[risk_assessment_form]

testblog.co.uk

It's been tweaked a bit and I've changed the CSS but it works and has made life for the charity a lot simpler.

The code isn't brilliant and there are some security issues that need fixing but it's not a bad effort.

I have been saying to you that its useful XD and that the prompt is key. Claude is defo better for these types of things over the other chat interface enabled ones or maybe Manus too but cost wise claude is better suited

fisicx · Mar 5, 2026

For a specific use I agree. But most people I know use it for fun. Making pictures and animations for example. They aren’t ever going to pay.

Many businesses are struggling to see any real increase in productivity and are now questioning the monthly cost.

As an aside, I wanted an update to an existing plugin and tried Claude. Took over an hour to get a prompt that worked but because the plugin uses a custom API Claude just gave up.

Data Swami · Mar 5, 2026

fisicx said:
For a specific use I agree. But most people I know use it for fun. Making pictures and animations for example. They aren’t ever going to pay.

Many businesses are struggling to see any real increase in productivity and are now questioning the monthly cost.

As an aside, I wanted an update to an existing plugin and tried Claude. Took over an hour to get a prompt that worked but because the plugin uses a custom API Claude just gave up.

Well the issue with productivity is because they just add something like Claude or Open AI and its just a new part of their process not actively looking at the true way of implementing AI and Automation. What have they actually done with AI is the real question and what charlatan have they bought into with their pack of prompts.

There are plenty of opportunities to completely get rid of the mundane repetitive stuff to hand off to automation and blending in AI for some of those things.

And in relation to Claude and the custom API did it have any documentation on the custom API to include within its knowledge base. But yes the prompt is king and you have to structure it well if you are just wanting to update something you already built. Or there are other ways in which you set it up through a proper code review and update process in a custom gpt or whatever claude call it so that it has the guardrails and understanding of what it should be doing with a specific usecase like that

fisicx · Mar 5, 2026

I just did the fix myself. Took less time that it did to craft the prompt.

I’m finding this a lot. With all the highly specialised work I do there is nothing the AI tools can do as the LLM doesn’t know about the edge cases. It struggled for example with calculating Norwegian energy bills for all the gazillions of possible options.

Data Swami · Mar 5, 2026

Well ye edge cases need the extra insight and extra knowledge so you need to download that knowledge into it and go through the thinking. Like everything if you havent got the extra insight you cant give a good output. With the prompts its needs those instructions to search and really research the answers. Yours is a very specific use case that needs more knowledge but in relation to many businesses not getting productivity out of AI uses it should be understanding what they are even trying to do rather than just blanket labelling the AI at fault for their poor planning/implementation. The majority of cases i see where "AI" has been brought in its just a OpenAI or Claude account to pass the work they are already doing through nothing like actually embedding it where its most impactful so they have more time elsewhere. But then thats what most did during the days of digital transformations too

fisicx · Mar 5, 2026

Or….

A company reviews their processes and simplifies and automates without using any AI.

The problem as you have alluded to is businesses have added AI without actually knowing what they want to use it for.

Data Swami · Mar 5, 2026

fisicx said:
Or….

A company reviews their processes and simplifies and automates without using any AI.

The problem as you have alluded to is businesses have added AI without actually knowing what they want to use it for.

Why does it have to be either or. You can combine AI with the automation to get even more.

And yes thats the point no point blaming AI for lack of productivity when they havent actually properly focus on what to do and if they dont have the expertise find that expertise rather than leaving it to a junior like so many of them have done with trying to find out what to do with their data

Paul Kelly ICHYB · Mar 5, 2026

fisicx said:
added AI without actually knowing what they want to use it for.

Data Swami said:
no point blaming AI for lack of productivity when they havent actually properly focus on what to do

And there lies the crux of the issue.

My businesses are so diverse, with many individual 'transactions' that I have no specific thing that would change my world, so, I am going to start simple and create a todo list with regular nudging. I will then probably look at simple blog creation (with approval) and posting and maybe some video creation.

I will not be looking at chatbots, receptionists, managing my accounts etc.

Data Swami · Mar 5, 2026

Paul Kelly ICHYB said:
And there lies the crux of the issue.

My businesses are so diverse, with many individual 'transactions' that I have no specific thing that would change my world, so, I am going to start simple and create a todo list with regular nudging. I will then probably look at simple blog creation (with approval) and posting and maybe some video creation.

I will not be looking at chatbots, receptionists, managing my accounts etc.

The way I work it is just going through the business processes making sure we know how long things take etc to then identify what areas should just be automated, what should be automated and include AI and what just isnt worth it.

Helps to show us whats a quick thing to do that saves plenty of time what would be more of a longer time to do but delivers alot of impact.

You want to make sure you can easily quantify how whatever you are doing is going to deliver both in the short term and long term. So sure it saves me X time now but if we increase whatever we put through it can it cope and if so it means we are saving even more time as we dont need someone on that so they can focus somewhere else

fisicx · Mar 5, 2026

I don’t really have anything that can be automated. I answer email and WhatsApp queries. I do quotes. I code and invoice. Every project is different and often needs lots of questions of the client before we get close to a quote. It’s unlikely AI could cope with the diversity.

Today I had an email from an old client who wants an update to a plugin. AI wouldn’t even know where to begin as the client just referred to a project name that only exists in my private git.

And I enjoy coding so don’t really want AI to do that for me. TBH it’s often the quickest part of the whole project. Most of the time is finding what the client actually wants from an opening request for a casino plugin.

Data Swami · Mar 5, 2026

fisicx said:
I don’t really have anything that can be automated. I answer email and WhatsApp queries. I do quotes. I code and invoice. Every project is different and often needs lots of questions of the client before we get close to a quote. It’s unlikely AI could cope with the diversity.

Today I had an email from an old client who wants an update to a plugin. AI wouldn’t even know where to begin as the client just referred to a project name that only exists in my private git.

And I enjoy coding so don’t really want AI to do that for me. TBH it’s often the quickest part of the whole project. Most of the time is finding what the client actually wants from an opening request for a casino plugin.

Ye each to their own. For my work my "agent" would be able to understand what the project was as its connected to my git and be able to pull all that context from knowledge bases and memory and be able to formalise the update but thats the way I have built the agents to ensure its got all the information and then utilise the different models more suited to code or content etc. So its possible and needs to be plumbed in. But yes the key it to find out what they actually want. Most of the people I have had come to me say they want one thing but in reality thats their perception of what their people are doing rather than what its like on the ground

fisicx · Mar 5, 2026

I deal with a lot of financial organisations with a lot of super secret stuff. I don’t think they would be happy knowing I let an AI agent access their data. Trying to limit what it can access is more trouble than it’s worth. And as many have discovered, AI agents often ignore the rules.

Using AI is always going to be something each person or business needs to carefully assess. Like anything, there is no single solution.

Data Swami · Mar 6, 2026

fisicx said:
I deal with a lot of financial organisations with a lot of super secret stuff. I don’t think they would be happy knowing I let an AI agent access their data. Trying to limit what it can access is more trouble than it’s worth. And as many have discovered, AI agents often ignore the rules.

Using AI is always going to be something each person or business needs to carefully assess. Like anything, there is no single solution.

I work with professional services and financial organisations too the AI Agents use either local or locked down LLMs and ive built a governance framework around it to make sure we know what the LLM is doing especially the AI automations we built for them, and practically all the bigger financial orgs are going very heavy into generative AI too but they have the likes of PWC et al charging them millions for utter dross

antropy · Mar 10, 2026

OpenClaw is great, I'm running it in a virtual machine and using Codex from ChatGPT Plus.

For simple tasks it saves a lot of time.

It doesn't seem to be able to complete more complex ones yet.

Paul.

Cookie Consent

Essential

Analytics

Marketing

AI does have its uses

Founder of UKBF

Founder of UKBF

Founder of UKBF