Subscribe to our newsletter to receive business tips, learn about new funding programmes, join upcoming events, take e-learning courses, and more.
Take the first step to successfully starting and growing your business.
Join for freePosted: Thu 8th Oct 2026
Do you have one AI subscription sitting on your desktop, barely scratching the surface of what it's actually capable of?
In this Lunch and Learn, Kiran Jayaram takes you through how to turn a single AI license into what feels like an entire extra team, without adding headcount.
Most people use AI like a search engine. But with the right setup, that same subscription can safely operate like a full agentic infrastructure, picking up recurring tasks, improving how it handles them over time, and freeing you up for the work that is taking too much time.
In this session, Kiran looks at the practical side of AI adoption and will run a live demo built around workflows submitted by attendees ahead of time, whilst getting the chance to practise applying it to your own specific use case during the session.
Walk away with an understanding of how to access funding that has been pledged for the Enterprise Nation community to upskill your business in AI.
Understand how to have your one AI subscription/licence safely and cost-effectively operate like an agentic infrastructure
How to implement AI so your business workflows self-improve
How to access funding to cover AI upskilling within your organisation
Kiran is Maikai's education lead, helping develop impact within small businesses and schools to upskill staff and ensure maximum impact of training opportunities. He loves sharing the transformations AI can bring to individual roles through small, practical tweaks.
Watch more expert webinars
Access a growing collection of expert-led webinars covering marketing, sales, finance, growth and more – ready whenever you are.
Lightly edited for clarity.
Beth Galloway (host, Enterprise Nation): Hello, everyone, and welcome to today's Lunch and Learn. My name is Beth, and I'll be your host today. For those of you attending a Lunch and Learn for the first time, Enterprise Nation is a vibrant community platform for start-ups and small businesses.
I'm very pleased to introduce Kiran, who is MyKai's education lead. In this session, Kiran will take you through how to turn a single AI license into what feels like an entire extra team, without adding headcount. If you have any questions during the webinar, please post them in the chat, and we'll do our best to answer them at the end of the session.
As always, the webinar is being recorded, so you'll receive the recording along with some follow-up resources later today. Over to you, Kiran.
Kiran Jayaram: Thank you very much. It's lovely to have everyone here, and thank you for taking time out of your busy schedules. My name's Kiran, and I lead all our education work within MyKai.
MyKai is a social impact organisation, and what we're here to do is help with whatever the major skills need is within particular sectors. Today, that means helping small and medium businesses use AI so they can really leverage it and compete with the big boys.
Today is about getting the absolute most out of your subscriptions, using an approach that lets one subscription act as any agent you need for any workflow. It'll be a bit of a speed run, so we'll definitely get to questions towards the end. I hope it's super valuable for you all.
What you'll take away is how to make the most of one subscription, whether that's Claude, Gemini, Copilot or whatever you're using, so that it acts as any agent for any workflow in your organisation. We sent around a diagnostic file and got some really interesting requests for workflows to illustrate in this session.
People asked about social media, including how to schedule posts and have the writing and heavy lifting done. We also had requests about cash flow and managing finances. There were a lot of different, really good ideas.
What you'll also get from today is a framework. I'm going to go through one workflow, but what's most important is that you can apply the framework to whatever workflow you're concentrating on in your business. That's the part to take away more than anything.
The workflow we'll go through is how to take a bank balance and all the transactions that have gone through the business and turn them into an effective forecast. We thought that would be relevant to as many small businesses as possible. You'll get a framework that lets you build this onto any workflow you need going forward.
We'll show you some different ways to run it, and how your one subscription can act at different layers of automation. We'll also look at how to build ethics into the process, so you can feel more comfortable and make sure your tool is operating in the right way.
Fundamentally, this is about setting yourselves up as an organisation so that if you use Gemini and Google went bankrupt tomorrow and Gemini got pulled, you could just turn on a different tool and run this through your infrastructure. It would be a 30-minute change, and you'd still have all your workflows ready to go and ready to automate. It makes you really resilient as an organisation.
One thing I'm going to add to the agenda is that we have funding to support small businesses with AI upskilling. We're running this in partnership with Enterprise Nation, so it's great to be able to offer it to the network. It's really simple: you book a call with us at the end of the session, and I've got a QR code for people to click into if they need to.
That's an expansion of the work MyKai does, but in and of itself, this session will be super valuable for you all. Taking a breath, I'm going to define a couple of things so that moving forward is really simple for you.
We have our AI models: Gemini, Copilot, ChatGPT and Claude, which is the one I use. What you'll see today can be applied across all of them. All these tools are just a model that one organisation has created and trained on massive amounts of data.
With every new model, they give it more and more data and test whether it gives the outputs they want to see. When the outputs are brilliant, they turn it into a new model and release it to the public. All these models are capable of acting as any agent you need.
What's important to understand, therefore, is context. When we talk about AI literacy and building the skills we need to be resilient, it's all about context. How can you break apart the context that AI needs in order to deliver a workflow?
You look at your workflow and ask what steps are required and what tasks are involved in each step. Then you decide which bit AI does and which bit you do, and how to make sure the process can be automated in the right way and checked regularly.
All an agent is is an AI model that has consistent context for one particular workflow. Lots of companies at the moment will say they have a social media strategy covering LinkedIn posts, Instagram posts, videos, and short and long content. They then ask whether each of those workflows needs a different agent.
Suddenly, companies are paying vast amounts of money to build an agentic infrastructure and calling it an amazing thing. For a lot of companies and individuals, that's a waste of money, because an agent's capability is simply specific, defined context given to your normal AI. We want to maximise the subscriptions we've got.
The language of an agent just means consistent, reusable context for one particular workflow, and that's all it is. What we'll show today is how to break up your context so your one subscription can act as many agents.
After a couple more slides, I'm going to jump off the slides and show you a demo. I've made a fake data sheet, like a bank report, showing all the transactions that have gone through a random business in a month. What we want to turn it into is a useful insight, which in this case is a forecast.
We'll run through how I've broken this process up. I started with some messy data and turned it into a forecast I can use to generate insights and put to work in my business. Separating out the workflow is really important.
The first stage is cleaning the data, so it's much easier for both the AI and ourselves to read. Within that stage, the AI will read the bank report, remove any duplicates and label each line. Whatever your workflow is, you break it into stages in the same way.
To produce a LinkedIn post, for example, I need to read 10 other posts in a format I like. I then need to understand the tone and the language as the next stage, and the third stage might be writing. You ask what tasks are involved, then decide that you'll do this bit and AI will do that bit.
For this process, we clean the data, then analyse it, and then produce our output, which is a forecast. That will be a slide you could show the CEO and colleagues and use to report across the organisation.
It's also really important to consider your ethics. AI ethics can sound vague, but you can embed them into the process I'll show you next. I won't go into loads of detail because it's a bit philosophical, but here are five principles of AI ethics from Luciano Floridi, a brilliant Oxford academic.
Beneficence simply means do good, so ask whether your AI is doing good. Non-maleficence means do no harm. Autonomy is your human agency: where are you in this process, what are you deciding, and how are you making sure it's your process?
Whenever you automate, you're responsible for the entire process. Justice asks whether it's fair: are you using people's data in the right way, where does the cost go for a process, and is it a fair use of AI? You can break these down and think about how they apply in your context.
The most important one is explicability. I'll add a paper to the follow-up that explains all five principles. If someone asks where all the information goes and why you've automated a process, you need to be able to explain it.
The process I'll show you next lets you break that up, so if someone comes to audit you, you can show the process and the elements we've broken out. You can explain why we made decisions at each stage.
Before I show you the spreadsheet, here's an element that's really important to get. It holds the whole approach to getting one subscription to act as many different agents. I've broken my workflow of cleaning, analysis and forecasting into files and folders, which is really simple.
Classically, you'd feed in the spreadsheet and say, 'Please analyse this and produce me a forecast for the next 13 years.' There's a lot of risk in that, and you also have to do it manually every time. Instead, you imagine your prompt and break it up into files and folders that separate the process out.
This is important because if you use Gemini and want to switch to Claude next week, the whole process is still embedded in your files and folders. That's really all it is. I'll quickly describe the types of files and how we've broken the process up.
First, we have a context file, saved as context.md. You might ask what .md is: it's just a markdown file, a type of file like a Word document. The only reason we use it is that it's much easier and more efficient for your AI tool to read.
Reading a Word document means the AI has to convert it into its own language and then convert it back into English for you. A markdown file is in a language it understands, and we can still read, analyse, and check it in English. It's really important for efficiency and for saving tokens and energy.
I've named the first file 'context', but you can name it what you like. All that's in this file is what the folder is for, what the workflow is, who we are, what our ethics are and what our company policy is. You can use your AI tool to help you write these files.
It's not a long process. I built this workflow in about an hour and a half last night, with everything you'll see, and I now never have to do that again. I can run the process every month if I want to, without putting in that hour and a half ever again.
The second file is the process file, which simply describes the process. You've broken up the context you'd typically give Claude in one prompt, and this file describes what happens and when the human check comes in. Then, to avoid overcomplicating things, we have three folders: clean, analyse, and forecast.
Inside each folder is an instruction file, another markdown file, setting out what to do in that folder. For a cleaning step, it sets out the instructions and stages the AI needs to go through. We have that for each piece.
Some processes also need a script, because when a process requires any mathematics, AI is not very good at maths and will make mistakes. Python is a coding language that will do the maths for you, and you don't have to know how to use it. This workflow doesn't include one, but I thought I'd mention it.
Large language models can't actually do mathematics. They go onto the internet, find examples of that maths and try to work it out by reverse-engineering it, rather than doing the mathematical process.
The final file, which isn't 100% relevant, is a claude.md file, though it might be a gemini.md file. It acts as a map: when you're cleaning, go to 01, and when you're analysing, go to 02. It might be only four lines.
I'm going to show you three layers of automation, though we won't have time for all of them. First, you might just do this process in the chat, which I'll show you now. Second, you have projects in Claude, notebooks in Copilot and gems in Gemini, and ChatGPT has a version as well.
A project is somewhere you can save context in one area, so you don't have to keep uploading it every time you start a chat. The third layer is editing the folder within a coding system, which is a bit more complicated and takes a bit of time to describe. If we get there, I'll illustrate what that looks like.
I'm now going to jump out. As a quick example, this is what the dataset looks like: a screenshot of something like a bank balance. We're going to produce this output, one slide ready to sign off, and I'm very aware this is a proper speed run, but I hope it's clear.
Here's our folder, and I literally mean folders. We've broken this up into clean, analyse, and forecast, and here we have our context file, which says what the folder is for, what our company ethics are and so on, and then our process file. Within each stage, we have the data sheets, which is the screenshot I just showed you, and the instructions.
Let me show you what a markdown file looks like. Don't worry about this app; it's not important at the moment. Inside is what the job is and what to do, then how to review the output, which will prompt me with some questions, and then the rules for this particular stage.
That's all that's in there, and we can share these afterwards so you can review them in your own time. The process then gives us an output, which I save into this folder. As you can see, every bit of the process is broken up like this.
Now, how do we actually get our tool to do this process? There are various ways, and as I said, we have three layers, the first of which is the chat. Typically, you'd give it a spreadsheet and say, 'do this for me', and you'd have to spend time typing out all that work.
Now that you've stored your instructions, you don't have to. You go straight into your files and say you want to run your first process, cleaning. I have my instructions and my data sheets, and I also have corrections and categories files, which are part of the cleaning process.
The categories tell the AI how to categorise information when cleaning the data. For example, a payment to a restaurant is an expense, while a payment from a certain type of business is recognised as income. You can add whatever categories you want.
The corrections file records the nuances that come up. Every time you run this process, it won't be perfect, so you iterate, and anything you add means the next run won't repeat the problem or will handle it better. Now I'll ask it to run the instructions, and hopefully this won't take too long.
I'll do this first step and then move on to the other way of doing it. At the moment, Claude is reading the instructions, and Gemini and Copilot would do the same. It goes into my data sheet and follows the notes in our instructions file.
It understands what's in the data sheet and goes through each step, and it's taking its time, which gives me time for a drink. It has found the opening balance, which is part of our process, and given us the totals. It's done the analysis I asked for, and some checking, including the maths, and it's got a zero.
For the cleaning, it has a couple of questions for me, and it's telling me the work it's done. I can also ask it for different outputs. You've got a couple of decisions to make in this process, which is the really important human in the loop.
First, there's an unknown car payment: £200 has gone somewhere, and it's asking what it was for. Then there's a double-charged train fare that has potentially done you over. The trains are terrible, so as you know, you've cancelled one, tried to buy another, and it's all a bit of a mess.
We've saved a lot of time already, which is great, but you'd have to do this for every single file and folder, which is a bit of a pain. We'd rather do it all in one. You can give the whole folder to Claude in the chat, but that requires quite a lot of energy.
What you want is to go into a project, notebook or gem and store all this context there. Here's one I made earlier, where I've written some really simple instructions. Remember, you can ask your AI to help you write this process out, and I'll send around a paper after the session that explains it really well.
You can also give your AI that paper and say, 'Help me write this for my process.' That's the brilliant thing about AI: you can work with it to do this for you. The first instruction is always, before doing anything, read my context file.
Instead of going through all the files and folders every time, I've added the entire forecast folder, which represents my whole process, to the project. I've got my new monthly dataset, so I'll add files: this is our data sheet, the screenshot. I'll tell it we have a new monthly cash dataset, and I can't spell, but Claude can, so it's okay.
Instead of running the entire folder straight away, Claude first reads the context file so it knows what we're doing, what the stages are and what we're trying to achieve. It then goes to stage one, reads that and executes, and asks me questions because that's what I've instructed it to do. It will go to stage two when I ask.
This saves a lot of energy and time, because instead of putting the whole folder into a chat where it has to read everything before executing, you go straight into the project itself. It's asked me some questions here, so we'll speed through them. One, this is an expense. Two, it's a duplicate.
The human-in-the-loop process is built into the instructions, because those nuanced situations come up. It's now asking whether I want to add this answer to my corrections file, and Claude will suggest things like this to you a lot, as will Gemini. It says it can't apply either answer yet and asks whether I want to keep both lines back, so it's constantly checking with me.
It's asking me quite a lot, so I'm giving it quick answers and moving on. I know we're getting close to time, but don't worry, we're nearly at the end. It's now running through each stage: step two, the analysis, is done, and it's given me some outputs and flags, which I'll answer.
What's really important here is that you're now spending your time analysing the process rather than doing the manual work. You're checking the output and using all your expertise to make sure these things are correct, and you can do this with any process. I hope that's coming across, because you're now spending your time on the most important judgment decisions.
I'm hoping it will now produce our output and look brilliant, and then we can call it a day. When I ran this before the session, it went a little better and didn't ask me all these questions. Now it's building the forecast and giving us the insights I asked for in the instructions file, and here we have our slide.
That's a lot of information in one go, and I'm very aware of that, but I hope it helps you divide your context up and shows how you can apply it to new workflows. I'm very happy to answer questions now and help with any particular workflow you want to apply this to. On a final point, you can then start to extend things.
The core message is to set yourself up in this way, with files and folders. I know it's boring, but this is what will make your business really robust to whatever changes are going on with the tools. We don't want to get into that rat race.
Brilliant, here's the final slide with the QR code. If you want to talk to us about the funding available for AI training, there are multiple programs for different roles and responsibilities within small businesses. The funding is limited, but we want to get it out to as many people as possible, as quickly as possible.
The most important thing is a diagnostic call, where we understand whether you're eligible and the funding applies to you. Thank you, everyone. Are there any questions?
Beth: Thanks so much, Kiran. That was great. We've had some really nice comments in the chat: Ina says this is super interesting and a new way of organising that is far more efficient, and Laura says it's a lot of work processed so quickly, and she loves the check-ins.
You talked about the human-in-the-loop element, which is a really important safeguard for checking the AI outputs. Beyond that built-in human oversight, what checks or safeguards should a small business have in place to make sure AI is being used accurately and responsibly?
Kiran: There are a few first things, and one is really important. The government has issued guidance on automation, which is a change in the rules. Automating anything that affects people, such as using their data or making decisions about them, used to be a complete no-go, and they now allow it.
This has made it a bit easier for you to run, because they know AI is coming. It's important to check those rules, so go straight to the government, which has guidelines and PDFs on all this. You can feed those into your tool and ask it to criticise your process in relation to them.
Whenever you break your process into steps and then tasks, it will be really clear which tasks are for you and which are AI tasks. Ask yourself the explicability question: where will the data go at this point, and is that okay? If it's financial or company data, is your AI tool a closed loop, or is it training bigger datasets?
There are frameworks out there, and many people have done so much work on safeguarding this stuff. It's a case of doing 10 minutes of deep research and finding some of them. Building your ethics into the files and folders will take a lot of the stress away.
Beth: Absolutely. Sorry, I know we're out of time, but if I could keep you for a few more minutes, that would be great. Elizabeth would like to know about the AI training and funding you mentioned, and what the process is around that.
Kiran: That's a really good question, because I've taken my slide off, so you can't actually do the thing I asked you to do. This is an initiative to support small businesses across the UK with the major skills they need. The first step is just a call with us, which is really simple: fill in the little form on the QR code.
We have about £1,500,000 of funding to go towards particular programs that can help small businesses with AI. The programs are worth about 15 to $18, depending on the type of program, but the funding covers everything. All we need at this stage is for you to fill in the form via the QR code.
If it doesn't work, we'll send these slides and everything afterwards, so just get in touch, and I can put my email on the slide as well. We'll book a really quick 20- to 30-minute call, and I'll go through things individually, because we check eligibility with everyone. There might also be some help we can provide that doesn't require any of that funding, so just follow that QR code.
Beth: That's great, Kiran, and sorry, I can see myself on your screen, which is a bit off-putting. Just to wrap up, could I ask one more question? What skills do employees need to develop as AI takes on more of these routine tasks?
Kiran: The most important thing is to ask yourself what your expertise is and what you're good at. Law is a brilliant example: there was so much writing, so much manual research and a lot of this work being done by hand. But the value of a lawyer is their judgment.
Within your particular sector and what you're interested in, your judgment is now the most valuable thing. In the short term, view AI as a way to break up your workflows. Make sure you get all your processes broken out so that you can be the one in your company doing this for your organisation.
You want to be that champion and the person driving the change, because that will keep you resilient in the short term. But if this isn't your thing and you're not an AI person, and you just want to automate some workflows, then you should concentrate on your area of expertise. If you're a lawyer, solicitor or accountant, ask where your judgment becomes really important in this process.
Get your basics, but then do anything you can to enhance that judgment and your human capabilities, such as leadership, your ability to communicate with people and deal with conflict. All of these will increase in value. People will pay you for that, because they're now saving money on all these workflows they don't have to pay for.
Beth: That's a fantastic response to end the session on, and I really like that. There are lots of thank yous in the chat, Kiran. Thanks so much for joining us today.
Please do connect with Kiran, as I've popped his LinkedIn profile in the chat. Scan the QR code, and as I say, we'll send some further resources along with this recording later today.
Thank you very much for joining us, and thanks, everyone, for having lunch with us. We will see you in the next one. Bye for now.
Kiran: Take care, everyone.
Unlock more on-demand sessions designed to help you sharpen your skills, grow your business and stay one step ahead. Find more Lunch and Learn webinars