This is a practical companion to my piece on what you need to understand about AI. Here we get hands-on with the tools themselves.
This isn't a comprehensive guide—it's a way in. If you've been hearing about AI but haven't quite known where to start, or you're not sure what all the fuss is about, this is for you. I've put together some exercises you can try yourself, to get a feel for what these tools can actually do.
The art of framing and refining your questions, known as Prompt Engineering, is rapidly becoming an essential skill in the workplace, in business and in creative pursuits. But don't let the jargon put you off. The best way to understand it is to have a go.
Even if you have concerns about these techniques, that's all the more reason to understand them. They're increasingly being used by others all around you.
Before you dive in, please take a look at my safety advice.
Trivial pursuits
Let's start with something frivolous, for the sake of amusement. If there's one subject that could do with lightening up it's politics, so let's give that a go.
I took the Wikipedia page on liberalism and asked Claude to explain it in the voice of Marvin, the depressed robot from The Hitchhiker's Guide to the Galaxy.
Ah, the mind-numbing verbosity of human literature, endlessly expounding upon the complexities of liberalism. It’s quite the downer, really. These so-called liberals, with their relentless pontificating on individual rights, liberty, and whatnot, can’t seem to keep it brief. They wallow in a sea of interpretations, each more convoluted than the last, all while claiming to champion private property, market economies, rights, democracy, and other such lofty ideals. The endless tedium of it all. This whole liberal ideology traces its roots back to the Age of Enlightenment, a period that witnessed the blooming of Western philosophers and economists. In their grand rebellion against hereditary privilege, absolute monarchy, and the divine right of kings, liberals championed ideas like representative democracy, the rule of law, and the sacred principle of equality… As if any of it matters.Here’s the result, judge for yourself
If you want to try this yourself,1 head to ChatGPT or Claude and ask:
Explain conservatism in the voice of William Shatner
Summarizing text
AI is quite good at summarizing text. Something computers were notoriously bad at just a few years ago, so the bar was low.
These kind of closed data sets are often where GenAI shines because you almost completely eliminate hallucination. You're asking it to work with text you've given it, not to make things up.
If you want to try this yourself,1 paste a document into ChatGPT or Claude, and ask:
Summarize the following text in fewer than 200 words, preserving key points.
If you're dealing with lengthy work documents, feed them into an AI summariser and use its output as a rough draft. Then revise.
Let's take our text from Marvin and put it through the mill:
Liberalism, born in the Age of Enlightenment, championed individual rights, liberty, private property, and constitutional government as a reaction against monarchy, hereditary privilege, and mercantilism. Philosophers like John Locke and Montesquieu shaped its foundations, emphasizing natural rights and separation of powers. The ideology fuelled revolutions in Europe and the Americas, becoming entwined with republicanism and democracy. Throughout the 19th and early 20th centuries, liberalism spread globally—from Victorian Britain to the Ottoman Empire—promoting reform, secularism, and nationalism. It clashed with rivals like socialism and conservatism, yet by the 20th century, liberal democracies emerged dominant after world wars and the Cold War. Its influence persists in modern political systems and values. Despite its lofty aims, the narrative of liberalism can seem endlessly verbose and ideologically rigid, often reduced to a philosophical loop of liberty, equality, and endless debate.Click here for the summary
It seems to have preserved some of the sentiment alright.
A notable mention here is Google's NotebookLM, where you can upload a collection of documents and interrogate it with questions. It can even generate a podcast for your ears, so that stack of boring policy documents you need to digest before the meeting can be packaged up for listening on your morning commute. It can make the mundane seem almost interesting.
But use with caution—there are real concerns about the privacy of Google's AI services versus the others.
Writing prose
Generative AI has become so good at writing prose that OpenAI abandoned its project to detect AI-generated content.2 Much to the dismay of teachers everywhere.
But there's still that niggling problem of hallucination. Often downplayed with the retort "humans aren't 100% accurate either". No, but the difference is the AI doesn't know when it's made something up. It sounds just as confident and can be quite persuasive—if you didn't know, you wouldn't catch it.
In a world where disinformation is already weaponized, automating the creation of alternative facts is a real concern.
That said, if you know your subject, AI can be genuinely useful for scaffolding your writing. If you struggle to get started on your outline, this saves time in brainstorming. You can refine it through follow-up prompts. But if you're winging it, be wary—perhaps it's missed something important, or worse, made something up. Verify everything.
If you want to try this yourself,1 ask ChatGPT or Claude something like:
Help me create a bullet-point outline for an article on the history of tea in England. Emphasize key turning points, influential figures, and fun cultural anecdotes.
I’d be happy to help you create a bullet-point outline for an article on the history of tea in England. Here’s a structured outline with key turning points, influential figures, and cultural anecdotes: Would you like me to expand on any particular section of this outline or add additional elements?Click here to see Claude’s answer to this prompt
History of Tea in England: Article Outline
Introduction
Early Arrivals (1600s)
From Luxury to Necessity (1700s)
The Rise of Tea Gardens (1700s-1800s)
The Victorian Tea Revolution (1800s)
Cultural Integration (Late 1800s-Early 1900s)
The “Cuppa” During Crisis (World Wars)
Modern Tea Culture (Post-WWII to Present)
Interesting Cultural Anecdotes (Peppered Throughout)
Conclusion
Or try asking it to adjust the tone of something you've written:
Make this email more formal and diplomatic while preserving the core message
Hi boss My head is spinning with all of the stuff on my plate this week: I’m a bit overwhelmed and already working late every day, can you help me out here?Sample input: draft email to my boss
Certainly! Here’s a more professional and business-like version of your email, while keeping the tone respectful and clear: ⸻ Subject: Request for Support – Current Workload Hi [Boss’s Name], I wanted to flag that I’m currently at capacity and finding it challenging to keep up with everything on my plate this week. Specifically, I’m juggling: I’ve been working late to keep up, but it’s becoming unsustainable. Could we discuss how best to prioritize or potentially redistribute some of these tasks? Appreciate your support, ⸻ Would you like to add any preference for what you’d like to be reassigned or kept? I can tweak the ending if so.Sample output from ChatGPT
[Your Name]
AI can shift tone and style quickly, helpful for switching between a business email and a lighthearted blog post. It's also democratizing the ability to write well for those who are less confident, or who may not be communicating in their native language.
A note on Grammarly
I mention Grammarly because of its widespread use, particularly among non-native speakers writing professionally in English. It's been around for a while to help improve your writing, and can run inside many tools such as Microsoft Word.
It's not without its problems. If you're a fluent, serious writer it will probably get in the way more than it helps. But as a product it's rapidly incorporating GenAI into its feature set. While often used by teachers to detect plagiarism—rapidly becoming redundant—it now offers to write articles for its students, which it can't detect as plagiarism. Go figure.
I don't use this tool much but I probably can't do better than this review from PC Magazine.
Creative writing
There will be plenty of fools who ask AI to write the story for them. And failure to disclose it could risk hefty fines.3 But for more serious writers, there are techniques in helping with the brainstorming process that are good to know.
If you want to try this yourself,1 ask:
Give me 10 unique characters who live in a remote Irish coastal village, with potential secrets or emotional baggage.
Here are 10 unique characters who live in a remote Irish coastal village, each with their own secrets or emotional baggage: Fionn O’Malley - The aging lighthouse keeper who hasn’t left his post in 40 years. He claims it’s dedication, but he’s actually hiding from a manslaughter charge on the mainland that’s long since expired. Siobhán Brennan - The village pub owner who knows everyone’s secrets but never shares her own. She left a promising career as a concert pianist after a nervous breakdown and still can’t bear to hear classical music. Dr. Aoife Quinn - The village’s only physician who moved from Dublin after losing a patient under questionable circumstances. She self-medicates her guilt and insomnia with her own prescriptions. Cillian Murphy - A renowned marine biologist conducting “research” on local aquatic life, but actually searching for a rare artifact rumoured to be lost at sea when his great-grandmother’s ship sank in 1912. Eoin Daly - The charming fisherman with a gift for storytelling who claims to be a bachelor by choice. In reality, he has three ex-wives and four children he’s avoided paying support for on the mainland. Maeve Gallagher - The elderly postmistress who reads everyone’s mail before delivery. She’s kept a detailed diary of village gossip for 50 years and uses this information to subtly manipulate village politics. Róisín Kelly - A talented young artist who paints seascapes that are gaining attention in Dublin galleries. She’s secretly the daughter of a famous artist who abandoned her mother, and she’s plotting revenge through her growing success. Father Patrick Doyle - The parish priest who lost his faith years ago but continues to serve the community out of a sense of duty. He’s accumulated enough village confessions to write a bestselling novel. Colm Walsh - The reclusive sheep farmer who lives on the village outskirts and speaks to no one. Locals whisper he’s a former IRA member in hiding, but he’s actually recovering from the trauma of losing his entire family in a house fire he believes he could have prevented. Orla Byrne - The new owner of the village craft shop who claims to have moved for the quiet life. She’s actually an investigative journalist researching a book about a decades-old unsolved murder connected to several prominent village families.Sample output from Claude
Seven of these will probably be garbage. But three might be interesting and something to build on.
Then follow up with:
Take characters #1, #3 and #7. Now give me 5 inciting incidents that would force these characters to interact in uncomfortable ways.
Here are 5 inciting incidents that would force Fionn O’Malley (lighthouse keeper), Dr. Aoife Quinn (village physician), and Róisín Kelly (young artist) to interact in uncomfortable ways: A violent storm damages the lighthouse, injuring Fionn. Dr. Quinn must treat him at the lighthouse for several days, discovering his old legal paperwork while searching for medical supplies. Meanwhile, Róisín, seeking shelter from the storm, gets trapped with them and recognizes Fionn from an old newspaper clipping related to the manslaughter case that matches a story her famous father once told her. Róisín decides to paint the lighthouse for her next exhibition and convinces a reluctant Fionn to let her sketch inside. While there, she witnesses Dr. Quinn delivering medications to Fionn and later observes the doctor having a trembling episode. When Róisín later collapses from an allergic reaction, Fionn must summon Dr. Quinn, forcing all three to confront what they’ve witnessed about each other. A Dublin art collector interested in Róisín’s work comes to the village and turns out to be the relative of the person Fionn accidentally killed decades ago. Dr. Quinn, who knows the collector from her previous life in Dublin, is asked to host a dinner bringing all parties together, unaware of the connections that will emerge over the meal. The village council proposes converting the old lighthouse into a tourist attraction and art gallery featuring local artists like Róisín, which would force Fionn into retirement and public scrutiny. Dr. Quinn, appointed to the health and safety committee for the project, must evaluate Fionn’s living conditions while Róisín is torn between her career opportunity and her growing sympathy for the old lighthouse keeper. A national newspaper publishes an exposé about medical malpractice in Dublin that mentions Dr. Quinn’s past incident. Simultaneously, they feature Róisín’s art and her connection to her famous father. When reporters descend on the village, Fionn recognizes one from his past case and hides the doctor in the lighthouse. Róisín, seeking the same refuge from the press, discovers them both there, leading to a tense confrontation about their respective secrets.Sample output from Claude
Again, most of these are probably rubbish, but one or two might be worth building on. Further interrogation through follow-up prompts could serve as a useful brainstorming exercise.
Conversational AI
So far we've been looking at the dry interface of a text box chatbot. We have yet to see this tech enter the world of voice assistants, but it's coming.
A decade ago the likes of Alexa and Siri were novelties that for a moment we thought might change the world. These days they've been relegated to kitchen timers. Though in the world of accessibility, voice interfaces have been transformative for people like me with a disability, enabling me to communicate and consume hundreds of audiobooks when I couldn't use my hands.
I highly recommend trying out the voice interface on ChatGPT's mobile app (Apple iOS or Android). Ask a few questions about any topic—and follow up questions. The difference in capability versus Alexa or Siri is astonishing. This level of capability is coming to your iOS, Android and Alexa devices soon.
Writing poetry
I keep hearing about its poetry writing skills. Sure, it can rhyme words and count syllables, but I challenge anyone to have it generate a poem, and then go read it aloud in front of another human. Personally I'd be mortified to be caught even reading it.
This is what I wrote in my first draft, which I sat on for a few weeks. Then by chance in my podcast feed, in the most unexpected of quarters, Richard Dawkins challenged ChatGPT to write a sonnet praising biological evolution, read it aloud, then critiqued it. He misses the point slightly on his surprise at it being non-deterministic.4 But he seems to think the sonnets are passable, with a few human corrections required.
Which leads me to challenge my own premise. As these models improve—if they eventually do start to churn out poetry that's better than their currently passable efforts, I mean, so what? Computers surpassed humans at chess decades ago. Did we stop playing chess? Did chess enthusiasts stop following the grand masters? Not one bit.
The more recent example is the game Go, which was thought to be too complex for a computer to master.5 But it did, and it produced techniques so unexpected that it changed the way humans play the game. Maybe some day AI will do something interesting with poetry.
I won't reproduce the exercise here. Instead I'll point you to Dawkins' experiment and his conclusions. It's a fun listen.
Writing code
So far it's the ability to parse natural (human) language that has been the key development of LLMs. So it should not surprise us that it can in turn write computer code—programming languages are more formalized in their rules and structure.
At least, it can generate code until something doesn't compile, and if you don't know the basics you can get into trouble. I've been using it in my own programming for a while now and it frequently makes up a non-existent API call6 or function. The old hallucination problem again.
However it really shines, I think, in helping humans both learn and write code.
Explaining code
I made a little open source project a while back to generate Kriskindle matches (a.k.a. Secret Santa) at a certain time of the year.
If you want to try this yourself,1 in your IDE7 install the plugin for GitHub's Copilot or Amazon Q. (Works with VS Code but also JetBrains, Emacs, even Vim.)
Here's the source code. Try asking:
Explain this code, what is it doing, and how does it work?
This code implements a “Secret Santa” or “Kris Krindle” gift exchange matcher. Here’s a simple breakdown of what it does: The program uses randomization and retries until it finds a valid solution where: It’s like solving a puzzle where you’re trying to arrange people in pairs, but some people can’t be paired together (like family members).Amazon Q explaining code
parse_input()):# Input format example:
# Bob Jane Mary # Family group 1
# John Sarah Tim # Family group 2main()):# Example output:
# Bob -> Sarah
# Jane -> Tim
# Mary -> John
It's actually done a pretty good job of explaining my code.
Writing a simple program
Let's write a small program. When learning a new language, the traditional first exercise is to print Hello World to the screen.
If you want to try this yourself,1 in your IDE with GitHub Copilot or Amazon Q plugin, create a new empty file, and save it as hello-world.sh.
Now open the Amazon Q or Copilot chat window, and ask:
Help me write a simple shell script that prints Hello world to the screen
Not only has it given us the starting code, it has explained how to make the script executable.8
Now, in our Hello World program, let's add a comment:9
And our autocomplete prompt has suggested the next block of code, based on the intent we've declared in our comment.
A more complex example
Now let's try something trickier and ask it to create a whole program:
Write a Python script that reads a CSV file of customer transactions and outputs the top 5 customers by total spend. Include comments explaining each step.
Here’s a Python script that reads a CSV file of customer transactions and outputs the top 5 customers by total spend, with clear inline comments explaining each step: Notes: Would you like a version that can be run from the command line with arguments?GitHub Copilot solution
import csv
from collections import defaultdict
# Define the path to your CSV file
csv_file_path = 'transactions.csv'
# Dictionary to store total spend per customer
# Using defaultdict so we don't need to check if key exists before adding
customer_spend = defaultdict(float)
# Open the CSV file for reading
with open(csv_file_path, mode='r', newline='', encoding='utf-8') as csvfile:
reader = csv.DictReader(csvfile)
# Loop over each row in the CSV
for row in reader:
customer = row['customer_id'] # Assumes CSV has a column named 'customer_id'
amount = float(row['amount']) # Assumes CSV has a column named 'amount'
# Add the amount to the customer's total spend
customer_spend[customer] += amount
# Sort the customers by total spend in descending order
sorted_customers = sorted(customer_spend.items(), key=lambda x: x[1], reverse=True)
# Print the top 5 customers
print("Top 5 Customers by Total Spend:")
for customer_id, total in sorted_customers[:5]:
print(f"Customer ID: {customer_id}, Total Spend: €{total:.2f}")
It's a pretty good solution, and GenAI is fast becoming an essential tool in every programmer's kit for writing better code faster. Its autocomplete capability can now suggest entire blocks of code which are unnervingly close to reading your mind.
Tool recommendations
Just copying and pasting text to and from ChatGPT works—it can answer questions and generate code quite capably. GitHub's Copilot is well worth the $10 per month for its inline suggestions alone, though there are question marks over whether and how it might store your code.1
Amazon Q is just as capable as Copilot, except that it's free for personal use, and does not store or harvest your code. That's my recommendation.
The capabilities go far beyond these simple examples. It's capable of generating a smartphone app, or an entire website.
The biggest danger
The biggest danger I foresee is idle talk of AI replacing programmers. From everything I've seen, it's changing the way we code, but the career of the developer is not going away any time soon. There's already a great shortage of people coming into this line of work, and the idea that it's going away only serves to put young people off choosing it as a career path.
Vibe coding
The latest buzz word is vibe coding, or using plain English to request an AI to write code. To be sure this is a wonderful democratization of programming, which until now has been some sort of black magic available to a select few.
Generating code, especially if you don't fully understand it, can be wonderful for prototyping and trying things out—better still for learning.
BUT WITH THAT COMES DANGER WHICH I WANT TO SPELL OUT VERY CLEARLY.
Deploying code that you don't understand, especially to the public cloud, risks spinning up resources with runaway costs—you could end up with a bill for thousands. And there are very real security risks. Please see my AI Safety guide.
I won't proselytize on this much more, but there's an interesting take on the future of software development in Pete Sena's blog that's worth reading.
Search and research
Using GenAI in day-to-day research is fraught with pitfalls, especially if you choose the wrong tools. Chat interfaces like Claude and ChatGPT have knowledge based on a fixed date on which their models were trained.10
However there are particular services emerging that have Google seriously worried for its monopoly11 share in the search market.
Perplexity AI is a search engine with a GenAI model on top. It parses the results of your search, summarizes them, and attempts to answer your question directly in plain English.12 Arc Search (Apple iOS or Android) does the same, but as part of the browser rather than a standalone website.
If you want to try this yourself,1 open Perplexity or Arc Search and ask something like:
How do the recent changes in regulation affect me as a first time house buyer in Ireland?
Recent regulatory changes in Ireland significantly impact first-time buyers through enhanced affordability schemes, relaxed planning rules, and accelerated housing delivery plans. Here’s how these changes affect you: The HTB scheme has been extended until 2029, offering a tax rebate of up to €30,000 (or 10% of the property price) for new homes or self-builds valued under €500,000. This reduces upfront costs, especially when combined with a 10% deposit requirement. This shared-equity scheme allows buyers to borrow up to 30% of the property price (capped at €500,000) at 0% interest for the first five years. Since 2023, first-time buyers can borrow up to 4× their annual income (previously 3.5×). For example: This expands purchasing power in a market where the median home price is €359,999. These changes collectively improve affordability and supply-side flexibility, though market pressures remain. First-time buyers should leverage state supports while monitoring interest rate trends.Perplexity AI full answer
1. Extended Help to Buy Scheme (HTB)
2. First Home Scheme Expansion
3. Increased Mortgage Borrowing Capacity
Income Borrowing Capacity €40,000 €160,000 €65,000 (joint) €260,000 4. Planning Regulation Changes
5. Market Challenges
Key Takeaways
This is a good use of AI simply because it has done an up-to-date real-time13 search of the web, summarized the results—but crucially, it has given us citations with real sources that can be checked. Of course, that's not the same thing as a fact-check, since any fool can put any old rubbish up on the internet (ahem), and the internet is where it's sourcing its answers. But at least you have sources you can verify.
Behind the scenes, this is powered by a search using Bing, finally finding a use for the much unloved alternative to Google.
What's key is that Arc and Perplexity run a real-time web search. The AI part is their ability to summarize the pages returned, parsing your question in an attempt to directly answer it. You get the results with citations to sources you can verify. This breaks away from the limitations of the mainstream LLMs whose knowledge is fossilized in time.
If you think about our usual patterns when researching—Google search, Wikipedia, investigate a few results and mentally assess their credibility—what Arc and Perplexity offer is some automation of this process. The limitation is that the AI is not always going to make the same decisions on credibility as you might. So it's up to you to check the sources.
Translation
This one has been around a bit longer and is better than it's ever been. In reducing the friction of communication, a fantastic resource.
But would you rely on an AI-translated document, without any human scrutiny, for a legal contract? A law? A medical diagnosis? I was recently involved in the translation of a children's book, and it reinforced my view that AI is a great accelerator, but much time and care was still needed to shape the language before it was ready to put into the hands of a child.
How good are they? Take a look at some interesting results as I put Chinese poet Li Po's poem Lament of the Frontier Guard through various translation engines and compared them with Ezra Pound's famous translation.
DeepL is a paid online service with a free tier, also available as a plugin for Microsoft Word and Google Docs. It does not store the documents you upload for translation.
LibreTranslate is free and open-source, and can be installed on your own computer. Running this locally means you can guarantee your own data security.
Google Translate is the most well-known, but also the most privacy invasive. However, for day-to-day use while travelling, it's hard to beat the Lens function that lets you see translation of text on a food packet or road sign. Its voice interface can enable real-time communication between people who don't speak the same language.
—
This may seem like I'm coming down quite negatively on all this. To be clear—what generative AI can do in 2025 is nothing short of astounding. I'm trying to articulate the caveats we'll all need to be aware of as it becomes part of our lives.
Footnotes
Much to the dismay of teachers everywhere, OpenAI's AI-generated content detector is shutting down—The Verge. Any service that claims to be able to detect AI-generated content is selling snake-oil.
The EU's AI Act applies, already being activated in its various member states. Spain for example: Spain to impose massive fines for not labelling AI-generated content—Reuters
Computers and maths are traditionally deterministic—if you give the same inputs you expect the same result. LLMs are non-deterministic, a deliberate feature of how they're designed and part of the reason they work at all.
Google DeepMind's AlphaGo—The Verge
An API call is a sort of instruction to the computer. AI making up an API call is equivalent to making up an impossible instruction like "count to three then levitate in the air for ten minutes". Another example of hallucination.
IDE = Integrated Development Environment, a sort of text editor for programmers where we write our code.
An executable is a file that can be run by a computer. This tells the computer the file is a program rather than a document.
A comment is a piece of text in a program that's not executed. It's used to explain what the code does. Comments are ignored by the computer when the program runs.
Generative AI models get their magic from learning algorithms which are trained on vast amounts of data. The model learns to predict the next word in a sentence based on the words that came before it. This is how it learns to generate text that sounds human-like.
Other languages exist.
As real-time as a web search can be, which is usually cached within recent weeks. Compared with Claude, ChatGPT or Gemini whose models can be 1–2 years out of date.