AI is evolving fast, and the latest buzz is all about Agentic AI. These digital assistants don’t just wait around for instructions, but take action independently (a little scary, but also pretty cool, right?).
From transforming industries to offering businesses and customers more proactive solutions, autonomous AI is changing the game. But while the possibilities are exciting, it’s important to consider the ethical challenges that come with them.
To break down what Agentic AI really means, how it can be leveraged, and what we should keep in mind when using it, we sat down with Alex Mold, ASquared’s Lead Software Engineer.
How would you define agentic AI, and how does it differ from other types of AI?
AM: Agentic AI is a type of AI that can make independent decisions using its own reasoning mechanisms.The key difference from typical AI is that agents don’t just generate a single response and stop there. Instead, they work through complex processes, adapting as they go. If something goes wrong, they adjust and try again before moving forward. This makes for a more hands-off user experience! Once you set the direction, the agent handles the rest with minimal human supervision.
Could you give us some practical examples of agentic AI in digital products that you think are doing well?
AM: Absolutely. We actually leverage this during our development process at ASquared through using Cursor (a fork of Visual Studio Code) which allows AI agents to interact with our code. By feeding Cursor with a defined set of rules, code styles and goals, we’ve managed to improve output across all our projects.
Let’s take a simple task, like creating a dialog component to display some text. While this is easy enough to do on your own, Cursor can generate it in just seconds. Not only does it produce code tailored to your framework of choice (in this case, NextJS), but it also makes sure everything has accessibility and reusability in mind. Plus, it also knows where other third party components are from, so will run commands to install any outlying dependencies too (saving you the effort of doing it yourself, which, let’s be honest, is the dream).
Of course, it’s not always perfect and the code should be carefully reviewed, but it provides a solid starting point and can often take a feature much further along in the process than starting from scratch. And, for more complex features, seeing the stream of its reasoning and the steps it's taking can also be a great insight for developers on how to approach something new. The key is knowing where to step in, spotting any mistakes, refining outputs, and making improvements. It should be used with the intention of enhancing efficiency and maximising client budget ROI, rather than relying on the agent becoming the developer.
Beyond the development side of things, AI agents are being used more and more for user-facing applications. A great example is Notion’s AI, which can autonomously draft content, summarise notes and suggest actions to users.
What human input is needed in the process of setting up an agentic AI system?
AM: It really depends on what you want to achieve. For example, using Cursor as part of our workflow doesn't require much in terms of setup. If you have a codebase and an IDE with an AI model enabled in agent mode, you're already good to go!
However, if you want the agent to perform at a much higher level, you would need to write instructions for how you want it to interact with your codebase to avoid any issues. Lots of AI agents will just assume certain packages are installed in a codebase and then fail when they run commands that use this, so there will be a bit of hand holding at the start.
In many cases, AI agents are designed to perform tasks with minimal human intervention. There are various services available to help businesses integrate AI agents, whether for customer support chats, automated meeting scheduling, or supply chain optimisation. Whatever the use case, the important thing is setting up the agent with access to the right data and giving it the ability to take action. This might involve communicating with an API or an external system to complete its tasks.
These workflows can be implemented through code, but no-code tools like N8N and Chat Thing make it far more accessible for non-technical users to build AI driven workflows.
.png)
What do you anticipate to be the biggest challenges in designing and building agentic AI systems?
AM: I think that one of the biggest challenges you might face is ensuring the agent performs the right tasks based on the input they receive. The key factors here are defining intent (clearly outlining what actions the agent should take) and providing context (the more the better). Unlike traditional AI that generates a single response, AI agents process data and make decisions that align with the system’s goals. So, it can get particularly tricky when the agent is dealing with ambiguous instructions or incomplete datasets. This is why you need to use it wisely to increase efficiency on tasks you could largely otherwise do yourself, so you can spot issues and fix them.
With that being said, it’s also super important to consider the type of data you’re working with. Ideally, sensitive information (e.g. personally identifiable information or API keys) shouldn’t be fed into Large Language Models. Data security must be taken very seriously and handled with care.
Do you think it's important for businesses to consider implementing agentic AI to stay innovative?
AM: I think it's definitely something to consider if you’re looking to automate processes that are perhaps time-consuming or costly. If we look at it from a birds eye perspective, the whole point of agentic AI is to make our lives easier, not harder. It has the potential to impact a wide range of industries, especially where people's time is spent on more menial tasks.
However, I do also believe that in order for it to have a great impact, it does come down to the system design. A poorly implemented AI agent workflow could end up taking more of the company's time to fix. So, it’s important to assess where your pain points are and determine whether agentic AI can genuinely alleviate them and add value to your business.
Can smaller-scale businesses benefit from agentic AI, or is it primarily for large enterprises?
AM: Absolutely they can! In fact, the initial set up might be simpler for smaller businesses, as their day-to-day workflows are often less complex than those of large enterprises.
For example, a small e-commerce business could use AI agents to manage customer enquiries, automatically update inventory, or handle appointment scheduling. Adopting Agentic AI is something a business of any size can do, but my suggestion would always be to implement it incrementally so that you can keep assessing its true value over time.
.png)
How does the user experience change when agentic AI is integrated into a product?
AM: This is quite difficult to answer because it really does depend on the implementation and execution.
AI agents aim to reduce the number of steps a human needs to take to complete a task by handling those steps themselves. In terms of user experience in business, this typically means an employee only needs to kick off the process by describing an issue in natural language. For example, if someone wants to set up a meeting, the AI agent handles all the heavy lifting, from finding a suitable time for all parties to then suggesting a time back to the user and sending the invite. You can also add frameworks which prompt you if the information you’ve submitted is unclear or missing certain details, to ensure you get the best quality responses.
The goal of implementing agentic AI into consumer-facing digital products is to dramatically reduce friction in the user flow. When it’s done correctly, it should provide a better overall experience for everyone involved.
Are there measures in place to ensure agentic AI solutions remain ethical and safe for users?
AM: Ethical AI design continues to be a hot topic. As a rule of thumb, I tend to think about it from two sides…
The first is the user interacting with the system; they should be aware that they are engaging with AI and understand what kind of data the process uses. The second focuses on how the system handles the data. It’s important to ensure that non-essential user data is removed from any processes and that guardrails (such as content moderation and data protection policies) are in place to limit any potential misuse.
There’s been some discussion around the term "AI agent" and whether "AI employee" is more appropriate to avoid implications of direct competition between humans and AI. Do you think terminology like this matters in shaping perceptions?
AM: I do think the language we use around these topics can certainly shape the culture and perception of AI. Personally, I’m not convinced by the term 'AI employee', as it frames an AI process as something much closer to a human employee, rather than a distinct process (in my mind anyway!).
I'm not sure what the best term is, but from a development perspective, we're using LLMs to help us in our day-to-day tasks and application processes, which I think is fantastic for productivity. However, they still need a human to oversee and make decisions where necessary, or to engineer them in the first place. For that reason, I believe it's important to compartmentalise these terms, at least for now.
Is there any advice you would give to companies exploring the adoption of agentic AI?
AM: Yes, start small and iterate. Implement agentic AI in low-risk workflows first, and assess how it's performing for your needs. If you clearly define your objectives, you can monitor how things are going and make tweaks as needed. Iteration often involves tweaking prompts and behaviour, then adding more features once you’re happy with the results!
As I mentioned before, it’s also crucial to use AI ethically, so make sure that your data protection and privacy policies are in place. We’re still in the early stages of agentic AI, so it’s a good idea to stay up to date with the news around these kinds of topics as things will always move rapidly.
-
At ASquared, we thrive on helping businesses design and develop digital experiences that help them stand out from the rest of the crowd. Whether you're just starting to explore AI possibilities or already have a vision in mind, we're here to help you turn that vision into a reality. Let’s explore how AI could elevate your current or future digital project.
Let's see what we can make together.
From startups to scaleups and enterprise, we're always happy to talk to impact-focussed and ambitious organisations. Use our quote creator to get a better idea of scope, timescales and next steps.
