Predictions: Artificial Intelligence 2

Agentic AI

Last year was all about large language models, the buzzword for 2025 will be Agentic AI and you will see new startups launching daily touting the latest and greatest agent technology.

Agents however are nothing new, in 2002 I won a contract with eresMas (Spain) to build intelligent agents in Java to route communications and grab and structure content for their internet platform. The system managed tens of thousands autonomous agents running across a network of three-hundred servers each carrying out their own designated tasks and had the ability to let agents create their own agents to delegate tasks further.

Then the world of agents went kind of quiet. The buzz was over.

But here we are in 2025 and agents are back, but with powerful AI to help them do their jobs.

Agentic AI refers to AI systems that can independently perceive, reason, act, and learn without continuous human guidance. This marks a significant shift from traditional AI, which primarily focuses on responding to specific prompts or instructions, and the ability to reason and learn is a significant shift from the original agent paradigm.

Agents are no longer dumb or controlled by simple if x then y logic.

Right now you can run an AI driven agent system right on your own pc. I have an open-source version of n8n setup on a $300 Linux box in my basement. You can also run it free on the web.

n8n allows you to build agents that can interact with any AI (including DeepSeek), and integrate with over 400 services on the internet. If you are hosting it yourself you can install hundreds of community written plugins to extend the functionality further.

But this is not an article about n8n, that is for another time…

Now companies like OpenAI, Anthropic, and Google are already touting agents and showing off their ability to take control of your screen to understand and assist you with what you are doing (look at Operator). But they are barely scratching the surface here.

Development agents such as Replit Agent can now create full-stack software applications and automatically deploy them in the cloud simply by chatting with an AI, and if you haven’t tried one of these out and are sceptical about their ability then you should have a go.

A couple of weeks ago I wanted to write a quick python app to scrape some data using a simulated browser, store the data in a database, then run reports on that data. Nothing too taxing for a developer, but then I ran into a bunch of problems I had to overcome because the site was doing its best to obfuscate the data returned to protect it. After a couple of hours I gave up.

A couple of days later when I was watching TV I tried the same thing with Replit Agent. After writing my requirements and getting back some initial code, I tested it and there were a few errors which Replit Agent quickly fixed, ten minutes later I had the app live on the web with the database setup and a simple but effective frontend deployed. Total time to develop and deploy the full app, less than thirty minutes.

Agentic AI now gives anyone the ability to develop applications, and this year it is only going to get smarter and more efficient as AI learns to interact better with non-developers and understand their needs.

But these are just the things that are front and centre right now. AI agents will make significant changes in almost all service related industries.

The first of which is Customer Service and Support. By now we are all used to chatting to bots. But these bots will make significant leaps forward with access to reasoning models they will be faster and provide a more personalised offering giving a more accurate and less bot-like interaction.

I expect many companies, especially in the banking, finance, ecommerce, and healthcare industries to roll out fully voice driven support services, all powered by AI agents capable of autonomously handling most queries and responding to proper questions rather than expecting a yes/no type answer.

This will especially be prevalent in Healthcare.

The automation and information access and provision across all manner of healthcare providers will make the system more efficient and this will include.

• Making appointments and reminding people of their upcoming appointments, together with the ability to reschedule appointments.

• Reviewing existing medication and following up with patients to make sure the medication is still suitable for them.

• Contacting the patient and providing test results and answering questions about the results together with the ability to schedule follow-up appointments.

• Monitoring patients in real-time. A patient would wear specific monitoring devices at home and agents could monitor their health status and alert necessary medical personal of any issues or report changes to their doctor.

• Making decisions on scans and other medical results. AI is used to analyse a lot of medical imagery, but an agent based system could be use to schedule follow-up automatically, or to look through patient data, results, or interactions to spot problems in their condition earlier and take action.

The Finance world is no stranger to AI and algorithmic trading and for years the finance industry has been using ‘bots’ to analyse companies, slurp social media posts and look for trending signals and sentiment so there will be less of an impact in this world than others. Where agents may begin to appear is in risk management and advanced trend prediction although they are already there to some degree.

Consumers will be exposed to personal finance agents that they can finetune to invest on their behalf and I expect a flurry of companies to launch this year offering this kind of service.

Much like the finance world, Manufacturing and Supply Chain Management already have some of the most advanced AI systems in-place. This is especially true for large manufacturing industries such as automotive and electronic device manufacturing. You can expect this technology to trickle down into smaller companies this year but it wont have as big an impact as some experts are predicting. Where this may be used though is in deploying agents to detect potential manufacturing problems or deal with Human Resources by automatically rescheduling certain manufacturing jobs or even staff rotas based on availability of humans in the work-flow process.

And speaking of Human Resources I can see this whole process of hiring and finding potential candidates moving to an even more automated process with the majority of potential candidates being screened by AI before a human becomes involved. AI and agents will also be used to make ‘first contact’ on sites like LinkedIn.

Currently, a lot of middle-tier and technical roles are screened by AI but this will filter down to even the most menial of jobs, with references being checked by AI before a potential candidate is even aware they are up for consideration. When they have been selected as candidates AI agents will automatically schedule interviews.

But the largest markets to be affected by agentic AI will be Research & Development, and Office Work.

On the R&D side AI is already being used to research new drug developments but going forward AI agents will analyse data, read scientific papers, and identify potential new drug targets, predicting their efficacy, assessing the financial risk of taking research further, and ultimately accelerating drug discovery.

AI agents will be used to analyse large datasets, identify patterns that humans just cant comprehend, and generate hypotheses to assist in new scientific discoveries. Scientists will also be able to chat and develop theories with agents using them to launch numerous agents to gather research papers, extract the understanding and meaning and to do deep thinking to present a hypothesis or just provide detailed knowledge on a subject.

This wont be limited to scientific or medical discoveries, all manner of industries will begin to implement reasoning based AI and agents to assist with research and development.

With Office Work I see agentic AI as both a positive and negative. From a positive point of view I see most of the menial day-to-day tasks being simply replaced with agentic AI; Collecting data, invoicing, chasing bills, creating schedules, booking hotel rooms and accommodation, writing reports, sending emails and writing follow-up, summarising information, processing receipts, banking and bill payments, paying wages, handling basic HR requests such as booking holidays, maternity leave, sick leave etc.

All of these can and will be replaced by agentic AI assuming a company wishes to do so. The saving on scale would be massive, but people might not be comfortable with it… and a lot of people will lose their jobs.

Even more technical jobs will be at risk. Developers even though they will still be require for the foreseeable future will see their jobs threatened or team sizes reduced. Right now AI is capable at worst of replacing interns, and in most cases replacing low end development roles.

But the question you have to ask is, if low-end positions and interns are being replaced by AI, then who will be in a position to take more senior roles in the years to come.

Another industry I also watch is Cyber Security. This is an ever evolving cat and mouse industry with adversaries developing new exploits and cybersecurity experts patching holes and defending against them. What has emerged over the last year and I suspect will explode this year is proactive agentic AI defence capabilities.

These will be agents that not only monitor networks for intrusions and watch for anomalies, but also proactively run tests against their network and try to actively find holes in software and hardware run on it.

Right now as a potential adversary I can use specialist uncensored AI to write undetectable exploits, or create new tools to test networks, and if I can do this you can bet there are thousands of people out there right now who have malicious intent doing the same thing.

This will be a major battleground this year and I can see a lot of disruption ahead.

But Agentic AI will also be a force for good. As you know I am pretty much a geek, and have used AI from the start, but I had a bit of a mind-shift last year.

I see a lot of people saying things like ‘I don’t want to pay $20 a month for a ChatGPT subscription’ and similar statements.

How about reframing this as ‘I can get a full-time employee for just $20 a month and I don’t have to pay any benefits on top’

Now for me, AI has always felt a little restrictive. It has been hard to get some results, hallucinations can be a pain, writing good prompts can be just as time consuming as doing the work yourself and so on.

I think this has now changed.

Using services such as Make automation, or my preferred tool n8n you can now build a full team of employees, each with a single role, and you only pay for them when you need them.

You don’t need an OpenAI subscription, you can buy some API credits instead, or as of now buy credits on DeepSeek which is way cheaper. If you need images then you don’t need MidJourney, again, there are a ton of companies offering APIs to models such as FLUX Ultra which in my opinion is better.

n8n allows a free account to run 2,500 tasks a month for free, each task can have an unlimited amount of steps so you can build amazing tools such a researcher that will gather all of the latest news in your niche, read all of the articles and summarize them providing content to put in your newsletter.

You can build personal agents that will schedule activities in your calendar, send emails on your behalf, answer customer queries on WhatsApp.

All of this you can do right now for pennies in terms of API costs… or if you have a machine with a decent GPU you can run all of the AI intelligence locally.

AI and now Agentic AI has opened up a lot of opportunities for us, I am using Gemini Deep Research (which costs $10 a month for the first couple of months) to accomplish hours and hours of research for a video project I am working on. Having even a basic employee to do this would cost $45,000 a year +

Start thinking what you could achieve with an extra team-member or two, and how you could begin to automate tasks.

If you would like me to share some of my workflows then let me know.

Up next: Part 3 of my AI Predictions.