Hey Reader,

2025 was said to be the year of AI Agents.

2023 was for chatbots,
2024 was for reasoners, so it made sense.

But 2025 didn't become the year of AI Agents, let's take a look at why.

Model Capabilities

AI tech companies have been busy trying to make better models to be the brains of the agents.

As a quick recap, an AI agent is a goal-driven system that can:

1. think (using an LLM),
2. remember
3. and do things (by writing code and sending requests)

AI tech companies were focusing on making better thinking models (called reasoners, like "GPT-5.2 Thinking"), and they also wanted to make these agent brain models take actions more accurately.

You see, the problem was that in an agent, the LLM Brain has to be able to accurately configure the request for the action to do.

In a nutshell, the LLM basically has to write flawless code for the request to go through without an error.

This is what we call "tool use" or "tool call". And in 2025, LLMs got significantly better at doing this, but not perfect.

What this means is that Agents now make far fewer errors in taking actions, but they still have a long way to go.

I predict this will keep getting better and better for the next few years.

Integration lags behind

According to this 2025 MIT Report, 95% of AI pilot projects created no measurable P&L benefits. Meaning that companies have invested $30-40 billion into AI pilots, and only 5% of them reached production stage with measurable value.

The bottleneck is not AI capability anymore, but how well it can be integrated into the business.

Most companies measured in this study treated AI as kind of a cherry on top, just sitting on top of everything as an afterthought, and not deeply integrated.

Instead of going at it from a strategic focus—what our problems are, where can we save time and money—companies chased shiny promises of a revolutionary AI solution that solves all of their problems and makes them millions.

The 5% of companies that have succeeded, however, were those that looked at back-office automations, where ROI is a lot more accessible and easier to measure than in customer-facing processes like marketing or sales.

This maps really well with my thinking (not to brag, but I was kinda right), that you should automate the mundane tasks you don't like to do first, so you can focus more on doing what you love.

I don't know about you, but I love to write educational marketing content like these emails or my Instagram carousels, and I wouldn't like AI to do that for me so I can do more admin work.

I want AI to handle the admin work so I can create more content.

What business owners need to do is to look at their systems & processes, and find operational bottlenecks, and invest into experimental AI solutions there.

For small businesses, it could just be a simple, well-written prompt that's stored in a folder for easy access to everyone, or a Custom GPT that executes the SOP the same way every time.

Once you have that, then you can think about fully automating it with a no-code solution. If this is the direction you want to go in 2026 and integrate AI into your systems to save time and money, check out the Paladin Guild, my new community offer to help small business owners integrate AI into their business in 5 hours per week without headache or a single line of code.​

People don't trust AI

I think this is another big roadblock against the advancement of AI agents.

People simply don't trust AI that much, because now they know that AI lies all the time (a.k.a. hallucination), and they wouldn't want to risk their business on this.

So integration projects go slowly as business owners are more careful.

And they are right to do so, AI still hallucinates, so we need guardrails.

This is why I recommend to start with back-office automations first and to use AI for tasks you're already good at, so you can see if it makes a mistake.

Back-office is also a lot lower risk. Who cares if an invoice draft comes out crooked? You fix it before sending to the client, no harm done.

Or if AI only expands on 4 ideas instead of the 5 in your 5 minute voice note, rambling about content ideas on your morning walk? Four ideas expanded on by AI and researched by the time you arrive home is a lot better than zero. And the last one you can do yourself.

Low risk, AI can't do you harm because the only harm it can do is within a closed environment only you and your team sees.

But these workflows are also the ones that can save a lot of time and money for you, as these are the things you do every day or week, and we only have a limited amount of time. Any time saved in back-office can be spent elsewhere in the business with higher yield.

My predictions for 2026

LLMs will keep improving, higher accuracy, less hallucinations, faster responses and more use cases.

If you have simple AI automations or just a few prompts collected as AI SOPs, you will benefit from newer models, just like how a new OS for your computer or phone makes things better without you having to do anything crazy or difficult.

If you have scaffolding for your AIs to work within, you can upgrade them and reap the benefits. If you don't, then you'll probably feel even more overwhelmed and restless as the world goes by.

AI is not showing signs of slowing down (although I wish it did so more people can catch up and we don't have to be afraid of a smarter-than-human AGI system), but until we have some breakthrough new approach, we will get these marginal benefits in performance and the big difference between companies will be how deeply is AI integrated into the systems.

Will we get AGI?

I don't know but I don't think so. AI companies will surely say they have it or they are close, so they can incentivize more investment money going their way. But even if we do, an AGI that's not integrated into your business is worth as much as a kiss to a dead person (Hungarian saying, means it's not worth that much, not useful).

I hope you have a Happy New Year and a very successful 2026!

Btw, I wanted to send a Christmas email as well, but I have fallen to the flu in the afternoon on the 24th and battled 40 degree Celsius fever (104F) for 2 days. Should've scheduled it, I know better next time! Sorry!

All the best,
Dave

Dave Talas

P.S: To help solopreneurs and small business owners integrate AI into their business safely in 2026, I'm launching a new community. Click here to check out the 5 Rank System to use AI in any business without code in only 5 hours per week. There are only 20 seats at launch, and it's application only!