Hi there,

Making predictions on AI is a terrible idea. Just ask the team that built ChatGPT - even they didn't see their own breakthrough coming.

And yet, here I am, making AI predictions for the third year in a row, because I've learned that even imperfect predictions can be useful in preparing what's next.

So, let's look at what worked (and what didn't) in my predictions for 2024, and what I think will shape AI in 2025.

Let's dive in!

Want to make 2025 the year where AI adds Profit to your business?

I've opened up reservations so you don't miss out on the 5 spots I'm opening in January to work directly with me.

If your goal is to uncover (and implement) hidden AI opportunities in your business that will add at least $10K+ in recurring profit, reply "January" and I'll get you details.

Looking Back: What I Got Right (and Wrong) in 2024

Let's start with what I got right: no GPT-5. Instead, we got o1, which showed that LLMs can produce better outputs by taking more time to generate them - not just by being trained on more data.

Speaking of putting AI to work - my prediction about augmentation beating automation held strong. Effective AI augmentation is still your best bet, especially when starting your AI journey.

I was also right about open source and RAG (Retrieval-Augmented Generation). The open source AI Stack offers a compelling alternative to proprietary solutions, and RAG has become the de-facto standard for getting proprietary data into enterprise AI systems.

What I got wrong was AI commoditization. Yes, AI became more accessible - but building AI solutions is still largely an expert's game. More tools actually added more complexity: RAG or fine-tune? Which model? Invest or wait? Building AI solutions seems more challenging than ever.

All these developments point to an interesting shift as we head into 2025. But remember: These predictions aren't about being perfectly accurate but about spotting trends early enough so you can act while there's still time.

AI Predictions for 2025

For next year, I see two big trends shaping AI in business:

Prediction 1: The Agentic Letdown

AI agents are currently all the rage and agents are expected to surge in 2025. However, the growing adoption of AI agents won't deliver the productivity gains everyone expects. That might sound pessimistic, but hear me out.

Agents are cool, there's no denying that - from scheduling meetings to writing software, or even collaborate with other agents to solve complex problems. It's pretty impressive stuff that gets a lot of attention these days along with a maturing tech ecosystem.

Why I Think We'll See Disappointment

But agents in 2025 will be what chatbots were in 2024. Everyone wanted one. Few really knew what they were getting themselves into. Agents fall into the same 80% fallacy as every other AI application. They work great in demos and controlled environments, but when you put them into production, things get really messy, really fast.

Organizations with shallow AI experience will treat agents like simple plug-and-play solutions, which is doomed to fail. Building reliable agents requires a deep expertise of

• the specific domain they're working in

• the edge cases they might encounter

• the consequences of potential errors

• the human oversight when things go wrong

After the initial hype dies down, people will realize that building truly useful agents is hard - much harder than current demos suggest.

What This Means For You

Does this mean you should ignore agents altogether? Not at all. Agents will have their place and many projects will succeed. But success will require a more nuanced approach:

1. Start with strong Augmentation
   Instead of letting your "agent swarm" loose from day one, gradually increase the level of automation and integration with appropriate levels of human oversight.

2. Pick Your Battles Wisely
   Choose specific, well-defined tasks where the consequences of failure are manageable. Document processing? Great start. High-stakes financial decisions? Maybe not yet.

3. Plan for Failure
   When designing agentic solutions, think about edge cases and failure modes as well as the happy path. What happens if things go wrong? How will humans step in?

The most successful agent implementations in 2025 won't be the most ambitious ones - they'll be the ones that found the sweet spot between automation, oversight, and business value. Which leads to my next prediction.

Prediction 2: The Year of AI Profit

2025 will be the year where AI needs to prove its worth in cold, hard profit. Let me explain:

This shift has been brewing for a while. 2023 was the year of prototypes, and 2024 saw companies trying to productionize their AI initiatives. But 2025 is when the bill comes due. Companies are starting to realize that scaling AI isn't just about technical challenges, it's about economics. The questions will shift from "can we build this?" to "should we build this?" and "is it worth maintaining?"

Why This Will Matter More Than Ever

Unlike traditional IT projects, AI initiatives come with a twist that many organizations are just beginning to grasp: they get more expensive the more they're used. This is a complete paradigm shift from the typical "zero marginal cost" model of software development.

Here's what's driving this reality:

• AI API costs scale with usage

• AI expert talent is rare and expensive

• Models need regular updates and retraining

• Edge cases keep appearing as usage increases

This means your AI project that looks great in a small pilot might become a cost center at scale. Suddenly, that chatbot handling thousands of customer queries per month isn't just about technical capabilities - it's about dollars and cents.

What This Means For You

To navigate this new reality, you'll need to adapt your approach:

1. Split Your Portfolio 
   Divide your AI initiatives into two categories: R&D projects (where immediate profit isn't the goal) and business projects (which need clear ROI). Be explicit about which is which.

2. Focus on Value Metrics 
   Some AI projects will still get funded for "productivity" reasons. Of course it makes sense to give people access to tools like ChatGPT or Copilot (with the right enablement). But for custom AI solutions? You'll need harder numbers.

3. Set Your Profit Threshold

   Instead of pinpointing the exact "value" of an AI use case, determine whether the problem exceeds your minimum value threshold. Then, evaluate potential solutions based on hard benefits, supported by soft benefits, and use these insights to guide your investment.

4. Think Maintenance First 
   Before building any new AI solution, calculate the full cost of running it at scale. Include everything: API costs, maintenance, monitoring, and human oversight. Then double that number - you'll thank me later.

The winners in 2025 won't be the companies with the most AI projects - they'll be the ones who figured out how to make AI truly profitable while keeping experimentation alive through smart R&D investments.

Bonus: Watch xAI

I don't typically call out single companies in my predictions. But I'll make an exception this time: Keep xAI on your radar in 2025.

Two reasons: First, never bet against Elon Musk (no matter what you think of him). Second, never bet against an AI company that just built the largest AI supercomputer on earth and raised a $6B Series C in record time. While OpenAI is busy pivoting from an AI research to an AI product company, xAI is building AI with a massive existing product already waiting for it.

Whether this will work out remains to be seen. But the combination of compute power, platform effects, and Musk's track record of making seemingly impossible things happen is worth paying attention to.

Conclusion

You'll hear from me in a year how these predictions turned out. So long, I hope you can use them to make better-informed decisions on AI today.

If you're interested in more deep dives on upcoming trends, watch out for the upcoming D2A2 Predictions Report, where Prashanth Southekal and a team of co-authors including myself explore 10 predictions that touch Digital, Data Analytics, and AI leaders.

Stay tuned for that – we'll keep you posted on LinkedIn and in this newsletter.

See you next Friday,
Tobias

PS: I've opened reservations to work directly with me to add $10K+ in recurring profit to your bottom line. You can reserve now and we'll start whenever you're ready next month. Answer "January" and I'll send you the details.