Earlier this year, Linas Beliunas asked Claude to write the code for a complete digital bank, something close to a Revolut replacement. The result was ambitious. It even had a confetti animation when the balance went up.

It also did not work. He shared the result with his followers anyway, with the humour intact and the lesson clear: a convincing AI demo is a long way from a working product.

On a subject where every week brings a claim that AI is about to replace an entire industry, a post from someone who actually tries it stands out. That habit of testing the claim, and saying plainly what he found, is what makes Beliunas worth following.

Who is Linas Beliunas?

Based in Vilnius, Lithuania, Beliunas writes Linas’s Newsletter, a daily briefing on the point where finance meets technology. Each day he picks three stories, adds things worth watching and the biggest money movements, and moves on. The newsletter has more than 156,000 subscribers on Substack, and he has more than 700,000 followers on LinkedIn.

His lens is AI and fintech, and he is fascinating on payments, risk, stablecoins and AI agents. He is hilarious on holding hype up to scrutiny. Treat him as a specialist guide to how AI is changing money and business, not a general encyclopaedia of the technology.

His credibility comes from working inside the industry he covers. He studied economics and started in financial services early, with roles at Western Union and Danske Bank among others. He moved into payments and fintech, leading Flutterwave’s expansion in Europe as country manager and running Flutterwave Lithuania, and later working at the crypto infrastructure company Zero Hash. He has also advised founders as a venture advisor at EQT Ventures.

In August 2025 Beliunas joined Oscilar, a startup building AI agents for fraud and risk decisions, as Head of Content Strategy. His reasoning at the time was simple: fraud keeps evolving, and AI that can read context and adapt in real time is better suited to it than fixed rules.

He is open about his professional work, and it gives his analysis an insider’s feel for what it takes to make the technology work. He is also funny about it. The confetti-covered bank shows a writer who can laugh at AI’s failures while taking its potential seriously.

How to tell AI hype from reality

The bank test is typical of how he works: look at the claim, then look at the evidence. The same approach shows up when he handles a big announcement.

In July, Moonshot AI released Kimi K3, a model with 2.8 trillion parameters and the largest open model of its kind. Beliunas called it wild and said it nearly matched the best US models. But the same post carried the caveats. The full weights were not yet public. Several benchmarks came from the vendor itself. One independent evaluation had found a hallucination rate of around 51%. He then explained why it mattered if Moonshot delivered: thousands of developers could fine-tune, audit and run a near-frontier model themselves. The weights did arrive about ten days later.

That pairing of excitement and doubt in the same post is rare. Most AI commentary offers one or the other.

He brings the same scepticism to big corporate claims. When Block cut 40% of its staff in February and CEO Jack Dorsey pointed to AI, Beli?nas agreed that AI had played a part but argued it was probably not the core reason. He asked whether this looked more like a correction after years of cheap money, and noted that plugging in AI is an easy excuse for companies to reach for. The real impact of AI, he said, is still unknown.

He is equally wary of the bubble question. When the investor Howard Marks published an 18-page memo asking whether AI is a bubble, Beliunas took from it that two things can be true at once: AI could be the opportunity of the century, and the excitement around it could be a bubble. You stay in the game because the technology is real, he argued, and you stay sane by remembering how much of the noise is hype.

Does the scepticism hold up? In the two cases above, the evidence so far is on his side. After the Block announcement, the Wall Street Journal and the Guardian both ran pieces questioning whether AI was the real reason for the cuts. And Moonshot released the Kimi K3 weights about ten days after his post, as the company had promised. That is a small sample, not a track record, but it shows his caution is aimed at the right places.

Why small teams can win with AI

Scepticism is only half of his appeal. He is also a builder’s reader of the news, and he usually ends up asking what would you do next?

A good example is his reading of Anthropic’s analysis of one million Claude conversations, which found that around 6% were people asking for personal guidance on things like health and money. Beli?nas treated this as a map of demand. People are already bringing these problems to AI, so the usage data shows where products are wanted. He pointed to Cal AI as proof that a narrow starting point can be enough. The calorie-tracking app began with one small problem, reached millions of downloads and tens of millions of dollars in annual revenue, and was bought by MyFitnessPal in March with a team of just seven.

Follow his posts for a few weeks and you understand that momentum is what builds giants. Companies that start narrow, ship early and keep improving tend to outrun those still polishing the perfect version, and AI makes that gap wider because the tools let small teams move so quickly.

In October he turned to Stripe CEO Patrick Collison’s argument that AI agents could change what it takes to win in business. For decades companies have spent billions competing for human attention. If agents start making purchasing decisions for us, better products could matter more than better marketing. Beli?nas thought Collison was mostly right, with a catch worth remembering that agents have biases too, and whoever controls the agent could control which businesses win. He followed it with a 14-point playbook for founders and investors.

How to sharpen your understanding of AI

The best way to use his work is as a regular briefing that trains your judgment.

Start with the newsletter. Three stories a day is easy to keep up with, and the value builds over weeks as you see which themes keep returning: agents, open models, AI in finance, the economics of building with these tools.

Read for the caveat. In his posts, look for the sentence that limits the claim. What is vendor-reported and what is independently tested? What is announced and what is actually released? That habit transfers to every other AI headline you read.

Use his guides as maps. His longer pieces include The Top 70 AI Agents You Must Know, The Agentic AI Survival Guide, Agents 20 on the top AI agent startups, and AI 100 on the top AI startups. Several sit behind the paid tier, but they are useful if you need to get oriented quickly in a crowded field.

Join the conversation on LinkedIn and Substack Notes. The quick takes appear there first, and he replies to challenges. In the Block thread, when a reader made a fair point, he said so.

Go to the original source when it matters. If a story affects a decision you are about to make, use his summary as the starting point and check the underlying material yourself.

Linas Beliunas will not tell you that AI is a fad, or that it is about to solve everything. He asks what is real, what is only claimed and what is still missing, and then tells you what he would do next.

For anyone trying to deploy AI rather than just talk about it, that is a rare and valuable expert voice.

Interested in the series, The Influencers? You might also like this…