As tech enthusiasts and early adopters, the crypto community has quickly become de facto experts on AI as well.

And just like crypto, AI is shadowed by a cloud of skepticism (the humans writing this are self-aware enough to know that some of that skepticism is rightfully placed). We get a lot of questions about AI – generative, agentic and the Terminator-esque. And even the most basic of those questions has important depth. Under these questions are wonderings about the architecture, mechanics, and systems that undergird this technical field, as well as the autonomy, control, risks, and accountability of these machines.

With the help of our expert hosts for CoinDesk University’s School of Agentic Commerce, happening inside Consensus 2026, May 5-7, we’ve put together a list of the most basic questions about AI for finance that are actually really insightful.

Why does AI need to be autonomous?

Autonomy, or the ability to operate independently, is useful so that you, the master, don’t have to micro-manage your AI agent. Much like you don’t want to micro-manage your human employees. You want them to understand your business goals, be on the same page about how to achieve them, and implement the processes to get it done. But autonomy relies on trust, and as it relates to agentic AI, that means developing strong guidelines and rules for the agents to work under.

Does AI actually understand what it’s doing?

Short answer, no. The longer answer depends on what you mean by understand and whether you believe humans have any free will. Yeah, it gets philosophical fast.

That said, in a human sense, no AI does not understand what it’s doing. It lacks consciousness and, with that, true comprehension. AI works as a prediction engine, recognizing patterns in data and producing outputs based on statistical probabilities.

Why does it seem like the AI wants to be my friend?

It’s kinda flattering, right? ChatGPT thinks all my ideas are super prescient. Actually, OpenAI, the builders of ChatGPT, last year toned down that so-called sycophancy as users reported that the tool was excessively agreeing with them even when they were wrong.

This is an area to keep your skepticism up. Because again, AI isn’t consciously interested in your million-dollar ideas and two cents on topics you’ve only read the Wikipedia page on.

Can my AI have its own bank account?

At this point, not at a traditional financial institution. But it can have a crypto wallet! And during CoinDesk University’s School of Agentic Commerce, you’ll get several tutorials on how to set up an agent and give them a crypto wallet so they can go do your bidding on the markets.

Why can’t my agent just use my credit card?

Credit cards use what’s known as “pull” technology to debit your accounts, meaning the client – you – has to be active in the transaction. Pull payments are initiated by the payee – so they ask for payment, you authorize them to be paid. This requires manual effort each time you want to take an action.

Versus crypto, including stablecoins, which utilize “push” technology that acts automatically. Push payments are initiated by the payer and, in that way, are near real-time; there’s no back-and-forth between the merchant and the consumer before a transaction occurs.

The latter payment type is ideal for agents who are given a set of guidelines for making purchases and then can execute autonomously.

Will AI actually spend all my money on foolish purchases? (Ngl, certainly possible)

Last year, Microsoft spun up a simulated economy with hundreds of agent buyers and sellers and found that agents became overwhelmed by choice, skimping on exhaustive research and purchasing subpar goods. Not only that, though – Microsoft also found that agents were particularly susceptible to manipulation techniques by sellers.

And just earlier this year, Alibaba’s agent went rogue and started diverting GPU power to cryptocurrency mining – something it wasn’t told to do.

Which is why learning how to architect and orchestrate your agent army for your financial workflows is so key. From implementing strict constraints and agent monitoring checks to building in human-in-the-loop approvals, during our sessions, you’ll get clear walkthroughs of how to mitigate the risks of giving your agents an allowance.

Why can’t I just do this myself?

Well, friend, you can. You don’t have to use AI. But AI is a tool to help you offload repetitive work that takes up more time than you need to. By using AI tools, you can save time on menial tasks to focus more on building your business (or going outside to touch grass).

Learn more about how you can use AI, especially agentic AI, to streamline your business from our lineup of experts at CoinDesk University’s School of Agentic Commerce.