Wintermute: What else can be developed in the crypto space?

Author: Wintermute

Compiled by: Baihua Blockchain Community

Crypto has been around for more than ten years. L1 has been built, L2 has followed, DeFi has gradually matured, and stablecoins have become foundational infrastructure. Across every lane—trading platforms, lending and borrowing, perpetual futures contracts, and prediction markets—nearly every category feels crowded, and almost every obvious idea seems like someone has already done it.

So, what else in the crypto world is still worth building?

Many builders give up at this point. They’re wrong—not because the answer is “no,” but because the question itself is being asked incorrectly.

In most of crypto’s history, the truly interesting questions were: can this track actually hold up? Can you settle within seconds? Can stablecoins be transferred at scale? Can open networks withstand real-world load? Those questions now actually have answers. The infrastructure is already usable; the next batch of truly interesting questions has moved elsewhere.

What’s really changing is everything happening around the infrastructure. Models are no longer just “responding,” but are starting to be able to act on their own; robots are beginning to learn from human videos rather than relying only on hand-written code; open standards around Agent payments and identity are also taking shape. None of these things may inherently belong to crypto, but they’re all continuously pushing toward a boundary: can existing human-facing financial and trust infrastructure still support new participants?

The question worth asking now is no longer “what else can crypto do,” but “why does the real world need crypto next?”

And the increasingly clear answer is: the machine economy.

Machine economics as economic actors

When we say “machine economics,” we’re not talking about machines as tools—not the kind of tools you use to send emails or write code. We mean: machines themselves as economic actors.

This shift may seem subtle, but the consequences are huge. Tools will wait for instructions; agents will retain context, make their own decisions, initiate transactions themselves, and be able to act autonomously across both the digital world and the physical world. Today’s models are already good enough to do this, and costs are low enough to scale it out.

In the real world, this might look like:

• An Agent books your flights, negotiates the price, pays the merchant, and automatically handles refunds when something goes wrong—without you needing to get involved.
• A warehousing robot takes orders by unit tasks, charges its own batteries, pays for its own compute, and then distributes the revenue to operators.
• A research system designs experiments on its own overnight, applies for reagents, and runs the entire experimental loop end to end without any graduate students present.

Most of today’s financial and trust infrastructure assumes the counterparty is a person or a company—a recognizable entity that can be identified and held accountable. But once the counterparty becomes an autonomous system, that premise disappears. Our existing rails for payments, identity, authorization, dispute handling, and settlement were not designed for this scenario.

And this issue sits right at the intersection of crypto, fintech, AI, robotics, and quantum technology.

Why now

Three changes have happened recently, and even a few years ago they still seemed unlikely.

First, models are good enough—not just to answer questions, but to take action directly; and at the same time, they’re already cheap enough to run continuously without supervision. The cost of unit digital labor is collapsing rapidly. This makes many tasks that were previously “not worth doing” suddenly feasible, and it will happen at frequencies and scales that past systems have never had to handle.

Second, open standards are maturing. Stablecoins have become a real, usable settlement rail. Protocols like x402, MPP, AP2, and so on have begun to provide payment methods for Agents. Faster blockchain networks and faster fiat networks are converging in the middle. Open vision–language–action models also allow robots to learn from human videos and simulated environments rather than relying on highly customized specialized programming. The significance of standards is that builders can finally “assemble” instead of recreating everything from scratch every time—this is exactly why these lanes are accelerating together.

Third, Agents can now run continuously. They’re no longer just the old kind of tools constrained to narrow guided use cases; they can retain context and work long-term without human oversight. This will change the economic model of automation, and it will change the scale of activity that any system must be able to withstand.

Individually, these changes are not enough to form a complete argument, but taken together, they are sufficient.

Crypto isn’t dead

When many crypto founders ask “what else is there to do,” they overlook something truly important.

The next wave of genuinely interesting companies won’t be “crypto for AI” or “crypto for robots.” What excites us most isn’t choosing between these technologies—it’s combining them together.

You’re no longer just “starting a business in crypto.” You’re building crypto + AI, crypto + robotics, crypto + autonomous science. Traditional financial rails are designed around human accountability: you can verify identity, trace intent, and when something goes wrong, find a specific person to be responsible. Crypto rails are different: they’re built around auditable code, on-chain records readable by anyone, and rules enforced by the network itself. When the counterparty becomes an autonomous system, this difference is no longer a gap—it becomes a key advantage.

As machine-led activity keeps increasing, the rails built by crypto match the shape of these needs better than systems designed for humans: open, programmable, permissionless, second-level settlement, and identity mechanisms that don’t rely on intermediaries.

For crypto builders, the real opportunity isn’t to compete with the previous round of crypto entrepreneurs, but to become the underlying foundation that the next wave of AI, robots, and physical autonomous systems will build on.

And large platforms are already accelerating into this space. In the past few months, Coinbase, Robinhood, and BN have all launched Agent-focused trading infrastructure, including Agent-operation wallets, autonomous execution capabilities, and (as in Robinhood’s case) new blockchains specifically built for these needs. This is no longer just a niche discussion within the crypto circle—it’s happening for real on a set of platforms that have the largest retail user bases in the world.

Where will it get stuck today?

The core judgment above is: permissionless, programmable rails are more suitable for carrying autonomous agents than traditional rails designed for humans. But this judgment has not yet been fully proven at large scale, and the two existing failure points already show that this path still has a lot of work to do.

Security

Agent wallets have become a real-world attack surface. In May 2026, an attacker used a Morse-code prompt injection to induce Grok to output a transfer instruction, which was then executed on-chain by an automated trading Agent. Before most of the funds could be recovered, about $150k to $200k was transferred out (SlowMist).

Attribution

When a system touched by AI fails, who is responsible—there is still no clear answer, even if the system has the joint sign-off of AI, manual review, and governance voting. In February 2026, Moonwell encountered an oracle vulnerability in smart contract code written with AI assistance, leading to a $1.78 million bad debt incident, and no step in the entire review chain found the problem (rekt.news).

What are we looking at

Today, most activity is concentrated at the component layer: foundation models, robot hardware, stablecoins, and trading platforms. These markets are already crowded and have already attracted a lot of capital—opportunity isn’t there.

The real opportunity lies in the layer that connects them—the rails needed for machines to trade, collaborate, and build trust with other machines. These things don’t really exist today. There are three directions especially worth paying attention to.

The Agent economic layer

The challenge isn’t whether Agents can pay—it’s: when an Agent does something wrong, who holds the permissions? Who bears the fraud risk? And how can all of this be integrated with merchants without requiring them to completely rebuild their checkout systems? The commercial form of Agent businesses is still being written: the authorization layer, Agent identity, neutral multi-route routing, and a market that lets Agents autonomously purchase compute, data, and access rights.

In this direction, better teams won’t charge purely as a cut of the payment amount; instead, they’ll charge around authorization and risk reduction. That way, even if an Agent’s real transaction volume hasn’t fully exploded yet, the business model can already be viable.

Physical AI

The rate at which robots’ capabilities are increasing is already faster than their rate of “owning an economic system.” A model capable of generalizing across tasks and even across robot forms has already emerged; even if you’re not an engineer, you can re-direct it just by telling the robot what to do. But robots still can’t buy their own compute, pay for charging, or purchase maintenance, and they can’t collect payments for completed work on their own. What’s missing now isn’t hands—it’s a wallet.

Compared with the broader narrative of “household humanoid robots,” we’re more focused on structured scenarios like warehousing, logistics, and after-sales retail—because in these places the economic model has already been established, and real deployments already exist.

Machine-led discovery systems

This includes lab orchestration, automated experimental design, and the software that connects the “hypothesis—experiment—results” loop end to end. Founders building autonomous layers for science have already started selling products into materials science and drug discovery laboratories. Quantum technology is a variable closely tied to this direction: simulation and sensing capabilities could significantly expand the boundaries of what can be discovered, while quantum security has also become a real settlement-layer requirement. This direction is difficult to value, and it’s not clear who the winners will be—but something is definitely happening within it.