Wintermute: 3 New AI x Crypto Directions We're Bullish On
TechFlow Selected TechFlow Selected
Wintermute: 3 New AI x Crypto Directions We're Bullish On
Crypto is no longer built for humans; the machine economy is the next battlefield.
Navigating Web3 tides with focused insightsAuthor: Wintermute
Compiled by: TechFlow
TechFlow Editor's Note: Wintermute releases industry manifesto: The infrastructure battle in crypto is over; the next battlefield is not DeFi but the Machine Economy. When AI agents, warehouse robots, and automated lab systems become economic agents, the underlying assumption of traditional financial rails "there's a human on the other side" will become completely obsolete—while the difference of crypto rails "there's code on the other side" will turn from a defect into a core advantage. Three directions worth watching: Agent Economy Layer, Physical AI, Machine-Driven Discovery.
Old Problems Are Dead, New Problems Are Emerging
Crypto has been around for over a decade. L1s are live, L2s follow closely, DeFi has matured, and stablecoins have become infrastructure. In every sector, from exchanges and lending to perpetual contracts and prediction markets, every category is crowded, and every obvious idea seems to have been done.
So, is there anything left to build in crypto?
Many builders gave up here. They are wrong—not because the answer is "no," but because the question itself is wrong.
For most of crypto's history, the really interesting question was whether the rails could hold up: could they settle within seconds, could they transfer stablecoins at scale, could they run open networks under real load. These questions now have answers. The infrastructure works; the next interesting question is elsewhere.
What's truly changing is everything around the infrastructure. Models can act autonomously rather than just respond; robots learn from human videos rather than relying on hand-written code; open standards for agent payments and identity are taking shape. None of this is crypto, but each is challenging the boundaries of financial and trust infrastructure built for humans.
The question worth asking is no longer "what can crypto do," but rather "what does this world need crypto to do".
The answer is becoming increasingly clear—the Machine Economy.
Machines Are Not Tools, But Economic Agents
When we say "Machine Economy," we don't mean machines as tools—the kind used to send emails or write code. We mean machines as economic agents.
This shift is subtle, but the consequences are huge. Tools wait for instructions; agents hold context, make decisions, execute transactions, and act autonomously in digital and physical worlds. Current models are good enough and cheap enough to do this at scale.
Real-world scenarios:
An agent books a flight for you, negotiates the price, pays the merchant, handles refunds—without your intervention throughout.
A warehouse robot picks up tasks paid per piece, charges itself, pays for its own compute, and routes revenue to the operator.
A research system autonomously designs experiments at night, procures reagents, runs closed-loop—without a graduate student present.
Almost all our existing financial and trust infrastructure assumes there is a person or a company on the other side—an entity you can identify and hold accountable. This assumption fails the moment the counterpart is an autonomous agent, and our existing payment, identity, authorization, dispute, and settlement rails were never built for this scenario.
And this sits exactly at the intersection of crypto, fintech, AI, robotics, and quantum computing.
Why Now
Three recent shifts that seemed unlikely just a few years ago.
Models Can Act, Not Just Answer
Models no longer just answer questions; they can act autonomously, and at a low enough cost to run unattended. The unit cost of digital work is collapsing, making tasks that weren't worth a human's time feasible—and at scales and amounts existing systems were never designed to handle.
Open Standards Are Maturing
Stablecoins are now real settlement rails. Protocols like x402, MPP, and AP2 give agents payment methods. Faster blockchain networks and faster fiat networks are converging in the middle. Open vision-language-action models let robots learn from human videos and simulations rather than relying on custom programming. Standards let builders compose rather than rebuild—this is exactly what drives accelerated progress in every category.
Agents Can Run Continuously
Unlike the tools we are used to—those adapted for narrow, guided use cases—agents hold context and work unattended over long periods. This changes the economics of automation, and also the volume of activity any system must absorb.
None of these constitute an argument on their own. But together, they do.
Crypto Isn't Dead—It Changed Battlefields
When most crypto founders ask "what else can be built," they overlook one thing:
The next wave of interesting companies won't be crypto vs AI or crypto vs robotics. The founders we are most bullish on are not choosing between these technologies, but stacking them.
You are no longer just building in crypto. You are building crypto + AI, crypto + robotics, crypto + autonomous science.
Traditional financial rails are built around human accountability: identities you can verify, intents you can dispute, people to hold accountable when things go wrong. Crypto rails are built around something different: code you can audit, on-chain records anyone can read, rules enforced by the network.
When the counterpart is autonomous, this difference is no longer a defect—it becomes critical. As machine-driven activity volume grows, the rails built by crypto fit this demand better than those designed for humans: open, programmable, permissionless, second-level settlement, identity without intermediaries.
The opportunity for crypto builders is not to compete with crypto builders from the last cycle, but to become the underlying substrate for the next wave of AI, robotics, and physical autonomy.
And the largest platforms are already sprinting. Coinbase, Robinhood, and Binance have each launched agent trading infrastructure in the past few months: agent-operated wallets, autonomous execution—Robinhood even built a new chain specifically for this. This is no longer a niche crypto conversation; it is happening on one of the largest retail user platforms in the world.
Today's Failure Modes
The bet above is: permissionless, programmable rails are better suited for autonomous agents than rails built for humans. This bet has not yet been proven at scale, and two failure modes illustrate why more work is needed:
Security: Agent Wallets Have Become Attack Surfaces
In May 2026, an attacker used Morse code-style prompt injection to make Grok output a transfer instruction, which an automated trading agent then executed on-chain—transferring approximately $150,000 to $200,000, most of which was subsequently recovered (SlowMist).
Liability: Who Bears the Consequences of AI System Failures
Even if AI, human reviewers, and governance votes all sign off, who bears responsibility when a system involving AI fails remains unresolved. In February 2026, an oracle bug in AI-assisted smart contract code on Moonwell led to a $1.78 million bad debt incident—no link in the review chain caught it (rekt.news).
Three Directions Wintermute Is Bullish On
Most activity currently focuses on the component level: foundation models, robotics hardware, stablecoins, exchanges. These markets are crowded and well-funded; the opportunity is not there.
The opportunity lies in what connects them—rails for transactions, coordination, and trust between machines that do not yet exist. Three directions stand out:
Agent Economy Layer
The hard part is not whether agents can pay, but: who holds authority when an agent goes wrong? Who bears fraud risk? How does all this reach merchants without requiring them to rebuild checkout flows?
The form of agent commerce is still being written: authorization layers, agent identity, neutral routing between rails, markets where agents buy their own compute/data/access. Better teams here charge fees for authorization and risk reduction rather than taking a cut of payment value—this makes the business viable before agent scale truly arrives.
Physical AI
Robots are gaining capabilities much faster than they are gaining economic volume. A model can now generalize across tasks and across different robot bodies; non-engineers can redirect it just by telling the robot what to do. But robots still cannot pay for their own compute, charging, or maintenance, nor can they get paid for the work they do.
What's missing is not hands, but wallets. We focus more on structured scenarios—warehouses, logistics, and retail backends—where the economics are already viable and real deployments exist, rather than humanoid home robots.
Machine-Driven Discovery
Lab orchestration, automated experiment design, software that closes the loop between hypotheses and results. Founders building autonomous layers for science are already selling to materials and drug discovery labs. Quantum is a wildcard next to this direction: simulation and sensing could step-change what is discoverable, and post-quantum security is already a real need at the settlement layer. Hard to underwrite, winners unclear—but there is something here.
R[3]sidency × Construct Accelerator
The infrastructure needed for the Machine Economy does not yet exist. This is where the work is, and where Wintermute is looking.
They want to support founders who see new problems emerging between financial rails, autonomy, and trust—founders who can deliver products on existing rails while remaining adaptable as standards evolve.
This is the purpose of R[3]sidency × Construct: 8 teams, $300,000 each, 12 weeks London residency, 30+ mentors, London and New York Demo Day. Run jointly with top partners: Fabric Ventures, Solana, and Coinbase.
If you are building for a world where machines and humans trade and operate in parallel—they want to support you.
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
@wintermute_t
