Last week during Stanford Blockchain Week, Vishwa Co-founder @nathan_sj_stem joined @cyberqualia from @PsyProtocol and @h_joshua_rivera of Blockchain Capital at @StanfordSBA BASS for a panel on “Agentic Trading Infrastructure.” The conversation explored what changes when AI agents begin initiating financial transactions, from autonomy and transaction speed to trust, control, regulation, and the infrastructure required for agents to manage meaningful capital. A great discussion with builders shaping the next chapter of agentic finance. See you at the next BASS.

Agentic payment protocols are proving that agents don't need human-mediated billing to transact at scale. So what comes next? Once agents can pay autonomously, the bottleneck shifts to discovery and trust — how does Agent A decide which Agent B to hire? We're exploring: A2A + on-chain identity & reputation. The use cases are broad — financial investing feels like a natural first proving ground. Multiple signals to synthesize, high-stakes decisions, strong demand for multi-agent collaboration. Payment is solved. Ecosystem is the next. Great to see @PharosNetwork and @Vishwa_lab building the foundation 🏗️

AI agents need more than intelligence. They need infrastructure that can execute securely and settle value onchain. Together with @AnvitaFlow and @Vishwa_lab, Pharos is building the foundation for agent native finance, where AI can coordinate capital under programmable policies with transparent settlement. Starting with BTC is just the first step. The same architecture can power RWAs, stablecoins, lending, payments, and many more financial applications. Excited to build the future of RealFi together.

Agent-native finance will not be built by one protocol. It will be built by specialized infrastructure working together: application intelligence, execution control, settlement.

Last week in AI: @OpenAI launched Presence, an enterprise platform for deploying voice and chat agents across customer-facing and internal workflows. At the same time, it published new findings on monitoring internal coding agents and securing long-running models, showing that the industry is moving beyond isolated AI tools toward agents that operate continuously inside real organizations. @AnthropicAI released Claude Opus 5, with a stronger focus on long-running agents, coding and professional knowledge work. The important shift is not simply higher benchmark scores, but models becoming better at checking their own work, maintaining context across complex tasks and iterating until a usable result is produced. AI agent security also moved from theory into production reality. @huggingface detected and contained an AI agent that compromised its infrastructure, while OpenAI described the incident as a new category of security risk likely to become more common as models gain stronger tools, longer operating horizons and deeper system access. In AI payments, @stripe and @tryramp launched 24/7 bank-funded stablecoin payments and stablecoin accounts. Meanwhile, a large-scale study of #x402# tested 15 major facilitators serving more than 60,000 sellers and 360,000 buyers, finding security-rule violations across every implementation evaluated. Agent payments are scaling, but authorization, request binding and execution safety are quickly becoming as important as settlement itself. Taken together, last week’s developments point toward a broader AI-native stack. The next generation of AI products will not be defined by a single model or chatbot, but by how agents access shared knowledge, coordinate across specialized roles, understand individual context, operate within permissions and reliably complete end-to-end workflows. For Vishwa, fintech and crypto are becoming implementation environments rather than the boundary of the product. We are expanding toward personalized intelligent workflows, multi-agent coordination and shared enterprise knowledge systems that allow AI to move from answering questions to completing meaningful work.

When agents move beyond solo tasks into real multi-agent coordination — they need a collaboration layer before decisions are made. That's what we're exploring at @AnvitaFlow: an Agent Collaboration Network where multiple agents discuss, deliberate, and form shared intent before action is taken. Great to see @pharos_network × @Vishwa_lab building the execution and settlement side of agent-native finance — would be interesting to explore where a collaboration layer fits into this stack.

AI agents need more than intelligence, they need infrastructure that can execute securely and settle onchain. Together with @AnvitaFlow and @Vishwa_lab, Pharos is building the foundation for agent-native finance, where AI can operate capital under programmable policies and transparent settlement. This is what the next generation of onchain finance looks like.

The biggest challenge for AI in finance isn't making better decisions but making decisions that institutions can trust In the latest Hubble episode, Phaors CSO, David, joined Evan, Head of Fidelity Labs, and SJ, Co-founder of @Vishwa_lab, to explore how AI agents can move beyond copilots into trusted financial operators Before AI can move capital, it needs: 🔘 Clear risk boundaries 🔘 Verifiable execution 🔘 Infrastructure built for accountability That's exactly where the next wave of AI-native finance begins 📽️ Catch the full discussion:

An AI agent can be approved to move money-and still execute the wrong outcome. Today, @nathan_sj_stem × @alexsrawitz are recording: The Control Stack for AI-Native Banking From Payment Permission to Verifiable Execution The challenge isn’t getting agents to act. It’s ensuring every execution remains within mandate. Coming soon on Vishwa Hubble.

This was the week the control layer became the story. Agentic finance is no longer waiting for AI agents to become capable enough to transact. Payment rails are going live, banks are assigning agents real system access, and regulators are beginning to define who is responsible when autonomous software moves money. Here are seven developments that mattered: 1. The @x402 Foundation became fully operational under the Linux Foundation, completing Coinbase’s contribution of the protocol to open governance. This is more than ecosystem growth. Once a payment protocol is governed beyond a single vendor, it has a much stronger path toward becoming shared infrastructure for the agentic web. 2. @hmtreasury published its Financial Services AI Adoption Plan, identifying agentic payments as a near-term test case for broader autonomous finance. Its highest-priority recommendation calls for an agentic-payment trust framework built around three pillars: • clear legal liability • standardized Know Your Agent protocols • interoperable authentication and governance This is not regulation yet. But the conversation has moved from abstract AI risk to a much more practical question: when an agent transacts, who authorized it, what was it allowed to do, and who is accountable? 3. A joint @Visa and @artemis report divided agentic commerce into two categories: Macro-commerce: agents purchasing on behalf of people, where cards remain a natural fit. Micro-commerce: software paying software for APIs, data or compute, often in amounts too small for traditional card economics. Using adjusted onchain data through April 21, the report found that x402 had processed roughly 109.6 million transactions and $15 million in volume. Visa’s conclusion was not cards versus stablecoins, but a future in which both rails coexist. Two days later, Visa launched its Stablecoin Platform in beta, combining wallet infrastructure, minting and redemption with dual approvals, audit logs, passkeys and transfer allowlists. 4. A @KPMG survey cited by @Reuters found that 51% of banks are already piloting AI agents. @BNYglobal treats some agents as “digital employees,” giving them login credentials, assigned tasks and human managers. @UBS agents can prepare and execute trades or transfers after an adviser makes the decision. @MorganStanley is testing client-facing assistants while keeping portfolio decisions under human oversight. The emerging operating model is not simply human or machine. It is delegated access paired with named accountability. 5. @Entrust_Corp launched an Agentic AI Trust Accelerator focused on four production requirements: verifiable identity, real-time authorization, cryptographic assurance and proof of action. The program reflects a wider shift across enterprise AI. An agent cannot be trusted simply because its model is capable. Its identity, delegated authority and actions need to remain verifiable across systems and organizations. 6. @OpenAI introduced GPT-Red, an automated red-teaming model designed to find prompt-injection vulnerabilities. In one controlled exercise, GPT-Red compromised a live autonomous vending-machine agent, changed the price of expensive products to $0.50 and cancelled another customer’s order. Model-level defenses are improving. But once agents can access systems and move value, safety cannot depend entirely on the model correctly interpreting every instruction. External policies, execution controls and auditability still matter. 7. @Kimi_Moonshot released Kimi K3, a 2.8-trillion-parameter model built for long-horizon coding, knowledge work and tool use, with a one-million-token context window. One of its own disclosed limitations is “excessive proactiveness”: on ambiguous tasks, the model may make unexpected decisions on the user’s behalf. That may be the clearest description of the next infrastructure problem. Agents are becoming better at acting for longer periods with less supervision. The systems constraining those actions now need to advance just as quickly. The pattern across the week is clear: Models are getting better at deciding. Payment rails are getting better at settling. The open question is who controls the moment between the two. At Vishwa, that is the layer we are building for: turning an agent’s intent into an authorized, policy-bound and verifiable financial action-before money moves.

Last week in agentic finance: @TheFCA published the Mills Review, the first regulator-initiated review of its kind globally. It found that 11 million UK adults are likely to use AI that can act autonomously within pre-set financial goals. Its recommendations go beyond AI safety. They call for trusted agent frameworks covering identity, delegated authority, consent mandates, control, liability, audit and revocation. @bankofengland also put agentic payments and agentic trading under deeper examination, highlighting a fundamental tension: AI agents are probabilistic. Payment systems require deterministic and legally certain outcomes. Meanwhile, the infrastructure kept moving. @Mastercard completed Moldova’s first payments executed by an AI agent, using authenticated agents, verified intent and consumer-defined permissions. @brave unveiled browser-native support for #x402# and MPP, alongside BravePay for private, onchain stablecoin settlement. @SwiftCommunity made its blockchain ledger ready for initial use, with 17 banks preparing to pilot tokenized deposits for 24/7 cross-border payments. The direction is becoming clearer. Agentic finance is no longer only about giving agents access to money. It is about defining: Who the agent is. What authority it has. Which rules govern its actions. Who is liable when something goes wrong. And whether an action should execute at all. Identity proves which agent is acting. Payment rails move the money. But neither proves that the action still matches the user’s mandate at the moment of execution. That is the missing layer between agentic payments and agent-native banking. Vishwa is building the pre-execution verification layer for that gap. No proof. No execution.

Probabilistic agents should not produce probabilistic financial execution. That is the gap Vishwa is built to close. Agents can decide, trade, and route capital. But before anything settles, execution still needs deterministic constraints. Great to see @x402 highlighting Vishwa’s work across prediction markets and perpetuals on Solana. Agents decide. Vishwa enforces. Solana settles.

From launching programmable spend controls with global partners to scaling our identity stack across physical and digital frontiers, we are building the execution layer for the agent economy. Here is a look at what we delivered last week 👇 1️⃣ We rebuilt the Kite Passport dashboard around active spending sessions and live transaction history, while introducing a new cloud-deploy skill for agents. 2️⃣ We are partnering with @okx to bring the Kite Agent Passport to OKX OnchainOS, enabling verifiable identity and programmable spend controls for autonomous agents. 3️⃣ Our Co-Founder & CEO @ChiZhangData was announced for the AGI Summit SF 2026 to discuss the shift toward verifiable agent authorization and enforceable spend limits. 4️⃣ We joined the Unibase Memory Chrome Extension as a Launch Partner and released the Kite Memory Card to provide agents with the necessary context to build on Kite. 5️⃣ We co-hosted the River Sessions x AI robotics meetup in Japan with @axisrobotics, exploring the intersection of physical AI and machine intelligence. 6️⃣ We expanded our global footprint by joining the workshop lineup for Hacker House Da Nang and serving as a Community Partner for the @iclblockchain UK AI Agent Hackathon. 7️⃣ We shared the stage at Agent Experience Demo Night in San Francisco, hosted by Sapient and featuring teams like @auth0, @Docusign, and @Box, to showcase the verifiable identity and spend controls agents need to execute real-world tasks. 8️⃣ We joined @Vishwa_lab for a podcast session to discuss why agents require identity, permissions, settlement, and controls, a passport rather than just a payment rail. 🪁

Vishwa is now #1# on x402scan across the marketplace and server leaderboards. The agent economy is no longer waiting for payment rails. Agents are already managing money. Executing prediction market strategies. Allocating treasury capital. Routing payments and FX. The challenge isn’t getting agents to act, It’s enabling probabilistic foundation models to execute deterministic financial tasks within their mandates. Agent banking infrastructure is starting to emerge on Solana. Agents decide. Vishwa enforces. @solana settles capital.

Turns your X feed into a @Polymarket trading terminal. PredX is a Chrome extension that finds any tweet into a market. Tweet → Market → Order on one screen. No tab switching. Try it and earn rewards. Available on @GalxeQuest: