Amrit Kang is a powerhouse in the transatlantic fintech corridor, currently serving as Vice President at London & Partners. With over 15 years of experience and a track record of supporting more than 200 tech companies in their international scaling efforts, she bridges the gap between the world’s most influential financial hubs, London and New York. As a recognized leader and former entrepreneur with a background in criminal defense law, she offers a unique perspective on the structural shifts currently reshaping how money moves. From the halls of major banks like Citigroup to the high-energy product launches at Stripe Sessions, she has been at the forefront of the transition to an agentic economy where machines are becoming the primary drivers of commerce.
The following discussion explores the rapid evolution of financial infrastructure to accommodate autonomous agents. We delve into the shift from user-led interfaces to API-driven logic, the rise of “agentic commerce,” and the specific ways incumbents and startups are building the tech stack of the future. The conversation highlights the transition of AI from a simple copilot to a fully realized economic participant.
As payments shift from user-initiated actions to machine-executed intent, how must the underlying infrastructure change? What specific steps should developers take to ensure secure delegated spending through agent wallets, and how can these systems handle real-time risk monitoring without human intervention?
The shift toward machine-executed intent is a total structural pivot that requires us to treat AI agents as first-class citizens in the financial ecosystem. To handle this, the infrastructure is moving away from human-centric interfaces toward deep, machine-native layers, as evidenced by Stripe’s 280+ product launches focused on AI-native commerce. Developers must prioritize the creation of “agent wallets” that use Link to enable delegated spend, ensuring that an agent can act within strictly defined financial boundaries. This necessitates a move toward tokenized payment flows that allow for secure delegation of authority without exposing the underlying credentials. Real-time risk monitoring is then embedded into the transaction logic itself, using continuous data loops to detect anomalies instantly, which is vital when you are managing thousands of microtransactions that a human simply couldn’t oversee.
Major financial institutions are now deploying internal agent platforms for complex tasks like risk simulations and portfolio analysis. What metrics determine the success of these deployments, and what are the primary trade-offs when transitioning from human-led workflows to AI-driven drafting of credit memos?
We are seeing heavyweights like Citigroup launch platforms like “Arc,” which empower employees to deploy agents for deep-tier tasks like risk simulations. The success of these deployments is measured by the speed of execution and the accuracy of the outputs compared to traditional manual models. When moving to AI-driven drafting of credit memos or closing books, the primary trade-off is the gain in operational efficiency versus the initial loss of nuanced human judgment. However, the goal is for these agents to handle the heavy lifting of data synthesis, allowing human experts to focus on final high-level strategy. It’s a transition from “doing the work” to “auditing the agent’s work,” which significantly increases the volume of analysis a firm can handle at any given time.
With payment networks now treating machines as first-class users, what specific technical hurdles exist for tokenizing autonomous transaction flows? How does an API-first approach to programmable money movement differ from traditional banking, and what anecdotes can you share regarding the integration of autonomous procurement tools?
The main technical hurdle lies in ensuring that tokenization can handle the sheer scale and velocity of machine-to-machine transactions. Traditional banking was built on “pull and push” mechanics initiated by people, but an API-first approach, like what we see with Increase or SoFi’s hybrid banking tools, enables programmable money movement where the transaction state is machine-readable. This means the money “knows” its own rules for when and where it can move based on pre-set logic. We are already seeing this in action with autonomous procurement tools that integrate directly with platforms like Stripe; these tools can identify a supply shortage, find a vendor, and execute a purchase via an API without a single person clicking “buy.” It transforms the bank from a vault into a dynamic, programmable engine.
Agentic commerce allows AI to discover products and execute transactions independently. What does the ideal tech stack look like for a startup enabling “agent checkout” experiences, and how do you balance autonomous optimization with the need for human oversight in large-scale B2B microtransactions?
The ideal tech stack for agentic commerce must be built on four pillars: real-time payments, robust tokenization, comprehensive APIs, and continuous risk monitoring. Startups in this space are moving away from traditional SaaS “dashboards” and instead embedding intelligence directly into the economic layer, often partnering with giants like Google to place payments inside environments like Gemini. To balance autonomy with oversight, especially in B2B microtransactions, developers are implementing “guardrails” within the code—rules that allow agents to optimize outcomes autonomously up to a certain financial threshold or risk level. This ensures that while the machine handles the 99% of high-frequency trades or purchases, human intervention is only triggered by significant outliers or high-value exceptions.
Personalized financial advice is scaling through AI-native tools designed to manage wealth at a granular level. How does this shift affect the traditional relationship between advisors and clients, and what step-by-step process ensures that automated wealth management remains compliant with evolving global financial regulations?
Traditional wealth management is being disrupted by tools like Citi Sky, which are designed to provide the kind of hyper-personalized advice that was previously only available to the ultra-wealthy. This shift changes the advisor’s role from a primary information source to a high-level relationship strategist, as the AI manages the granular, day-to-day portfolio adjustments. To remain compliant, firms follow a rigorous process: first, they bake global regulatory requirements directly into the agent’s logic; second, they use specialized agents from providers like Anthropic to assist compliance teams in real-time auditing; and finally, they maintain a “human-in-the-loop” system for final approvals of complex wealth strategies. This ensures that even as we scale to millions of users, every action remains within the legal frameworks of both New York and London.
What is your forecast for agentic finance?
My forecast is that we are moving toward a completely machine-native financial world where the distinction between “fintech” and “AI” effectively disappears. In the very near future, the vast majority of financial transactions will be initiated and settled between autonomous agents, with human interaction reserved only for setting high-level intent and goals. We will see a massive explosion in micro-transactions and B2B flows that were previously impossible due to human friction, and the companies that win will be those that provide the most secure, “agent-friendly” infrastructure. Payments will become invisible background tasks, and we will look back at the era of manual “user-initiated” payments as a slow, archaic stage of economic history.