A payment instruction carries responsibility for somebody else's money. At Agentic Finance Hack, scheduled for December 5, 2026 in New York, builders will have a day to explore how much of that responsibility software can handle. The in-person event spans trading, payments, financial crime and finance operations, with approval required to attend.
tokens& is hosting alongside Alessandro Amenta and Jacopo Piazza. Listed speakers include Cognee's Head of Forward Deployed Engineering, Veljko Kovac, and Monk co-founder Joe Zhou. The mix gives the gathering a practical point of interest: people building financial agents will be able to examine the distance between a successful demonstration and a workflow another operator can trust.
That distance is becoming a commercial question. Visa's recently announced invoice-payment pilot offers a current example of agents operating within approved financial transactions. For founders and finance practitioners considering the December gathering, the useful question is how a prototype makes its authority visible to the people who will have to answer for its actions.
What Agentic Finance Hack asks builders to explore
The organizer's project themes cover markets and trading, agentic payments, compliance and financial crime, and finance operations. That last category includes invoicing, collections and treasury. Teams of 2-4 are invited, while solo participants can look for teammates through the organizer's Discord before the event. Space, compute credits and food are offered; specific credit amounts and providers are not listed.
Each theme brings a different version of the same engineering problem. A research agent can produce a recommendation, while an agent with payment permissions can change a financial position. Giving a system that second kind of authority makes its limits part of the product. A useful demonstration might show the instruction, the permitted action and the record that allows another person to reconstruct what occurred.
Those are editorial evaluation questions, rather than published judging criteria. A team could choose a small invoice workflow and make an exception visible: a disputed amount, an instruction outside a permitted range, or a request that needs review. Narrow scope gives an audience more room to examine the mechanism. A spectacular interface can move a demo along; a clear exception path gives a finance practitioner something specific to interrogate.
Why approved financial action is a current market question
On October 16, 2026, Visa described a live pilot with BILL and Billtrust enabling trusted agents to complete approved buyer invoice payments through supplier portals. Visa emphasizes authentication, supplier oversight and transaction visibility. The pilot provides relevant commercial context for this hackathon's themes, without establishing any relationship between those companies and the event.
Visa's Trusted Agent Protocol documentation separately describes cryptographic verification of agent requests. Identity and authorization therefore give builders concrete design territory to explore. Recognizing an agent, understanding whose instruction it carries and deciding whether a particular action is allowed are related questions, each deserving a visible answer.
For an operator evaluating a prototype, the productive conversation begins with a specific workflow. What information does the agent receive? Where does its permission come from? Which action requires approval, and what record remains afterward? These questions help translate engineering choices into the language of operational responsibility. They also make feedback more useful than a broad debate about whether finance is ready for autonomous software.
The hackathon offers a bounded place to expose those choices. A day of building can produce an inspectable example, leaving deployment decisions, integration work and organizational approval for the people responsible for the eventual system. Readers following DevCuration's coverage of agent reliability and finance-focused hackathons in New York can explore related operating questions across the city's build events.
The people on the current event lineup
The official event page lists Veljko Kovac as both a speaker and judge. A Tessl speaker biography corroborates his role leading forward deployed engineering at Cognee, including technical implementation of agent memory for enterprise clients. His presence puts an engineer working on persistent agent context into the conversation about financial action.
Joe Zhou is also listed as a speaker. Monk's own interview with Joe Zhou identifies him as co-founder and CTO and discusses the reliability demands of accounts receivable automation. That is a useful operating perspective for a gathering where collections and invoicing are explicit build themes. The event page does not publish individual talk titles or promise particular technical lessons from either speaker.
The other listed judges are Pradeep Dhananjaya, Tech lead at AWS; Douglas Luman, Tech at Coder; Noah Jenkins, AI Enablement Engineer at Coder; Andry Lebedev, Swipelux co-founder and CEO; Rohan Prasad, AI Engineer at Monk; and Jeffrey Chi, Software Engineer at Monk. These are organizer-listed affiliations, with no sponsor relationship established by the roster.
For a builder, the potential value is practical feedback across the problems represented in that lineup. For a founder, it is a chance to make the proposed workflow understandable to someone outside the team. Neither benefit requires pretending the room guarantees investment, customers or a production deployment.
What to prepare before December 5
The official organizer calendar confirms December 5 in New York. Luma's event data gives a 9:30 AM EST start and a 7:30 PM EST end on that local date. The written agenda schedules hacking at 11:00 AM, lunch at 1:30 PM, submissions at 4:30 PM, finalist demos and judging at 5:00 PM, and awards and closing at 7:00 PM. Applicants should confirm the final schedule with the hosts.
Registration is free and approval-based. The organizer says pending or unapproved applicants will not be admitted, and the exact address is available to registered guests. Engineers, quant builders, fintech founders and finance practitioners are the stated audience. A prospective participant should apply through the official event page and resolve team arrangements before arriving.
An effective preparation exercise would be to choose one financial task and describe the handoff in plain language. Write down what the agent may do, what information it needs, and who handles a case outside its authority. Those choices can guide a prototype without implying that the event requires a particular architecture or compliance standard.
NIST's voluntary AI Risk Management Framework provides broader context for considering trustworthiness during design, development, use and evaluation. At the hackathon scale, the work can stay concrete: a defined task, a visible decision and an exception another person can follow. By the afternoon submission deadline, teams will have an opportunity to put those choices in front of the people watching the demo.
Kinneberg Management Group Adds Thompson Street Growth Backing|Ancep Raises $8.5M for Commercial Insurance Broker AI|Sparxity Series A Backs Sparse AI as Total Reaches $90M|Lombard Systems Raises $4M for Tokenized Asset Financing|Pier's Allocator Market Event Puts Fundraising in Focus|Twilio Assemble NYC Brings AI Communications to Brooklyn|Mercor Raises $750M Series D for Enterprise AI Evals|Bay Raises $37M Series A for Enterprise AI Agent Security|Alterra Labs Raises $12.5M Seed for Medical AI Evaluations|Blue J Raises $176M Series E for Agentic Tax Workflows|