Money Talks 3 Brings Finance Mentors to New York AI Builds

MONEY TALKS #3 · AI × Finance Hackathon is scheduled for November 21, 2026, from 9 AM to 6 PM EST in New York. Block Convey is hosting the in-person event, with Arun Prasad listed as host. The announced program combines small-team building, finance-mentor input and a demo whose underlying agent activity can be inspected.

For founders and engineers, the attraction is a chance to put a finance problem beside someone who understands the work before spending the day coding around an assumption. For finance operators, the room offers a way to make a workflow problem tangible. The commercial question running through the format is how much a prototype improves when business judgment and technical evidence arrive before the final presentation.

Finance mentors enter before the demo

The organizer plans teams of 2 to 4, finance-mentor support and a 30-minute opening workshop. Money Intelligence, Money Operations, and Trust & Risk are the announced themes. Those categories leave builders room to work across financial research, operational tasks and safer financial AI without prescribing a single product.

A mentor can change the economics of a short build by challenging its starting point. An engineer may see an opportunity to automate a sequence of steps; the operator may know that the exception buried in step 3 consumes most of the working day. That conversation can produce a smaller prototype with a more credible purpose. It can also stop a team from polishing an answer to a question the intended user never asks.

For example, a hypothetical invoice assistant might extract the visible fields accurately while leaving a disputed charge for someone else to reconcile. The useful discussion would concern the disputed charge, the authority to act on it and the information needed to resolve it. That is an illustration of the design problem, rather than an announced hackathon project. In a one-day setting, choosing the right boundary can determine whether the demo has an operating story to tell.

Agent traces give the room something to inspect

Block Convey says teams will use PRISM and judges will be able to inspect traces. Its PRISM documentation describes records of model exchanges, conversations and agent runs. That makes the route to an answer relevant to the demo, alongside the answer itself.

For an agent that retrieves information and calls tools, a plausible final sentence can conceal several different failures. A team could retrieve the wrong document, lose a changed instruction or perform a step that no longer serves the user's goal. Those are different engineering problems. A visible execution record can give a mentor or judge a more specific question to ask, provided the relevant activity was captured.

The company's agent-observability product page describes views of model calls, tools, retrieval, state and outcomes. These are vendor-described capabilities, rather than an independent assessment of PRISM's performance. Block Convey also has a direct commercial interest in the event because it supplies the platform participants are expected to use. That relationship makes it useful for builders to ask what the trace actually reveals and where interpretation still depends on the person reviewing it.

The finance workflow remains the hard part

The wider relevance sits in the handoff between software behavior and business responsibility. A team can record that an agent took an action while still needing to establish whether that action was appropriate. The person who owns the process brings knowledge that a sequence of successful tool calls cannot supply on its own.

NIST's voluntary AI Risk Management Framework provides broader context for considering trustworthiness in the design, development, use and evaluation of AI systems. NIST has no stated involvement in this hackathon. Its framework is useful background for the distinction between observing a system and deciding how its risks should be managed; attending the event or instrumenting a prototype does not confer a certification.

That distinction gives finance operators a substantive role in a technical room. A useful exchange might concern which inputs can be used, which decisions require review or what evidence would persuade the next person responsible for the workflow. For a founder, hearing those constraints while the product is still flexible can sharpen the next development decision. The prospective benefit is a better-informed build, with its limits visible enough for another person to challenge.

What participants can establish before November 21

Registration is free and subject to host approval through the official event page. The exact venue and address are shared through registration. The current listing does not name the individual mentors or judges, so applicants should use the organizer's current instructions for those details rather than assume a particular institution or executive will be present.

Earlier Money Talks editions have separate announcements. Their sponsors, prizes and participants should not be carried into the November program. For this edition, the supported reason to pay attention is the announced relationship between finance-mentor input, a working prototype and an inspectable run. Any later hiring, investment, customer relationship or production deployment would require its own evidence.

Within New York's technology-event calendar, this format gives builders and workflow owners a specific piece of work to share. The most useful artifact may be the version the team changes after a mentor points out an assumption, with the trace still open and enough of Saturday left to build again.