Inside Arca's Long-Horizon Agents That Run the Business - Aman Singh, Arca

AI Engineer · 13 min · 151 sentences · from YouTube's caption track

Each timecode opens YouTube at the start of that sentence. Line anchors (#s42) are the cue ids in the WebVTT, and every line carries its start and end seconds. All transcripts has every talk, and the whole corpus as one file.

00:01[music]
00:12My name's Aman.
00:14Uh I lead AI at Arca.
00:16Um and today I'm going to talk to you about Arca's long horizon agents that run our business.
00:22Arca is an AI native services firm.
00:26We provide our clients with financial services like financial planning, tax planning, asset management, investment advice.
00:35Um and we serve a range of clients.
00:37Some of them are like technology executives that um have amassed a lot of wealth and need a real plan to retire early.
00:44But also that family in New Hampshire that made their first million dollars and and want to buy that house finally.
00:52Um I said we were AI native.
00:56There's a lots of There's lots of definitions of this this term these days, very overused.
01:00But what we mean by AI native is that agents run our business.
01:07Not that they assist the business from some co-pilot corner.
01:11They They execute the decisions that shape the outcomes that our business has for our clients.
01:19Work in this industry, it's interesting to know what it looks like.
01:23In financial services, the work is more long horizon and less long running.
01:28And what I mean by that is this is less agents grinding tokens for like hours or days on end, right?
01:36This is more one client goal that spans 10 days.
01:40It's very stop-start.
01:42There's many actors in between, many decision points, many hand-offs.
01:47Um and ultimately that goal lands like 10 days later.
01:50That is the kind of work that that we're working on.
01:54And so here's an example.
01:55Jane's our client, fairly wealthy, but has a big tax bill.
01:59Oh.
02:01Um has a big tax bill in uh in in April 15th, right?
02:06So, we're going to help Jane figure out what is the most tax optimal way for this cash to be to be raised to pay the tax bill.
02:16So, there's the advisor that will do all of the the statement analysis and the analysis of of how to make this tax optimal.
02:23There's the the trader that would actually liquidate the the assets.
02:28They the assets themselves will take like days to settle, and then the ops people will come in and kind of like trigger the ACH.
02:34That will itself take days uh to land.
02:37So, this is kind of the work I'm I'm talking about in this industry, which takes like days to settle.
02:44And of course, if you do this at scale for several several clients, the amount of time our firm takes to do this sort of work exponentiates, right?
02:51So, at Arca, we took a different approach.
02:56It's helpful to know that Arca is not a SaaS company.
03:00We are not selling software to other businesses, to other firms.
03:04We are the firm.
03:06We're an applied AI company that happens to be running a services business in this industry.
03:11And so, the approach we took was, all right, we're going to have agents own this whole long horizon outcome end to end.
03:18And for that, we gave them a couple of key components, I'll give you a sneak peek of.
03:23We gave them signals from the external world that keeps them really aware of what is happening.
03:28So, that email will be read by agents.
03:31We give them all of our operational context on what to do about that email.
03:35What does this industry do with this situation of taxes?
03:38Right?
03:38We gave them a harness that gives them all of the tools and uh the guardrails, and more importantly, persistence of memory over the the kind of long horizon period
03:49to carry that work out.
03:51And of course, we gave them the ability to collaborate with people.
03:57So, that work with um with agents compresses a lot because all we did was we took humans out of the critical path of of this work.
04:06Nobody has to babysit this this uh this goal over days.
04:11Agents are are on it.
04:12In this talk, I'd love to describe to you how we actually do that.
04:17Let's follow the journey of this email.
04:21Jane owes on the $100,000 for taxes.
04:24This email, much like many other signals in our in our platform, will be will be read by background agents.
04:33So, there's like phone calls our firm is making, there's documents that are coming in, there's meetings being had, there's events coming in from our custodian, there's market movement that's happening.
04:42These are all signals that our background agents are continuously evaluating to progress work.
04:48So, this email is one of those signals.
04:50And when the signal is is by agents, they actually get to the work, right?
04:54Um we have a harness that gives them the knowledge to know what to do about the situation.
04:59They have deep client memory, so they know, "Hey, let's never touch Jane's Apple position cuz uh she said that in a call like 8 months ago, right?"
05:08Um we of course have a sandbox for all of the financial computation that needs to happen for this analysis.
05:14Um but largely, I think harnesses today are are coalescing on a standard shape.
05:18So, um it's more interesting what happens outside of outside of this harness.
05:25So, after they've done the work, they produce this plan.
05:28This plan by itself cannot be executed without people.
05:32So, in our product, we build a space, agents build a space, where they put out this plan, and there's you can chat with the agent to like kind of know what's happening with the plan.
05:41There's mult- multiple people that can be chatting with the agent.
05:45And more importantly, this space is where also they pull people in when they need people to execute some work.
05:51So here in this case, we would approve uh the plan and the email out to Jane.
05:57And then there's this whole dance of has Jane replied?
06:01What does her What does her answer mean for the plan that we had, right?
06:05And then all of the these other questions that will have to be answered throughout this whole like long horizon work.
06:11At Arca, we were like, of course we're not going to have people babysitting it.
06:15So we let our agents watch the world.
06:18So people don't have to, right?
06:19So we give them tools to be able to register interest in anything happening in our world.
06:25So they're able to say, let me know when Jane responds.
06:29Let me know when that ACH lands.
06:31Right?
06:31And this is powered by our signal system in the in the background where agents are routing uh signals to agents that are interested in it.
06:41And of course, if the signal never lands, agents can like wake themselves up on on a schedule as well to keep track of work.
06:52But then, of course, the real world is very messy, right?
06:56People are waiting for that email, but Jane never responds to that email.
07:01Say we we meet her in person and we learn some information that affects the plan.
07:07So the person goes in and they they add this information into our UI.
07:11But this act of adding information or doing work in our UI is also a signal that will wake up an agent.
07:19In fact, when people work in our in our product, when they reschedule a calendar event, you know, that's a signal.
07:26When they mark a task complete, that is a signal.
07:29So this system lets us be both aware of everything happening in the external world and the internal world as work is progressing so our agents can participate in work.
07:43And of course, agents are communicating with each other as well.
07:46So, this signal read by a background agent will let it notify another agent about this so that it can update the space to say, "Hey, we just need a need $80,000
07:56instead of the $100,000."
07:57So, the work has changed because of because of these signals.
08:02And the last piece of the story is that agents are able to pull in the right people at the right time.
08:06So, uh the agent on the left has read our operational context to be like, "Okay, this the next step here is to pull in the the person that trades
08:15these uh these assets."
08:17And so, that person now gets a task and gets pulled in to participate in this long-horizon work.
08:24What we've done overall, if you take a step back, is we've given our agents the primitives for driving long-horizon work.
08:32What do people do at work, right?
08:33They They need to stay aware of what is happening, so we've built this very intense like signal system for our agents to be aware.
08:40They need to know what to do next, so we've given them all of this operational context.
08:44They need to do the work, so it's our harness that lets them actually perform the work and keep track of it, right?
08:49It's long-persistence um memory systems that that hold the state.
08:54And bringing in the right people at the right time.
09:00This is a pretty simple idea on the face of it, but there's a lot of emergent complexity.
09:06You can imagine at any point there is hundreds of these signals coming in, and they are, you know, tasks tasks are being added on people's plates, they're being pulled in, background agents are communicating with each other,
09:18right?
09:19So, over the last 6 months, we've we've been working on getting this system to be really good.
09:24We've been closing all of our context gaps so that agents are fully aware.
09:29We have been working on the precision of tasks, so pulling in the right person at the right time.
09:35Work that needs to be tracked for the future is tracked in the agent's memory.
09:39Work that is doable now becomes a task, right?
09:42So, we've been working on on our ability to hold state over long horizons as well.
09:47And because we are the firm and not just a SaaS company, we've been changing and standardizing how work happens in the firm in the first place and operationalizing that context into
09:59our agent's state.
10:00And because these are our colleagues that use our product, we collect rich reasonings from them that become evals for for our system to improve.
10:11So, I guess there's there's a question to be asked here, like why do this, right?
10:15What does this unlock?
10:17What is the takeaway here, right?
10:19For for us, this is our whole thesis.
10:23Is that we want our platform to be so AI native that agents drive the business.
10:29Agents run the business.
10:31They they are making the decisions that shape our outcomes.
10:35And so, at Arca, when people log in, right, they see a list of tasks that have been prepared for them by agents.
10:43They've already read that email.
10:44They've like looked at the document that came in.
10:47They've done like 90% of the work and they're pulling you in just when it's time for you to act.
10:54So, building this system lets us get to a point where at Arca, agents prompt people more than people are prompting agents, which lets our agents drive the work that our our firm
11:07does.
11:09And because agents are driving the work the firm does, agents can drive the business.
11:15Let me give you an example.
11:16Remember that email that Jane sent us?
11:18This is a bit of a failure, right?
11:20Because we should have just been proactive about this in the first place, right?
11:24So, Jane should never have to send this email.
11:27So, say at you know, our AI native business, we take on this new goal.
11:32Our new goal is proactivity.
11:37Every January, prepare each client's tax situation and proactively reach out to them before anyone even asks.
11:45This is a new business goal for our business to drive.
11:47And in an AI native firm, all we have to do is make this goal part of our agents' set of goals.
11:56And because they have the primitives to be able to drive work through people for the business, they start driving this goal, and just like that, you have you have these analyses being done in January,
12:10and these these one-click approval emails with the full plan for every client showing up on our users' plates.
12:18And because they're aligned with the goal of proactivity, they can one-click approve and drive the business towards this this new outcome.
12:28So, this is sort of I think the key difference between a a company that uses AI tools versus a a company that is AI native.
12:42In an AI native company, you know you've built one when you can steer your business by steering your agents.
12:50And that is our ambition at at Arca, and we are we're hiring.
12:58Um so, please uh please check out our website.
13:01Please reach out to me if this is something that you you're you're interested in working on these sorts of problems.
13:06And um thank you for listening.
13:09[applause]