By Anurag Bhardwaj, Product Engineer at Auquan
Written with Mohd Ali Rizwi, Product Engineer at Auquan
At Auquan, forward-deployed engineers work directly inside financial institutions to build AI agents that eliminate manual work from investment workflows. We sit with the analyst teams, learn how they actually operate, ship solutions on the ground and integrate those back into our product. This post is about what that looks like in practice, based on our latest on-site customer deployment.
My colleague Ali and I are both product engineers at Auquan, but when we're on-site with a customer, we're doing forward-deployed engineering work. At Auquan those two things aren't separate jobs. We build the technology that powers our AI agents and we deploy it on-site inside financial institutions. The same person who writes the code sits with the customer, watches how they work, and brings what they learn back into the product.
We were sent to a global investment firm to develop a two-week proof of concept in close collaboration. The firm's analysts were spending days manually producing investment reports. Every deal required pulling data from scattered documents, synthesizing it, and assembling dense, highly formatted presentations. On top of that, they were paying an outside firm just to handle the output document formatting (PowerPoint this time), which added another bottleneck of days to every report cycle.
Our job was to show that Auquan's agents could take on the manual heavy lifting and produce complete, ready-to-review reports in minutes instead of days.
We flew in on short notice and spent the weekend planning. David Ardagh, who runs product at Auquan, had been building the customer relationship for months, but Ali and I were walking into this customer, a major financial institution, for the first time. This is always intimidating. These are serious, sophisticated organizations, and we were there to prove something.
We showed up Monday with a plan for what we would demo. The customer's IT lead looked at it and was straight with us: what we showed wasn't going to work for their analysts.
That was a tough moment, and a clear reminder that every firm works differently. We had to throw out our prepared approach and start listening instead of presenting.
The thing that I have now come to expect but is still surprising: every financial institution is extremely specific with their requirements for output formatting. These teams didn't just need the right analysis. They want it in exactly the right format, with exactly the right colors, exactly the right density of content per slide, exactly the right chart types. Every detail matters.
As Ali put it: "They were very adamant about color formatting and achieving the same visual layout."
This makes complete sense once you understand why. A report generated by our agents feeds into a much larger process. If we change the formatting in any way, it is likely to disrupt something downstream. The customer wanted to eliminate the manual effort, not change how their organization works. That distinction shapes everything about how we build.
You can't understand this from a Zoom call. You have to see the actual documents, sit with the people who use them, and understand how each report moves through the organization. These nuances are invisible from the outside.
During a POC, you don't need to deliver a finished product. You need to show the customer that you understand their specific problems and that you can address them and that you're heading in the right direction. That's a different kind of engineering challenge.
After the Day 1 reset, we shifted to rapid iteration. Being physically close to the customer's office meant we could meet in person twice a day, get immediate feedback, and adjust. On a call, you might misunderstand a requirement and not realize it for a week. In person, you catch it in minutes.
Midway through the engagement, the customer's IT lead actually ran some general-purpose AI coding tools over a weekend to see if he could solve the specific PowerPoint formatting problem himself. He got partway there, but it was frustrating for him, and he struggled to replicate it with subsequent runs. The output we generated using Auquan's purpose-built approach was significantly better, and the team could see the difference immediately. That was a critical turning point.
Ali described the final days of the POC: "We got to a state where we could generate PowerPoint presentations with exact custom instructions per slide. They could customize, edit, do things after the slide had been generated. That was a wow moment. The analysts and associates were genuinely impressed."
Getting from "this isn't going to work" to that reaction in two weeks took focused, in-person collaboration and a lot of late nights. But the feeling of watching someone's workflow go from days of manual effort to minutes of review and refinement is hard to beat.
For anyone considering an FDE role like this, I want to be clear. This is not consulting. We're not there to take orders and write custom code that only one customer will ever use.
Every customer engagement teaches us something about how the product needs to evolve. The PowerPoint generation capability we built during this POC became part of our core product. The integrations we built did as well. When an analyst tells you their report needs to include specific chart types and you have to figure out how to automate that, you're not just solving their problem. You're making the product better for every future customer.
The hardest part of forward-deployed work is not the technical problem-solving. It's resisting the urge to immediately build the thing the customer asks for and instead asking why they need it. Ali and I talked about this a lot during this deployment. Customers will tell you they want a specific feature. But if you take that at face value, you might build something that solves the surface-level request but misses the actual outcome they're trying to reach.
Our job is to understand the outcome and then work backwards to the right solution, one that works for this customer and also makes the product stronger overall.
Ali's advice: "In forward-deployed work, you have to understand what the customer actually requires from the solution rather than just building the fastest or most technically elegant fix. The business outcome matters more than the engineering."
I'd add: take a step back before you start building. Any engineer can write code and get something to a working state. The question is whether what you built actually gets the customer the outcome they're seeking. During an FDE on-site deployment like this, it's more important to show you understand their use case and that you're moving in the right direction than it is to quickly deliver a polished feature that, in the end, might miss the mark.
And get on-site whenever you can. Being in the room with finance professionals changes how you think about the work. You learn how they talk about problems, what they actually care about, and where the real friction is. That understanding feeds directly into a better product.
Ali told me something during this trip that stuck with me: "Before Auquan, I had never met or interacted with customers face to face. I can now see firsthand the impact of what I build. I know what's working and what's not."
For me, I get energy from challenging situations where you have to figure it out under pressure. I don't enjoy the exam where I have all the answers prepared. I enjoy the one where I have no idea what's coming and have to work through it in real time. Auquan puts you in those situations constantly.
We're also at an interesting inflection point in engineering. Writing code used to be the bulk of the job. Now, with AI tools accelerating that part of the work, the real challenge shifts to building end-to-end solutions that actually solve business problems. That's what forward-deployed engineering is about, and it's a much more interesting job than writing code in isolation ever was.
Anurag Bhardwaj and Mohd Ali Rizwi are Product Engineers at Auquan, where they build AI agents that eliminate manual work from institutional finance workflows.