In our Meet Addepar series, we sit down with the people at the heart of our success — those who enable us to consistently deliver excellent outcomes for our clients. Today, we speak to Arda Sahiner, Engineering Manager on our Core AI team.

Arda leads a team building AI products across Addepar's platform, working at the intersection of large language models and some of the most complex, high-stakes financial data in the world. Below, he talks about what drew him to Addepar, what makes engineering here different and why he thinks context engineering — not the next model release — is where the real opportunity lies for AI engineers.

What drew you to Addepar?

The size and impact of the mission were what really drew me in — maximizing the impact of the world's capital is a mission I'm genuinely excited to get behind. Beyond that, it's the complexity and scale of the problems we work on. Operating across trillions of dollars of assets on the platform means every decision we make has real weight, and you inherit all the constraints that come with that kind of complexity.

What separates the people at Addepar?

I've worked in a lot of different environments, and what stands out most at Addepar is the care and love we have for our clients. That is baked into everything we do. We're a genuinely client-centric company, and that shapes how we solve problems together, because the client's perspective always comes first.

What's it like building AI for financial data specifically?

What's unique about building AI for financial data is how accuracy-sensitive everything is. That pushes you to innovate at the product layer, figuring out how AI products surface themselves and get used in ways that build trust alongside accuracy. We're constantly finding creative ways to deliver the value of AI while keeping our users in the driver's seat. It's a fascinating set of constraints to work within, and solving for them is one of the more rewarding parts of the job.

What's different about managing an AI team?

Managing an AI team means thinking differently about verification and accuracy, and building systems that support that at scale. We move fast and experiment constantly, so the real challenge is balancing rigor, and its verification, accuracy and performance, with the freedom for individual engineers to move quickly and take risks. That tension is one of the more interesting management problems I get to work on at Addepar.

What's surprised you most about Addepar's culture?

Given our mission, you might expect our culture to be finance-first. What surprised me is how tech-forward Addepar actually is. We care deeply about experimentation and moving fast, paired with a rich domain of financial expertise and a real love for our clients. That combination means we get to ship quickly and then see the direct impact of that work on clients, and celebrate it together as a team. That's been a genuinely rewarding surprise.

What's your AI hot take?

My hot take is that the models themselves are already good enough — you don't need to spend your energy tracking every new model release. For the vast majority of enterprise workflows, the real bottleneck is context engineering and context management, not model capability. Unless you're doing frontier research, the new model that comes out next month isn't what determines your success. What matters is how well you encode the right context and specify the right information into your systems. That's where the leverage actually is.

The deeper issue is under-specification: we haven't yet given these systems enough context about the world they're operating in, or been precise enough about what we actually want them to do. If you gave a person a task with only vague instructions, they wouldn't succeed either — not because they lack capability, but because they lack information. A lot of the industry conversation stays fixated on the next model drop, and for the hardest frontier problems, that's still relevant. 

But for the much larger set of everyday enterprise problems, the constraint isn't the model — it's how we, as humans, organize and deliver the right information to it. That's the layer Addepar is focused on: encoding semantic context into our systems rather than waiting on the next training run. There's enormous value still to unlock simply by getting better at specifying what we want.

Why is now a special time to join Addepar, especially in AI?

Joining Addepar in AI right now is a genuinely special moment. We're at an inflection point where AI has real capability to do disruptive, high-value work, and Addepar has already built the foundation to take advantage of it. We’ve built the data foundation, the scale, the workflows and a deep understanding of what matters to our clients. That combination makes this the right domain and the right time to show real impact with AI across every part of the business. You're solving new problems constantly, and the work has a direct, visible effect on the company and our clients.

What's the worst coding advice you've ever received?

The idea that coding is its own separate activity, disconnected from the actual problem you're solving, is probably the most misleading advice I've encountered. That's especially true now, with so much of the mechanical work compressed by AI. What you're really doing is problem-solving. Engineers who treat coding as a distinct, isolated skill tend to focus on the wrong things entirely.

What's your AI toxic trait?

I'll admit that when I'm coding with AI, I can be fairly blunt with it — probably more than I should be. I say thank you when I'm happy with the result, but if I'm not, I'm not exactly diplomatic about it either. I'm not emotionless in those interactions, for better or worse, and it's something I could stand to soften.

What's your team's superlative?

Most nerdy, without question, and I mean that as a compliment. I count myself among them, and honestly, I think the world could use more nerds.

What makes your team different?

My team has been taking on more of a Gen Z energy lately, and I'm fully here for it. I'd call myself a Zillennial, technically a Millennial, but spiritually aligned with Gen Z. I'd love to see even more of that energy show up on the team.

What's your post-ASI dream?

After artificial superintelligence arrives, I'd want a ranch in Montana — completely off the grid, surrounded by nature. It's probably the go-to answer for a lot of people in this field, which is a little ironic, since most of us thinking about a post-ASI future are also the most chronically online or always working. I suppose I'm no exception.

How do you decompress after a long day of work?

I have a dog, and walking him is a reliable way to unwind.

Interested in building AI products at the frontier of financial data? Addepar's core AI team is hiring.
Explore open roles

A note on context engineering

Arda's hot take echoes a shift playing out across the AI industry right now: the discipline that determines whether an AI product actually works is increasingly context engineering, that is the practice of deciding what information, tools and instructions to put in front of a model, rather than chasing the models shipped most recently.

A few sources worth reading if this resonates:

  • Anthropic's engineering team lays out the case in detail in "Effective context engineering for AI agents" (Sept. 2025), describing context engineering as "the natural progression of prompt engineering" and outlining techniques including compaction, structured note-taking and sub-agent architectures for managing context as a finite, high-value resource.

  • Shopify CEO Tobi Lütke is widely credited with popularizing the term. In June 2025, he posted to X, ”I really like the term ‘context engineering’ over prompt engineering. It describes the core skill better: the art of providing all the context for the task to be plausibly solvable by the LLM.”

  • Andrej Karpathy, former Tesla AI director and OpenAI co-founder, amplified Lütke’s post saying,  “People associate prompts with short task descriptions you'd give an LLM in your day-to-day use. In every industrial-strength LLM app, context engineering is the delicate art and science of filling the context window with just the right information for the next step.”

  • Independent AI researcher and writer Simon Willison tracked the term's emergence in his post "Context engineering", arguing the phrase captures the real work of building reliable LLM applications better than "prompt engineering" ever did.

  • Google Developer Expert Philipp Schmid makes the practical case in "The New Skill in AI is Not Prompting, It's Context Engineering," arguing that "most agent failures are not model failures anymore, they are context failures."

For engineers weighing where to spend their careers in AI, the throughline across these sources is the same one Arda points to: the biggest unlock is figuring out how to feed today's models the right information, in the right form at the right time. That's exactly the kind of problem Addepar's AI team works on every day, applied to some of the most complex and consequential financial data in the world.