Engineering leadership knows the standard playbook for product teams: the Product Trio, the Spotify model, outcome-based roadmaps, and so on. I've seen teams adopt these practices and still struggle when they add responsibility for advanced algorithms -- search, recommendations, predictive modeling, or generative AI -- without changing how the team is led. The processes and role expectations that work for a typical product team break down when the domain is heavily algorithmic. This post pulls together what I've learned and what others have written about building teams that own these systems.

The Product Trio Meets Advanced Algorithms

A Product Trio gives you a standard set of operational processes: how the team communicates, how decisions get made, and what each role is expected to contribute. Product managers synthesize user needs and business goals, designers own the voice of the user and the interface design, and the engineering lead ensures that the system can be built, scaled, and maintained. What happens when the domain a team is responsible for is highly algorithmic -- search, generative AI, or other advanced algorithms? The Trio is no longer sufficient.

This isn't a new observation. Teresa Torres writes that if you work on a data-intensive product, your trio might become a quad: you may want to invite your data analyst to participate in most key decisions. Kevin Holland argues that data product teams need a data lead, not just occasional support from analytics. In search specifically, Daniel Tunkelang has written about how product and engineering leads on search teams have particularly overlapping roles. And James Rubinstein has unpacked the search PM role and why it's often misunderstood. These authors are all pointing at the same thing: when the core of your team's responsibility is an algorithm, the standard trio doesn't map cleanly onto the work.

What Does This Look Like in Practice?

So what skills, exactly, are needed? And who represents the voice of the user?

For search and similar algorithm-driven features, the voice of the user doesn't come primarily from interviews or usability tests. It comes from diving into a highly complex and stochastic set of logs and data sources, and from thinking through how algorithms shape user behavior. You need someone who can study query logs, click-through rates, reformulation patterns, and engagement metrics, and connect that to algorithmic changes that might help. That person needs both data skills and a deep enough understanding of the algorithms to reason about cause and effect. Many product teams have business analysts who can pull data and build reports, but for advanced algorithm teams, deep quantitative skills and the tools needed to rapidly discover and iterate are more-or-less the whole game.

Reganti and Badam's recent O'Reilly piece makes a related point for generative AI: evals are not enough. You need a process of continuous improvement, a "flywheel," to deal with the complexity and stochasticity of an advanced algorithm. You can't just run a few tests before launch and call it done. The same logic applies to search and ML systems. The work is inherently iterative and data-driven.

Compared with classical product teams, the code written for advanced algorithm teams is much less likely to be user-facing production code. There's a much higher need for internal tools such as team-specific reporting systems, interactive analysis tools, and prototypes for exploration. In search, that might mean tools to test new algorithms or UI patterns, alternative administrative UIs for specific business or technical problems, or custom dashboards that surface metrics that matter for your ranking and relevance. In ML, it means tools for building, scaling, and monitoring predictive systems. Generic analytics platforms rarely give you the right abstractions -- every advanced-algorithm product is too different, too context-dependent.

Given the advances in easy-to-use development frameworks and AI-assisted coding, it becomes easier than ever for the same person who is diving into the data and algorithms to build their own tooling. The data and algorithm lead becomes less dependent on a separate tools team, or on their team's spread-thin engineers, and more able to close the loop from insight to prototype to iteration.

Who Leads, and Who Doesn't

You still need Product -- for synthesis, coordination with external teams, prioritization, and making sure the algorithm work connects to business outcomes. You still need Engineering management -- for supervision, career development, and technical direction. But maybe you don't need design in the same way? Or at least, design plays a different role. The data and algorithm expert (whether you call them a data scientist, search scientist, or something else) provides the voice of the user, via the data. I think it's important to have the same person looking at the data and being capable of thinking through complex algorithmic changes. Splitting those responsibilities across a designer and an analyst who can't reason about the algorithm tends to create handoffs and lost context.

So, for an advanced algorithm team, the Product Trio becomes Product, Engineering, and Data & Algorithm leads. Design, as traditionally understood, becomes a supporting role, pulled in as necessary rather than driving the experience.

The Gen AI Era

How does this change when we add Generative AI into the mix? Two angles matter: how Gen AI affects software development, and how it affects teams that deploy Gen AI systems.

On the software development side, the conventional wisdom is that in the world of Gen AI, many more developers can be "10x" developers compared to a few years ago. You don't need as many developers, and the bottleneck is likely to be figuring out what to build rather than building it. Marty Cagan and the SVPG team have written about how people on product teams will focus more on discovery, letting AI agents take more responsibility for delivery.

For advanced-algorithm teams, you still need the data science capabilities and the deep understanding of algorithms in order to determine what to build. The need for deep expertise doesn't go away, even if AI tools speed up aspects of data analysis, brainstorming, and more. Teams focused on Search or other advanced algorithms will get smaller and faster.

When it comes to deploying Generative AI systems specifically, even though integrating an LLM is just a few lines of code, the implications for software lifecycle management are similar to other advanced algorithm systems, particularly Search systems. Teams building Gen AI products may find they have more in common with Search teams than with teams that ship traditional ML models. Gen AI systems, like Search, take free-form linguistic input and produce complex output, so evaluation, monitoring, and iteration look different from predicting a likelihood or a forecast from structured, tabular data. Even though the algorithms used in Generative AI and predictive modeling may be similar (deep learning), the product implications are very different.

UI/UX considerations for Search and particularly chat-based Generative AI have much in common, as well. In both cases, the UI needs to help users understand what's possible, to meet them where they are, and to handle extreme variations in use cases and expertise.

All of which brings us back to team structure. If Gen AI products share more organizational commonalities with Search than with traditional ML, then the same team model applies: Product, Engineering, and Data & Algorithm leads working together, with Design in a supporting role. The "voice of the user" comes from understanding linguistic patterns, evaluation metrics, and iterative feedback loops.

I’ve seen these patterns play out repeatedly on search, generative AI, and machine learning teams. If you’re trying to structure one of these teams differently, I’d be interested to hear what’s working for you. If you're looking for organizational help or a data and algorithms expert, I'm a freelance consultant with many years of relevant experience who can help your Search, Generative AI, and other teams be maximally effective. Get in touch!

Note: This post was primarily human-authored, with AI assistance for research, editing, and organization. The AI filled a Secondary author role. The core ideas and final voice are mine.