What it’s like to build AI agents at Decagon

At Decagon, we build powerful AI agents that transform customer experience for companies like Hertz, Chime, Duolingo, and Notion. Our AI agents can be easily trained, prompted, and controlled through plain language, solving incredibly complex use cases—from intricate order management issues to nuanced product functionality questions.

While our platform abstracts much of the complexity of working with LLMs, deploying an AI agent that consistently performs in production is still a process of exploration. Businesses often begin with a clear goal, but getting from idea to outcome requires iteration, context, and care. That’s where the Agent Product Manager comes in.

So, what is an Agent Product Manager?

Agent Product Managers are Decagon’s in-house experts on building, deploying, and scaling AI agents. We work directly with brands to bring their AI agents to life, and then grow them into core parts of their businesses. Think of us as part product manager and part AI consultant.

I've personally lived this role, having joined Decagon as our first Agent Product Manager. I've helped major global brands deploy AI agents that significantly enhance their customer experiences. This means working hands-on with everyone from CTOs to individual support agents to take their initial idea to fully operational LLM-powered agents.  

Other Agent Product Managers have found success helping build agents that:

• Empower Rippling customers to quickly resolve complex payroll and HR inquiries
• Simplify subscription, billing, and publishing for Substack creators and readers
• Help ClassPass members automate refunds for unexpected circumstances  

Owning the agent lifecycle

Agent Product Managers work as the connective tissue between Decagon’s platform and the brands who rely on it to deliver exceptional customer service. Because we own every stage of the agent build process, we’re a blend of many functions rolled into one. Our job requires us to be malleable to the needs of both our customers and the business at large. As a result, on any given day, we might be:

• Collaborating with engineering to design a new product feature
• Writing and testing prompt logic for a specific use case
• Working with a customer’s executive team to define their AI roadmap

The pace is fast, but fundamentally collaborative by design, both with our customers and our internal teams. We spend as much time building inside the Decagon product as we do charting a course for each AI agent we support. As a result, we’re often the person in the room who has to translate between technical expectations and business impact. 

But let’s be a bit more concrete: imagine a business wants their AI agent to explain a user’s loyalty points, based on account activity. That use case sounds simple: a few API calls, maybe a knowledge lookup. But in practice, it can involve much more: 

• Meeting directly with a customer to understand their scoping, documentation and use case to develop the initial Agent requirements document
• Exploring the “question beyond the question” to not only help users understand their loyalty points, but also advise them on any follow-ups they might have
• Partnering directly with Decagon Engineering and Design to scope and build out the use case from end-to-end 
• Testing the workflow for functionality, tone, correctness and other core KPIs to ensure enterprise-ready functionality
• Rolling out the feature to users across the customer’s target audience, measuring success and identifying areas for iterative improvement

Our team takes ownership of that process from end to end. Our goal in this situation isn’t just to build a new feature, it’s to ensure the AI agent performs reliably in a way that aligns with our customer’s goals and scales with our product. 

A function built on range

As we’ve scaled, we’ve found that there’s no one specific background that defines a successful Agent Product Manager. Our current team includes former founders, product managers, operators, and investors from companies like LinkedIn, Tesla, Salesforce, and General Catalyst.

Rather than focus on specific experiences, we optimize for unique traits. An all-star Agent Product Manager needs to be an intuitive product thinker and strong communicator. They have to be comfortable switching between technical and non-technical audiences while maintaining  an organized approach. On a given day we can be presenting to C-Suite leaders in one moment, and managing the AI roadmap for a publicly traded company the next. It’s a wildly vast range of skills and experiences that can lend itself well to many different types of people.

As a general principle, we optimize for hiring smart people, giving them a ton of agency and resourcing, then getting out of their way to let them do what they do best - solve problems. 

To account for this, we intentionally bring on people with deep “spikes” in different areas, whether that’s product thinking, customer strategy, systems design, or technical implementation. And our team is structured to take advantage of those differences. In practice, that means leaning on each other when needed, and taking full ownership when your expertise is required.

Optionality at the edge of AI

One of the most unique parts of this role is the optionality it creates from a career perspective. Because the work spans so many domains, the Agent Product Manager role is a natural launchpad for a range of paths.

Since members of our team live at the intersection of Engineering, Operations and Customer Management, we get to hone our skills across nearly every facet of the business at large. A good Agent Product Manager can jump into any problem and instantly be one of the most effective people in the room thanks to their broader context of our business.

Our team doesn’t define success as staying in the same lane. Rather, we value and hire  for curiosity, adaptability, and depth. That means that when someone joins our team, we’re also committed to investing in helping them grow across every skillset, regardless of their previous experience. 

What comes next

Before joining Decagon, I’d worked across multiple functions, including Product, Sales, and Finance. I’d built teams, launched products, and founded a company. But from my first day, it was immediately clear this role stands apart. It sits at the frontier of how AI is being deployed in the real world, not as a research project, but as a core part of how businesses operate.

The Agent Product Manager function is a cornerstone of our company, and we’re actively growing each day. If you’re excited by high-ownership work and building something new we’d love to hear from you.