Introducing Agent Development: How Decagon is redefining forward deployment

Forward deployed teams have quickly become standard at modern AI companies. Nearly all have one, bridging customer needs and product.

Traditionally, these teams have hired for drive, technical ability, and customer empathy. The archetype is the hero who embeds, hustles, and does whatever it takes for the customer. This is still true, but as these teams scale, another important trait emerges: the systems thinker.

I've watched that shift up close. I joined Decagon when we were around 20 people, as a founding Agent PM. Today, we’re near 500 employees with an ever-growing forward deployed team. Here's what I’ve learned about what makes one succeed.

Forward deployment as a bottleneck

Most forward deployed teams start the way Decagon’s did. In my first months, the simplest description of the role was as an orchestrator: I sat between engineering, product, and our customers, turning what a customer needed into engineering tasks or explaining the product's limits back to them.

Early on, it was obvious this couldn't scale. Every customer generated an endless stream of requests that all routed through engineering, and customers couldn't reach what they needed on their own. The work was almost entirely custom: wiring, workflows, edge cases, handled one customer at a time.

At ten customers, this barely matters; you can brute-force every launch. At a hypergrowth company like Decagon, a wave of demand was building: the kind every company wants and the kind that breaks a team built on brute force. The next phase of growth wouldn't come from more people. It would come from better systems.

Forward deployment as systems design

In software, systems design means architecting an application to hold up as load grows. The best forward deployed teams treat their craft the same way. A delivery motion should be designed to absorb increasingly complex customers and improve with every launch.

Systems thinking starts with the customer’s agent. AI gives you near-infinite room to build; you can configure the same outcome a dozen ways, and most of them clear the bar. The scarce skills are restraint and foresight: seeing the second-order effects and choosing the design that lasts over the one that looks most complex or sophisticated. The failure mode is everywhere right now: agent builds that decay into a black box of prompts and patches no one can safely touch. A systems designer builds the agent so the customer can still understand and own it a year later.

The same instinct widens out from there. A systems designer builds the forward deployed team into a machine, where everyone draws on the same playbook and adds to it, so one person's learning becomes everyone's. They design how the whole motion connects to the rest of the company: how it partners with engineering, feeds the product roadmap, and advises sales strategy.

How we're building it: Agent Development

A little over six months ago, we built the Agent Development team for exactly this: a team ruthlessly focused on the zero to one motion of getting agents into production, and on making that motion compound.

One of our first projects was less about any single customer than about all of them. We mapped every place the work was needlessly custom, the edge cases and gaps that forced bespoke engineering into each build, then partnered with product and engineering to turn that analysis into a self-serve roadmap. 

It cut the custom engineering work per agent by 80%. Reworking a customer's promotional policy used to take an engineer a full sprint of custom code; now an ADM designs, debugs, and tests it in plain English in an afternoon. Agent Development Managers now own almost all of the build work that once required engineers, and customers feel it: they go live faster, they're happier, and they have far more visibility into what they're building.

In a world where traditional product moats get copied within days, the delivery work that was once our bottleneck is becoming a new moat.

Come build with us

The role today comes down to two things:

• Nail zero to one: build agents customers love and be the team they want to work with, owning the outcome until the agent is live and resolving real business issues.
• Compound zero to one: make the process easier every time, turning every manual step you take once into something the next customer never repeats.

The people who thrive in this role are great problem solvers with strong business intuition. They're data-driven, and they instinctively ask “and then what?”, tracing how one decision ripples downstream. Above all, they're obsessed with building beautiful systems that last. We hire for these instincts, and they show up the same whether someone has designed software, an org, a process, or an agent.

We believe this is one of the most important problems for AI companies today, and we're building the best team in the world to solve it, here at Decagon.

Interested in shaping the future of agent development? Explore careers at Decagon.

‍