How to Build the Foundations for Agentic Development

GlossGenius VP of Engineering Brady Allchin talks with Stephen Poletto about the foundation his team built for agentic development, why it matters more than what tool or model you’re using, and how it keeps spend in check and reshapes the engineer's role as coding compresses. An audio version of this interview is available on Spotify.

Three Takeaways

• The foundation is the lever, not the model or tool. What makes an internal agent like GlossGenius's Geni work is the layer underneath: an architecture that doesn't waste context, a codebase documented for agents to navigate, and training that aligns engineers on shared workflows.

• The guardrails that matter are set by engineering. GlossGenius runs an autonomous coding agent aggressively, enough to show up on an Anthropic leaderboard, and still hasn't hit its guardrail on internal development. What keeps spend in line is the foundation its engineers built.

• Engineering's value is shifting from writing code to owning outcomes. As coding compresses, the work moves to the edges: deciding what to build, and checking that it led to the desired outcomes.

Introduction

It's tempting to put most of your energy into deciding which tools and models to use when it comes to agentic development. Instead, GlossGenius spent months building the foundation that would effectively support agents before it built its own, called Geni.

Just three weeks into production, Geni had all but cleared a vulnerability backlog the team had never found time for. The heavy usage earned GlossGenius a spot on an Anthropic leaderboard, but it still hasn't come with the runaway bill you'd expect. When asked why, Brady Allchin, GlossGenius's VP of Engineering, states that when you lay the right foundations and set the right guardrails, the payoff spreads beyond the engineering team.

The Foundation Is the Lever, Not the Tool or Model

Allchin argues that the real leverage isn't what tools or models you're using. Rather, it's significantly more important to build a solid foundation upon which agents can operate effectively.

GlossGenius's engineering organization spent months documenting its codebase hierarchically in markdown files across more than 80 repos. During that time, it also published around 75 skills in a shared internal marketplace and trained engineers to align on common workflows. The result is that if you're working in those repos with those standardized skills, you're already operating inside guardrails set by engineering. Teams that struggle tend to do the reverse, handing out powerful tools with no pre-existing foundation.

None of that scaffolding appeared because someone mandated it. About a year ago, GlossGenius gave engineers Ramp cards and told them to go try the tools, then watched for the power users. Leadership then pulled those power users into a virtual squad called Power Up, cleared their calendars, and gave them one goal: establish the best practices for agentic development that can be rolled out across the org. After about two months of dogfooding, they rolled it out team by team, run like on-call training, and the adoption curve went vertical.

What It Means to Lead Engineers Now

As GlossGenius dives deeper into agentic development, Allchin keeps coming back to one principle: engineers were hired to be problem solvers, and now they have tools that free them up to do more impactful work.

He frames it as a funnel. Engineering used to sit in the middle, writing the code, with product at the top and QA at the bottom. As coding compresses, the middle shrinks and the value moves to the edges: building and planning at the top, validating that the work landed at the bottom. The job becomes less about producing code and more about driving outcomes.

This transition is reshaping how Allchin also thinks about team structure. On Stack Trace, Allchin admitted that he used to sort engineers into archetypes: the zero-to-one builder, the scalability expert, the investigator. But he found it nearly impossible to compare at review time.

Now he expects something more like strata: an infrastructure org that owns scale, reliability, and the guardrails, with builders working closer to the customer on top. 

What This Means for Leaders

It's easy to read this as a story about one clever agent, or a smart way to manage spend. The biggest takeaway, however, is about what building a strong foundation for agents can do for your org. GlossGenius can run agents hard without a runaway bill because its engineers own the guardrails, built into the codebase and the architecture, instead of waiting for finance to set a ceiling. A guardrail from finance caps what a team is allowed to do. A guardrail from engineering is a foundation that lets it move. 

The tools are powerful, but the teams that pull ahead aren't the ones with the best tools. They're the ones that built the foundation around them, and own it.

Watch the full conversation on YouTube. Listen on Spotify.