Key Takeaways

  • Ramp built an internal AI agent called Glass that connects directly to Snowflake data, customer research repositories, design systems, and the production codebase.
  • PMs at Ramp no longer interrupt technical leads to ask if a feature is feasible or if it will break existing architecture; they query Glass directly.
  • Glass produces working prototypes that run inside Ramp's actual software rather than static Figma mockups or abstract documents.
  • Engineers reject standalone specs and isolated prototypes; they require verified data, coding-agent-ready requirements, and interactive functional prototypes delivered together.

Stop Giving Engineers Raw Specs and Blank Mockups

Most product teams treat AI like an upgraded text editor. A product manager opens an external chatbot, writes a vague prompt about a billing feature, and copies the resulting five-page document into a ticket. The engineering team ignores it. A generic prompt cannot anticipate edge cases, database constraints, or existing user behavior.

Geoff Charles, Chief Product Officer at Ramp, watched this disconnect slow down product shipping. In an organization running at breakneck speed, passing text files back and forth created friction. Engineers did not want wordy descriptions, nor did they want non-functional design files that failed to reflect actual system logic.

“Your engineering team don't want any of these individual things,” Charles explained. “They don't want your prototype. It's useless. They don't want your long spec. That's useless. They want the combination of qualitative and quantitative data to convince them that this is a problem. The actual requirements that they can use for their coding agents and a prototype that they can actually get inspired by.”

How Ramp Connects Glass to Production Systems

To bridge the gap between product strategy and engineering execution, Ramp developed Glass. Instead of acting as a standalone chat interface, Glass operates with direct access to Ramp's underlying technical infrastructure.

“We built our own AI agent called Glass. And we gave it all the context it needs to do the work,” Charles said. “It connects to all our systems. So it understands both the data with Snowflake. It understands our user research and it can nail exactly the job to be done that our customers are asking for.”

Because Glass indexes Ramp's live codebase, product strategy documentation, and historical spec formats, it acts as an on-demand technical lead. When a PM outlines a feature, Glass checks whether the proposed change conflicts with existing services.

“Glass also understands our product strategy. It understands how we define product specs. It understands our codebase. And so instead of a PM bothering an engineer asking them, is this possible? Would this break something? What am I missing? You ask AI,” Charles noted. “You can actually define and build a prototype that actually works within your product.”

Glass forms part of a broader internal suite at Ramp, working alongside specialized agents like Inspect, Review Buddy, Testo, and Gadget. By anchoring AI directly to institutional memory and live data pipelines, product managers can validate customer demand and deliver functional code context before an engineer writes a single line.

What to Do With This

Audit your product team's handoff workflow this week. Identify the top three questions your PMs repeatedly ask engineers during sprint planning, such as database schema limitations or API behavior. Hook an internal LLM to your engineering documentation, repository index, and customer feedback tracker so PMs can self-serve technical feasibility checks before writing their next project brief.