Work

forward deployed

Gumloop

Forward-deployed engineering for a creator marketplace

Gumloop·YC-backed · $23M+ raised
MarketplaceSEOPlatform ArchitectureForward Deployed

Full-stack marketplace build — creator publishing, SEO architecture, scalable template infrastructure. One month, forward deployed.

The Problem

Gumloop is a no-code AI automation platform. Its community had built thousands of workflows — sophisticated automations that non-technical users could configure and run. But these lived behind the login wall. There was no public-facing way to discover, preview, or use a community-created workflow.

The founders, Max and Rahul, wanted a public marketplace: a place where creators could publish their automations, new users could discover Gumloop through search, and the template ecosystem could become a growth channel rather than an internal catalog.

This wasn't a design-then-build engagement. We flew to San Francisco, sat in Gumloop's office for a week, gathered context directly from their product and engineering teams, then sprinted for a month.

Week 0Fly to SF
meet the teamgather context
Week 1Embedded Design
whiteboard sessionsdata modelcomponent inventory
Weeks 2–4Sprint
daily syncsshared branchesintegrated deploys
Production
marketplace livehanded off clean

The week in-person transferred more context than a month of Slack threads would have.

Total time from first meeting to production: one month.

Forward deployment compresses context transfer. Design and build in parallel, not in sequence.

The Architecture

Data Model

The first task wasn't code — it was figuring out what a "published template" actually is. Gumloop has three entity types: workflows (multi-step automations), flows (simpler sequences), and agents (autonomous AI workers). Rather than building three separate publishing systems, we designed a unified template abstraction that wraps any entity type with publishing metadata: title, description, category tags, input parameters, creator attribution, and usage stats. Every pre-built and internally published template is automatically compatible — the abstraction extends the existing system rather than replacing it.

Creator Publishing

The publishing flow needed to be low-friction for creators but maintain quality. When a creator publishes a workflow, the system auto-generates a description using an LLM that analyzes the workflow's structure, node types, and connections. The creator can edit this, add context, and categorize the template.

The key constraint: creators should never have to leave the editor to publish. The publishing interface is embedded in the existing workflow builder — you build, you publish, you're done. The LLM-generated descriptions turned publishing from a 15-minute writing task into a 2-minute review-and-edit, and creator publishing rates increased significantly once the friction dropped.

SEO Architecture

Each published template gets a dedicated, server-rendered page optimized for search — built for Gumloop users browsing templates and for people who've never heard of Gumloop discovering it through a search like "automate Gmail summarization." Existing templates are rendered at build time; new publications render on-demand. The URL structure is flat and descriptive: /marketplace/gmail-summarizer, /agents/research-assistant.

How We Worked

This project was forward deployed in a way that's unusual for an external team. We weren't working from a spec that was handed to us — we sat with Gumloop's designers and engineers for a week, mapped the full user journey on a whiteboard, iterated on wireframes until the data model was settled, then sprinted on code.

The week in-person wasn't a kickoff meeting. It was the design phase. By the time we flew home, we had a finalized data model, a component inventory, and a clear build plan. The remaining three weeks were execution — daily syncs, shared branches, and integrated deploys to Gumloop's existing infrastructure.

We wrote code that Gumloop's own engineers could maintain and extend without us. The goal was to leave infrastructure, not create dependency.

What We Learned

Forward deployment compresses context transfer. The hardest part of any external engagement is understanding the client's system — not just the codebase, but the decisions behind it, the constraints that aren't documented, the things the team tried and abandoned. Sitting in the office for a week transferred more context than a month of Slack threads would have.

Extend, don't replace. The marketplace needed to feel like a natural extension of Gumloop's existing product, not a separate app. That meant building on their authentication system, their data layer, and their design language.

Reduce friction, increase adoption. Auto-generated descriptions turned a 15-minute publishing task into a 2-minute review. That single change moved publishing from "something creators intended to do" to "something creators actually did."

Outcome

Gumloop Templates launched as a public marketplace supporting workflows, flows, and agents — transforming an internal catalog into a growth engine with community-driven content, organic search visibility, and the foundation for a creator economy within the platform. One month from first meeting to production.

Rubric gave us the tactical engineering firepower we needed as we rapidly scaled. Professional executors who came in, crushed the task and handed it off gracefully.

Max Brodeur-Urbas, CEO of Gumloop