Executive Summary

Cofounder (equity) and full-stack engineer of PartnershipTool, an ad-tech business intelligence and partner management platform. With a two-person founding team, I built the product and the agent-driven development system behind it — taking the platform from concept to customer-validated beta and continuing to advance both together.

The Challenge

Company: PartnershipTool — ad-tech BI and partner management platform Stage: Pre-seed / early customer validation Team Size: 2-person founding team (business + technical) Funding: Bootstrapped

The Problem:

• Ship a credible beta to early customers with a single engineer
• Make architecture and stack decisions that survive from MVP to scale
• Keep infrastructure costs near zero during validation
• Cover product engineering, infrastructure, and AI features simultaneously

The Approach: Agent-driven development. One engineer running a disciplined agentic workflow covers ground that used to take a team — with the verification discipline to trust the output.

The Agentic Development System

This project is where my agentic engineering practice runs at full depth. The development system is a first-class product alongside the platform itself.

Current workflow:

• Harness: Cursor + Claude as the core development harness
• Agent memory: Self-hosted Honcho, moving past ad-hoc memory toward durable agent context
• Agent team: Hermes orchestrating semi-autonomous agents that behave like team members, not CLI tools
• Supporting tooling: GitHub apps and actions, code review automation, MCP servers, custom agents and skills

How it evolved: The workflow started as AI-assisted coding and matured into longer-running autonomous sessions. As session length grew, review burden became the ceiling — which pushed the practice toward better knowledge architecture, agent memory, and verification workflows. That progression is the core of my agentic engineering practice today.

Verification discipline: Agent output is held to the same standard as human output: typed interfaces, review gates, integration tests, and deployment checklists. Autonomy without verification is just risk at higher speed.

Key Contributions

Product Launch

Beta product enabling partner marketers to manage relationships and act on business intelligence.

• Beta launched on schedule to early customers
• Positive feedback from initial cohort
• No critical technical issues during launch
• Smooth onboarding experience for early users

AI Features in the Product

• AI agents for account configuration using Gemini
• Performance-insight agents surfacing actionable analytics
• Vendor-transaction data product in progress: transforming raw transaction data from different vendors into a common data model

Architecture Decisions

Technology Stack:

• Frontend: Next.js (App Router), TypeScript 5, Tailwind CSS + shadcn/ui
• Backend: Vercel Edge Functions, Neon (PostgreSQL)
• Payments & Analytics: Stripe, PostHog

Decision Framework:

1. Time-to-market: Prioritize speed to beta
2. Cost efficiency: Minimize fixed costs during validation
3. Scalability: Easy path from MVP to production scale
4. Developer experience: Enable fast iteration — human and agent alike
5. Ecosystem maturity: Avoid bleeding-edge risk

Key Trade-offs:

• Chose: Vercel over AWS (faster deployment, simpler ops)
• Chose: Neon over RDS (serverless scaling, lower minimum cost)
• Chose: Next.js full-stack over separate frontend/backend (faster development)
• Deferred: Kubernetes, microservices (premature for MVP)

Lessons Learned

What worked well:

• Small-team leverage is real: product velocity that previously required multiple engineers
• Agent memory and knowledge architecture pay compounding dividends as the codebase grows
• Verification workflows caught agent errors early and built trust in longer autonomous sessions
• TypeScript caught bugs early — for humans and agents

What I’d do differently:

• Ship alpha even earlier; more customer interviews before building
• Start architecture docs from day one — they double as agent context
• Set up automated testing earlier; more end-to-end tests for critical paths

What This Demonstrates

Cofounder-level ownership: Concept to customer-validated product with a two-person team — architecture, product, infrastructure, and AI features all owned end to end.

Agentic engineering in production: Not tool adoption — a working development system with harnesses, agent memory, orchestration, and verification, applied to real shipped software.

Modern full-stack execution: Next.js, TypeScript, PostgreSQL, Stripe, PostHog, with AI features built in from the start.

Engagement Details

Duration: September 2025 - Present (ongoing) Role: Cofounder (equity) & Full-Stack Engineer Status: Active — product and data-product development continuing