Problem
Program management ran entirely on Google Suite: program plans in Sheets, strategy in Docs, quarterly planning in more Sheets. Every artifact was maintained by hand, nothing fed anything else, and there was no source of truth. Leadership couldn't see across the portfolio without someone manually rebuilding the picture each time, which meant decisions consistently lagged the data.
Approach
I designed and built an executive portfolio platform covering 17 concurrent initiatives, each mapped to a swarm team of 3–5 engineers, roughly 50–85 engineers of delivery capacity in a single view.
I shipped it in two deliberate stages:
- Prototype: a Google Apps Script front end with Sheets as the data store, authenticated through Google Auth. Secure, no new infrastructure, and fast enough to prove value and earn approval.
- Production: re-platformed to Next.js and React, containerized in Docker with a real database, hosted on Earnest's existing AWS infrastructure.
The production step required standing up something the company didn't have: internal hosting for internally built applications. Earnest's only infrastructure was external and client-facing, which nobody wanted to touch. I established the internal path, which then became the pattern other teams used to deploy their own AI-built tools. I owned the architecture and hosting decisions end to end.
Impact
- Gave the senior leadership team a live, org-wide view of the portfolio for the first time, replacing manual roll-ups with a single source of truth.
- Surfaced engineer and engineering-manager over-allocation early, swarm teams carrying more than they could absorb, plus cross-program dependencies and risks, so leadership could rebalance capacity before delivery slipped rather than after.
- Established Earnest's first internal application hosting capability, unlocking a new class of internally built (including AI-built) tooling across the org.
The framework, in hindsight
In October 2026 I completed the Harvard Data Science Initiative's AI Strategist and Builder intensive, built around the A.G.E.N.T. operating model (Hofmann and Kruhse-Lehtonen, Harvard Data Science Review): Audit, Gauge, Engineer, Navigate, Track. I didn't have that vocabulary at Earnest, but this program followed the arc almost exactly:
- Audit: mapped how portfolio reporting actually worked: who maintained each Sheet and Doc, what fed what (nothing), and the outcome that mattered, which was leadership decisions keeping pace with delivery.
- Gauge: the manual roll-up was the classic automation sweet spot: repeatable, structured, high impact, and mostly humans compiling and re-checking other humans' spreadsheets.
- Engineer: the stage this write-up covers in detail: redesign the workflow around one data store instead of accelerating the old one, prototype cheaply, then re-platform for production.
- Navigate: leadership kept the decisions. The platform surfaced over-allocation, dependencies, and risk with the reasoning visible, so trust came from transparency rather than a black box.
- Track: measured the outcome, not the output: over-allocation surfaced early so capacity could be rebalanced before delivery slipped, and a source of truth leadership actually used.
The rollout side maps to the companion B.U.I.L.D. framework (Blueprint, Unlock, Iterate, Launch, Drive): spec the production system, make the stack decisions deliberately (including standing up internal hosting the company didn't have), iterate the prototype with the people who used it, prove value in a constrained launch before asking for production investment, then keep measuring. Drive is why the hosting path outlived the dashboard: other teams reused it for their own tools, which is exactly the compounding the framework is after.