Thomas Tran

Founding Designer building MongoDB’s Agent Platform

Building the platform from the ground up with 15 engineers and 2 PMs

MongoDB Agent Platform — project overview

The problem

Companies can build AI agents. They can’t ship them.

A demo agent stalls before going live on three things: security & governance, memory & context, and vendor lock-in.

So the agent never ships — this is the missing layer we’re building.

Config source of truth decision doc
Build and deploy design proposal

01

Shaping direction

Creating clarity and alignment across engineering, product, and design to define what to build and why.

02

Simplifying complex systems

Turning complex workflows into intuitive experiences across multiple tools and touchpoints.

First-time user experience journey map
Agentic CLI — auth login and deploy

04

Designing end-to-end platform experiences

Creating a cohesive experience from CLI onboarding to cloud management UI.

05

Bridging design, AI & code

Using design, AI, and code to move from idea to production.

Cursor planning and implementation workflow
GitHub pull requests shipped with AI and code

03

Humanizing enterprise software

Making powerful enterprise platforms feel simple, approachable, and thoughtfully crafted.

Monitoring dashboard
Coverage monitoring view
Agent workspace detail

An app I built and launched to help managers prepare 1:1 conversations, grow their people, and build lasting habits. Designed and shipped in 8 weekends.

florish.me ↗

Florish landing page

01

Feedback & coaching captured automatically

Florish surfaces feedback and recognition moments from your 1:1 notes, and helps you say the hard things well.

Florish 1:1 dashboard with Ada

02

Frame difficult conversations

Not sure how to say it? Florish helps phrase a sensitive message clearly and respectfully, using AI.

Help me say it — input and suggestion modals
Early AI-generated build — Mentra
Intermediate dashboard iteration

03 – HOW I BUILT IT

From rough build to a real design system

I iterated through ChatGPT, Lovable, Figma and Cursor — from a first AI-generated build to refined components and tokens.

Refining the flows and visual design in Figma

Figma file — flows and visual design

Set up colors, tokens, and reusable components in Figma

Design system — tokens, typography, components

04

Next steps

This was meant as an experiment for me to design, build and ship a product end-to-end. While I have no plans to market it, I still want to further improve it for personal use and learning purposes.

Changing the way we rent, buy and sell homes — centralizing the agent’s entire transaction into one place: Deals.

Compass Deals — deal detail on web and iOS

01 – The problems

Their workflow lived on paper

Before Deals, agents tracked listings across folders, notebooks and a wall-sized whiteboard — nothing connected, everything manual.

Handwritten listing board
Paper deal folders
Compass Deals dashboard on web and iOS

02 – The solution · Deals

One home for every transaction

Centralize all information related to a deal, and create magical moments where the tedious parts get automated — on iOS, web and desktop.

03 – Create a deal

From listing to active in a few taps

A guided flow turns a new listing into a live deal — pick a type, add the details, and it’s ready to work.

Create a deal flow across four iOS screens

04 –On the go

The whole deal in your pocket

Comps, listings and insights for every deal — the same source of truth, now on iOS for agents in the field.

Compass deal detail across three iOS screens