Esther Suebsubanant · Portfolio

How I workThe same loop I used on deals

The loop I ran on every deal is the loop I run on every product. Only the last step changed.

Ask the question nobody asked
In a founder meeting it was the question they hadn't rehearsed. In my RDI study it was why AI tools that executives bought never reached the people doing the work.
Write down what would prove me wrong
A deal memo is a bet with a kill condition. So is a product spec. I decide what the evidence has to show before I go looking.
Go find the part that hurts
Across 30+ cross-border deals, the story that mattered was rarely in the pitch deck. I look hardest at whatever weakens the case.
Then build it
As an investor, an idea that survived got a check. Now it gets a prototype.

Selected workFour builds and the calls behind them

Field research · Stanford RDI

Why AI stalls inside companies

Companies keep buying AI and the people doing the work keep not using it. I wanted to know where it breaks.

What I did: About 40 interviews with people building and deploying AI inside companies, including Databricks teams, vertical SaaS founders and a16z-backed startups. I built agents on a VPS to run the outreach, scheduling and notes.
What I found: AI enters from the top, through the CEO or a vendor selling to the C-suite, and stalls with the staff. Nobody I spoke to had a working fix.
Where it led: You can't plug AI into an organization. You either adapt the AI to the people or redesign the organization around it, and the real gains come from the redesign. I'm now writing the implementation playbook for that.

Read the full study

Bought at the top. Stuck at the desk.

Running daily · Personal build

Jarvis

A voice-first command center I talk to from my phone, built on my own Hermes agent.

The decisions: One Telegram group as the only interface, so there is nothing new to open. Specialist agents for my morning brief, recruiting tracker, meeting follow-ups and club operations. Hosted in the cloud instead of on my laptop so it never sleeps. Spoken replies through free-tier text to speech, because the whole system had to cost nothing to run.

How it's built

Stanford x Anthropic Hackathon · 2nd runner-up

Anchor

A WhatsApp-native AI companion for Alzheimer's caregivers.

What happened: Placed 2nd runner-up.

See the demo

In progress · CS 224V with Kenaj Washington

DealRoom

An AI analyst that works through an M&A data room and answers due diligence questions.

Why I'm building it: I've sat in enough data rooms to know which questions decide a deal and how long it takes to answer them by hand.

Follow the build

This is the tool I wanted on every deal I worked.

AI RadarMy weekly read on the AI stack

Every week I track the three things that decide what AI can actually get built.

Constraint 01

Compute

Chips, cloud capacity and what it costs to rent them.

Constraint 02

Capital

Who is funding what, at which price, and the rules that decide where money can go.

Constraint 03

Energy

Power deals, turbine queues and grid prices that decide where data centers can switch on.

Each price, deal, rule and power deal on the board is tagged by the constraint it moves and which way, with a link to the source. Founders, investors and operators use it to see what changed and why it matters.

Open AI Radar

ProofOne example per skill, or it's not listed

Customer empathy: About 40 interviews on how AI really gets adopted, and a finding nobody had a fix for. RDI study
Execution: Forty rounds of outreach, scheduling and notes, run by agents I built instead of by hand. RDI study
Technical fluency: A multi-agent system on my own Hermes agent, running every day. Jarvis
Business judgment: Built Thai Union's corporate venture function from zero under a $30M mandate, and ran McKinsey and Bain engagements from the client side.
Communication: AI Radar, a weekly board that tags every price, deal and rule by which constraint on AI it moves. AI Radar