Case study · June to October 2026
Vantage: founders and investors in the same workspace
A complete product I designed and built in four months by leading a team of AI agents. I started from a real problem, wrote a clear spec, made the hard calls, and kept checking the product until it was right.
One click as a founder or an investor. No sign-up.
4
months, start to finish
600+
reviewed changes
3
sides: founder, investor, and admin
7
AI agents, each with one job
The problem
Reporting that wastes everyone's time.
Most startups report to their investors through emails, PDFs, and spreadsheets. Every month the founder rebuilds the same numbers, nobody can tell who read what, and every investor gets the same view, even though some should see the full financials and others only the headlines.
Investors have the mirror problem: a dozen companies, a dozen formats, and no single place to see which ones need attention.
What it does
Built for both sides of the table.
Each investor sees their own slice
The founder grants each investor full, guest, or custom access to financials, hiring, and projects.
What you sent stays sent
A published update keeps its figures. Correct a number later and the dashboard changes, but the update investors read does not.
Reporting is graded
Each company's grade drops as updates run late. Only a published update restores it; a saved draft doesn't.
Trouble, in sentences
Fixed rules decide what's flagged as a risk. A plain-language explanation follows, and the flag stands even if that writing fails.
The whole portfolio
Holdings compared across eleven sortable columns, a shared calendar, and capital calls split across each LP's commitment.
AI where it helps
Seven AI features, from drafting the monthly update to meeting prep. The public demo shows prepared answers, so nothing pretends to be live.
How I built it
I led it the way I'd lead a product team.
I'm not a career engineer, and I didn't write the code by hand. I set the direction, made the decisions, and held a high bar. The work started in Google Antigravity and moved to Claude Code.
01
I wrote the spec first
A living product specification that every feature and screen had to match. It's public.
02
I built an AI team with roles
Product manager, marketing, UX, graphic design, a developer, a senior reviewer, and a test user playing a hiring executive. Only the developer changed code.
03
I tested it myself
I used every screen as a founder and as an investor, and caught many real bugs just by looking closely.
04
I set the brand
The name, the wordmark, a strict, readable color palette, and a rule that every claim must match the product.
Hard calls
The decisions that shaped it.
Honesty over impressiveness
A reviewer who catches one inflated claim stops trusting the rest. Every number on the homepage was checked against the product, and the demo openly lists what's switched off.
Free for anyone to try
I had it rebuilt to run entirely in the visitor's browser: no server, no database to pay for, and no sign-up. Each visitor gets a private copy of the demo data.
Same look, new foundations
That rebuild replaced how the whole app runs underneath. Every page was compared against the original before it was accepted.
What I learned
AI is a team, not a tool.
Results improved most with clear roles, a record of what had already failed, and agents reviewing each other's work.
Check everything.
AI answers can sound confident and still be wrong. The habit that mattered most was checking claims against the actual screen.
The spec is the product.
When the specification was clear, the build went fast. When it was vague, no amount of AI fixed it.