Investment thesis · early-stage AI agents

The Founder-First Scorecard

At pre-seed you back the people, not the slide. The verdict leans ~two-thirds founder, one-third thesis — and a weak founder can't be rescued by a clean thesis. Two gates decide whether it's on-thesis at all.

Deal

Gates

fail one → off-thesis

G1Stays off the record

Sits on top of the incumbent's core tools — the customer keeps their ledger, CRM, helpdesk. Fail if it requires replacing the system of record. Category conviction; applies at any stage.

G2Path to its own record

A pre-seed test of trajectory, not today's product: is there a credible path to owning a new record layer — action history, cross-system context, an audit of what was done? Fail a founder building toward a permanent stateless wrapper.

Founder conviction

Primary0–5 · weighted

Thesis fit

The lens0–5 · weighted

The bet, in one picture

Action & intelligence layer

orchestrates · remembers · executes

LEDGER

SPEND

PLANNING

BANKS·SHEETS

Verdict

0.0 / 5

Founder

—

Thesis

—

Unscored

Set both gates, then score the founder and the thesis.

Composite = 0.65 × founder + 0.35 × thesis · founder ≤ 2 caps the verdict at Watch.
Millberg Ventures. — a private decision tool.