Backing the AI build-out, at every stage.
Milton Maxwell invests across the full life of a company: angel checks and fund LP positions at the earliest stage, private growth-stage names, and a concentrated public-equity book. We're in San Francisco, blocks from OpenAI and Anthropic, and near many of the public companies we follow.
Close to founders, early.
Angel checks into operators we know, and LP positions in a select set of startup funds. We get to know most of these founders at SF Patchworks, our incubator, years before their companies reach the public markets.
The companies defining this cycle, still private.
We research them closely and invest where we have both conviction and access.
Long the frontier, short the incumbents they're displacing.
A concentrated long/short book.
AI in the investment process.
A methodology, not a separate strategy. The same process runs underneath everything above, from the earliest checks to the public-equity book.
- 01
Ingests unstructured data.
Earnings transcripts, SEC filings, podcasts, social signals, web traffic, ad spend, credit-card aggregates, news and trade press. Most of the information that moves a stock arrives as text or speech, not numbers; large language models let us turn that flood into something we can search, compare, and act on.
- 02
Extracts and structures signals.
Each piece of unstructured input gets reduced to the things that actually matter: revenue drivers, unit economics, management commentary that changed from the last quarter, technical milestones, competitive dynamics. The result is a structured view we can rank and revisit.
- 03
Surfaces comparables and analogs.
Across thousands of companies and decades of filings, finding the right precedent used to be a senior-analyst skill: the historical analog for a margin trajectory, a regulatory event, a competitive dynamic. AI makes it routine.
- 04
Monitors positions continuously.
Conditions that would matter to a held position (a competitor's earnings call, an SEC enforcement action, a supply-chain disclosure) are flagged the moment they arrive, not at the next quarterly review.
AI hasn't replaced judgment. It's compressed the wait between question and answer from days to minutes, so we ask more questions and explore more theses than we otherwise could. The bottleneck moves from research throughput to research quality, selecting the right questions, sizing positions with discipline, and being honest when the data says we were wrong.
The pipeline runs on a small set of internal services we maintain. They exist to serve our own research, not as a product.