Building an AI-Native Investment Firm

We spend a great deal of time pressing founders on whether they are genuinely AI-native or merely AI-adjacent. It would be incoherent to ask that question of others and not of ourselves. So this is an account of how we are rebuilding our own decision loop around AI — what has worked, what has not, and where we have deliberately kept a human firmly in the chair.

An investment firm is, at bottom, a decision engine. It ingests information, forms judgments, allocates capital, and learns — slowly and imperfectly — from outcomes. That description should make any honest practitioner uncomfortable, because the loop is leaky at every stage: information is missed, judgments are inconsistent, and the feedback from outcomes arrives years late and heavily confounded. AI does not fix all of this. It does meaningfully tighten parts of it.

Where intelligence earns its place

The clearest wins are in coverage and consistency. No human team reads everything relevant; an AI-native research layer can monitor a far wider surface — filings, on-chain activity, market movements, the primary sources behind a narrative — and surface what deserves attention. The point is not to replace judgment but to ensure judgment is applied to a more complete picture rather than to whatever happened to cross a partner's desk.

The second win is consistency of process. Diligence questions get asked the same way every time; comparable companies get compared on the same axes; the analysis of a deal in month one is retrievable, structured, and testable against what actually happened by month thirty-six. This is the operational-data discipline we preach to portfolio companies, turned on ourselves: capture every decision and its outcome in a form that can teach the next decision.

AI widened what we can see and made our process consistent. It did not, and should not, make the decision.

Where the human stays

We are deliberate about the boundary. The allocation decision itself — the judgment to commit capital and the conviction to hold it through drawdowns — stays human, because it rests on things the system handles poorly: reading a founder, weighing a downside that has never occurred before, and taking responsibility for being wrong. A system optimised on past outcomes is structurally blind to the genuinely novel, which in venture and emerging markets is precisely where the returns live.

• 01Coverage and monitoring: delegated to the system, supervised by humans.
• 02Structuring and consistency checks: system-assisted, human-owned.
• 03The conviction to commit and to hold through stress: human, full stop.
• 04Accountability for outcomes: human, and never laundered through a model.

The honest scorecard

Not everything has worked. Early attempts to over-automate produced confident summaries that flattened exactly the nuance a good analyst exists to catch. We learned that the value is in widening the aperture and enforcing consistency, not in shortcutting the thinking. An AI layer that lets you skip the hard reasoning is not an asset; it is a liability with good production values.

The firms we most want to back are the ones doing this honestly inside their own walls — using intelligence to see more and decide more consistently, while guarding the irreducibly human core of judgment and accountability. We hold ourselves to that standard not as a marketing posture but because we believe it is, increasingly, the standard. The investment firms that compound over the next decade will be the ones that rebuilt their decision loop deliberately — and were equally deliberate about what they refused to automate.