Myrto Lalacos recently asked whether AI that enters a venture firm as an analyst will ultimately replace the general partner. It is a useful question because it exposes three different kinds of work that are often compressed into one title: producing research, exercising investment judgment, and carrying institutional accountability.

AI can take on more of the first category and support the second. That does not automatically transfer the third.

The replacement question hides three different jobs

An analyst gathers evidence, structures dealflow, compares companies, prepares meetings, and maintains follow-up. A partner also interprets incomplete evidence, changes the thesis, allocates scarce capital, manages LP and founder relationships, and accepts the consequences of a decision.

These responsibilities overlap, but they are not a ladder of the same task. Faster research does not become fiduciary accountability simply because the system has processed more companies.

A better question is therefore: which decisions can the system prepare, which judgments can it challenge, and which authority should remain with named people?

Where AI changes analyst work

A well-bounded AI analyst can continuously handle work that is expensive mainly because it is repeated:

resolve company identities and remove duplicates;
collect current sources against the fund's thesis;
separate company claims, third-party facts, and analyst inference;
surface contradictions and stale evidence;
compare a new opportunity with prior decisions;
prepare meeting questions and follow-up drafts;
monitor approved signals and revisit conditions.

This can compress the path from a deck or URL to a reviewable case. It can also make the work more consistent because the evidence and rules remain visible.

But a system doing these tasks is replacing coordination and research effort, not necessarily replacing a person. The investment team still decides which evidence matters and what to do with it.

A GP is not simply a more experienced analyst

General partners do more than score opportunities. They decide when the written thesis is wrong, weigh a concentrated portfolio, negotiate ownership, manage reserves, earn founder trust, explain decisions to LPs, and operate through conflicts that cannot be resolved by adding more public data.

They also carry consequences. A model can produce a recommendation. It does not independently bear the reputational, legal, relational, or financial responsibility attached to the fund's action.

That distinction does not make partner judgment mystical or immune to scrutiny. It means the system should make judgment easier to challenge without pretending accountability has moved to the model.

The compounding asset is fund-owned decision memory

The strongest AI teammate may not be the one exposed to the most funds. Cross-fund experience can produce broad patterns, but a fund's edge is often hidden inside its private history:

why the team passed despite an attractive market;
when it broke its own thesis and why;
which partner identified a non-obvious risk or opportunity;
which evidence later proved unreliable;
what a founder conversation changed;
which revisit condition eventually mattered;
where the team was confidently wrong.

This is decision memory, not a folder of investment memos. It connects the evidence available at the time, the disagreement, the chosen action, and what happened next. When that history remains owned by the fund, an AI analyst can use it to question consistency without flattening the firm's judgment into a generic market score.

What the AI teammate should remember

A useful memory record does not need every meeting transcript. It needs the parts that can improve a later decision:

the opportunity and evidence snapshot;
the relevant thesis version;
the advance, hold, pass, or revisit rationale;
material disagreement and unresolved uncertainty;
corrections made by the team;
the authorized next action;
later evidence and outcome updates.

This structure lets the system ask better questions. Are we applying the same standard as last quarter? Are we treating a missing fact as negative evidence? Did we previously override this rule? Has the condition behind a pass changed?

The goal is not to preserve one partner's intuition as permanent truth. It is to make the fund's evolving judgment inspectable.

Authority should be earned by task, not granted by title

Calling a system an AI GP creates the wrong operating debate. Authority should expand task by task, based on evidence quality, correction history, reversibility, and consequence.

A fund may allow an agent to update an internal research field when the source and identity are clear. It may allow continuous monitoring within approved sources. It may let the agent prepare a meeting brief or draft a founder email.

The same fund can still require a partner to approve outreach, change a deal stage, resolve a portfolio conflict, communicate a pass, or commit capital. Approval to research is not approval to act. Strong performance in one task does not silently expand authority in another.

A human-led fund can still be AI-native

The near-term winner is unlikely to be defined only by whether it employs fewer analysts or gives a model a partner title. It may be the fund that builds the clearest division of work.

AI keeps evidence current, tests consistency, prepares decisions, and compounds the firm's decision memory. Human partners change the thesis, own relationships, resolve consequential ambiguity, allocate capital, and carry the outcome.

That is not a temporary compromise on the way to full replacement. It is an operating model worth designing deliberately. The AI compounds the process. The fund owns the edge and the responsibility.

Will AI Replace the VC Analyst—or Eventually the General Partner?