Vic Singh

The through-line

Vic Singh's recurring argument is that venture capital has drifted away from its own craft, and that the way back is through the very technology the industry funds. He traces the thought to a memo he wrote in 2023 called "the operating system for venture capital" [1]. His diagnosis: an industry that was once "a cottage industry like a thousand GPs in all of North America" has seen real democratization, but also "a lot of commoditization. I think firms got quite bloated and I think we sometimes forget the craft of the business which is GPs working with founders under toughest problems" [1]. The resolution he keeps returning to is deliberately paradoxical: "ironically, I figured that we could return to the craft of venture capital by actually using cutting edge technology" [1], since "strangely our business invests in like the most like cutting edge tech companies and we're kind of a file cabinet business" [1].

Underneath that sits a personal arc he tells openly. In the summer of 2009, sitting in Madison Square Park after startups that "did not turn out to be big outcomes," he and a peer concluded that as former VCs turned founders they "had no usable skills to offer anybody" [1]. His framing of the present moment inverts that: "VCs are builders again" [1]. He is now simultaneously a GP at RRE Ventures and co-founder and CEO of Originalis AI, and he insists the two are inseparable: "to build this startup you actually have to actively be investing" [1].

On why venture is a hard product to build for

Singh is blunt that the domain resists productization. Venture is "the most like unstructured" business, defined by "unstructured data, unstructured and non-repeatable processes" [1]. He illustrates with the Monday partner meeting, where the agenda might be "the CTO wait, he needs to get fired or like the company's running out of money or like there's this hot deal and we have two days" [1]. His second structural observation is that "all of venture is like an edge case," which he names directly as the reason the problem is hard [1]. The prize is correspondingly large in difficulty terms rather than market terms: "if you could like productize like 80% or so, like you solved a really hard problem" [1].

He is equally candid that the market itself is unattractive by his own investing standards. Having run the company's own deck through the platform, he puts the TAM at roughly two billion with the planned modules, and says plainly that "as a VC I'm like, 'This is not an attractive TAM'" [1]. What makes it worth doing is the extension beyond venture into the rest of the private capital stack, plus a personal motive he states without dressing up: "I wanted to be a founder. I wanted to build again... I'm in my 40s now. Like this my last act" [1].

On sequencing the product: analysis and diligence before sourcing

Singh breaks the GP job into "seeing, picking, winning, building, and harvesting" and stresses that these are genuinely different jobs [1]. Choosing where to start was, in his telling, the hard strategic call. Sourcing exists in the roadmap, built around what he calls "builder sourcing," but "sourcing will be the last thing that we launch" [1]. He deliberately began with the part that would draw the most resistance: "I actually made a strategic decision to go with what I knew would have the most objection handling, which is analysis and diligence" [1]. He expected pushback of the form "Hey, all right, I do my analysis my way. I have my own taste in this," and treats that as a legitimate view rather than an obstacle to steamroll [1].

The justification is a tension he sees as central to modern dealmaking: "if you don't move fast, you're going to lose the deal. But, if you move fast and don't do the work, you're going to make a bad decision. So, how do you like square that circle, speed and depth, and move really quickly to build conviction?" [1]. Automating diligence deeply, in his account, lets a GP "build conviction like faster with depth" [1]. A side effect he values as much as the speed is visibility inside the firm: instead of a partner quietly working a deal and reporting back, "they're seeing the stuff come in. So like they're getting into like a prepared mind too" [1]. He is realistic about how differently firms actually decide, from Founders Fund's "do not you like infringe on my intellectual freedom" to consensus houses, voting firms and firms that simply say "I trust you like do the deal," and notes the knowledge asymmetry problem where partners comment on spaces they do not hunt in [1].

On network intelligence and why relationship graphs are broken

Singh calls networks "the backbone of the whole thing" and returns to the theme more than any other [1]. His starting point is that existing "who do I know" platforms have struggled because "graphs are pretty broken," and that "who you know is not really your LinkedIn anymore" [1]. His analogy is Facebook's failed attempt to replicate Foursquare: the Foursquare graph worked because it was curated around who you actually wanted to run into in person, while dumping every contact into a system "is not an accurate reflection of who the network actually knows" [1]. He was warned off the problem early, recalling a board member saying "don't do that no one's ever cracked that nut," and answered that "for this thing to really work I have to do it" [1].

His approach layers several signals. He treats the Dunbar number as a real constraint: you cannot remember who is hunting where, and the coffee-meeting regret of "oh man I should have sent you this deal" happens "all the time" [1]. So the system maintains an ontology mapping people to sub-sectors, scores their strength there from activity, and cross-references that against what he calls "the holy grail, which is relationship strength": meeting frequency, email volume, "the depth of that exchange," whether it mattered, "recency, time decay, and all this stuff" [1]. Crucially, expertise is not confined to investors. On a robotics deal, "a founder of a robotics company that took it raised $100 million probably knows more than both of us," so the ranked list surfaces that founder ahead of specialist investors at Eclipse and Lux [1]. The output feeds straight back into work: expert feedback gets "wrapped into diligence notes, updates to memo" [1]. And the same graph serves adjacent jobs, from deciding who to invite when travelling for YC, to filling out a round when "I just need like 500K checks now," to identifying "the proper follow-on investor" for a Series B and replacing "all those shared Google sheets that people make," which he notes people still do [1]. He also thinks the taxonomy problem matters: you cannot look up on Crunchbase or PitchBook which investors are "willing to do weird" stuff, though he says you will be able to in his system, and points to leading a BCI brain computer interface company, Science, as his own example of weird [1].

On scoring, nuance and scaling taste

Singh rejects generic metrics. Analysis in his system runs on a rubric of "team, market, product, traction, mode" with subfactors rolling up to an overall score that updates as primary diligence arrives, drawing automatically on sources like Granola notes [1]. But he insists the scoring must be context-aware: penalising an open source company for low revenue is wrong, and you should instead examine "4 to 1 stars ratio," download growth, and for consumer businesses "engagement... the cohort retention" [1]. His shorthand rule is "don't look at a deep tech company to have like a 100 million revenue," and he extends it to the meta-question of "who are the investors that look at it like that" [1]. He describes personally encoding this nuance, "up like 3 in the morning just tearing down stuff," to establish a baseline [1].

The baseline is only the beginning, because everything is an edge case. When a partner disagrees, "you change it, you tell us your rationale, and we update your memory. And the next time something like that comes in, we know what Charlie likes for that thing" [1]. He frames this as the mechanism for the thing the industry has never had: "that's the only way you could like get scale taste at the edges. Otherwise, it's like impossible" [1].

On accountability to founders

A moral argument runs alongside the efficiency one. Singh is animated by founders who pitch VCs and never hear back, and says of the portfolio intelligence module that it "is going to make VCs accountable," including by design shaming those who do not deliver [1]. The mechanism is shared teams: a firm team, a team for co-investors such as YC investors, and a board-level team that includes the founder, where you "press the easy button on fundraising the A, getting customers, getting talent. And you can see who's doing the work and who's not" [1]. His conclusion is unambiguous: "we need to be accountable to our founders" [1]. The same logic applies during the deal process, where the system flags deals that are fading and drafts a pass note in your own voice for you to edit, so that at minimum "you're doing good by like the founders" [1].

On why he is not keeping the edge to himself

Asked the obvious question of why not run this privately inside a fund and enjoy the advantage, Singh says he struggled with it himself and lands on two answers [1]. The first is temperament: "philosophically I just think this industry needs it. I'm not trying to help people. I'm just like I kind of get like annoyed like how things were" [1]. He frames the outcome as market efficiency rather than charity [1]. The second is that he does not believe the edge is transferable in the first place: "your edge doesn't really go away because your edge is really your edge," and no software will settle whether one investor is better than another at looking a founder in the eye and judging whether they are a killer [1]. Tools amplify what is already there, which is why "the work is generated off of your quality" and pressing the easy button for a Series B does nothing if "you don't know anybody" [1]. His analogy: AI "takes up 10x engineer and makes him 100x, but like a 2x engineer they only make like 5x. It's like that with this system, too" [1].

On building it as a practitioner

Singh keeps insisting the product cannot come from outside the trade. On the network problem specifically: "I think it takes a practitioner to build this... if you don't understand how networks drive this entire business and how clubby it could be like you're not going to honor the craft" [1]. He is explicit that this is leverage rather than replacement: the goal is "not in a way where you're building like a robo investing platform, but in a way where you're giving leverage and time back to GPs to actually do the real work," including automating pieces of what platform teams do [1]. He is also frank about the build timeline and the difficulty, describing prototyping through April of the prior year, founding engineers arriving then, and "tinkering a lot" because "this is a hard product to figure out" [1]. RRE Ventures is a founding investor and led the seed round, and the team he assembled includes a head of product who was his co-founder at his last startup and came out of Squarespace, a technical co-founder who is a quant from Citi, and engineers from Duolingo, Amazon and Expedia [1]. The original plan was to build a software-powered firm called Originalis VC, which is where the company name came from, before he returned to RRE, where he had started his venture career in 2006-7 [1].

Takeaways

• The stated purpose of applying AI to venture is to restore the craft, not to automate judgment: give GPs leverage and time back "to actually do the real work," rather than build "a robo investing platform" [1].
• Start where the objections are hardest. Singh built analysis and diligence first precisely because it would meet the most resistance, and left sourcing for last [1].
• The central operating tension he designs around is squaring speed and depth: move slowly and lose the deal, move fast without the work and make a bad decision [1].
• Contact-dump networks do not work, because they do not reflect who you actually know; useful network intelligence requires sub-sector expertise scoring cross-referenced with relationship strength, including recency and time decay [1].
• Scoring must be context-sensitive by asset type: judge open source on stars-to-forks and download growth, consumer on engagement and cohort retention, and never expect $100 million of revenue from a deep tech company [1].
• Firm-level taste scales only through correction and memory: override the score, give the rationale, and the system learns that partner's preference for next time [1].
• Portfolio and deal tooling should make investors accountable to founders, surfacing who is doing the work on a board team and prompting a real pass on deals that are fading [1].
• Sharing the tooling does not surrender edge, because tools amplify existing quality: the 10x engineer becomes 100x while the 2x engineer becomes 5x [1].