The two-year thesis: VCs are triaging portfolios, not reviewing them

A Fortune column published on August 20, 2026 reported that venture investors are running portfolio reviews as triage rather than as health checks, on the premise that deals which looked good in 2024 are obsolete by 2026 ^[1][2]. That implies a working thesis life of roughly two years ^[1], and according to Kamran Ansari of Kapital Ventures only two categories are structurally insulated from that clock: regulated license businesses, and businesses sitting on proprietary data that is difficult to access ^[4].

"No doubt that AI is changing really fast, and it's changing the half-life of a thesis," Eric Archer, cofounder of Monashees, told Fortune ^[3]. Ansari's version is blunter: absent those two things, he said, everything else feels exposed ^[5].

The operational shift is in what the review asks. Lily Lyman, managing partner at Underscore VC, told Fortune the question is no longer whether a company is on track but, in her words, "What track are you on, and what does that mean?" ^[6] She sorts startups into four lanes: the soaring AI consensus bets, efficient-growth companies, companies with a path to product-market fit in tough industries, and companies caught in the crosshairs of OpenAI and Anthropic ^[7]. For that fourth group, she said, the models have moved what is possible far enough that the startup's original value proposition is no longer as valuable as first thought, and the salvage question becomes whether the residual value is people, product, or distribution ^[8].

The numbers are the part worth acting on. Ansari estimated that 10 to 20 percent of his portfolio feels very vulnerable right now, in addition to normal venture dropoff ^[9]. An early-stage VC speaking to Fortune anonymously put roughly 10 percent of their portfolio as specifically imperiled by foundation model shifts ^[10]. Robert F. Smith told a Fortune Brainstorm Tech audience that a small but real portion of his software companies no longer have a right to exist ^[11]. Taken together, that is between one in ten and one in five companies failing for a reason that did not exist at entry ^[2], which is substitution rather than a market correction.

Perplexity is the illustration of how fast substitution runs. Ansari said the company was "molten-lava-hot" and that he no longer thinks it is that special, because Google caught up extraordinarily fast and its AI-powered search is now pretty good ^[12]. Zachary Aarons, cofounder of MetaProp, who has been tracking what he calls the SaaSpocalypse, argued that software companies have to agentify their own platforms to survive, because mission-critical industries such as construction do not want to run everything on raw foundation-model products ^[13][14].

The most consequential suggestion is procedural. Ansari said he would welcome founders coming back 18 months in with much of the cash intact, returning 60 or 70 cents on the dollar and raising again later, rather than taking the gas tank to zero and sending the inevitable email ^[15]. That is a voluntary 30 to 40 percent loss instead of a total one ^[3], and it turns a dead line item into recyclable capital plus a relationship that survives.

Watch whether the 10 to 20 percent band moves at the next review cycle; it currently rests on one named investor and one anonymous one ^[9][10]. Watch for the first fund to publicise an actual returned-capital figure, which is what would turn Ansari's proposal from a preference into a norm ^[15]. And watch pricing in the two supposedly insulated categories, because if licenses and inaccessible data are the only durable moats on offer, they will be bid as such ^[4].