Gradient Ventures GP Darian Shirazi: 'I'm bullish on Salesforce puts' — why SaaS is in jeopardy and where AI-era seed investing is heading

Darian Shirazi, Gradient Ventures

Darian Shirazi arrived at seed investing by an unusual route: Facebook employee in the early days, when pushing to production meant rsyncing from Dustin Moskovitz's laptop, then founder of Radius, an enterprise data company targeting small business intelligence as a competitor to Dun and Bradstreet. That founding experience, he says, gave him something most VCs lack — genuine empathy for how hard running a company actually is. "The mess is just masked by growth," is how he puts it.

Gradient Ventures, Google's early-stage AI fund, closed its latest vehicle at $220M. Shirazi joined eight years ago specifically to avoid crypto, which he didn't understand, and to invest in AI before the category existed. Lambda Labs, RadAI, Streamlit, and Writer were among the early bets. CaseText, which he backed personally as one of its first investors, was acquired by LexisNexis. Legora, another early position, he describes as currently "on a tear."

“I believe in shorting SaaS if I can. I generally think that software is in jeopardy. We did this robotics company at sub 50 [valuation]. The real reason is just that if you wanna build anything, you can in a few hours — and that is a major change, not just for software companies but generally for the bottom 40% of work that you don't sit behind a desk for.”

Bullish on Salesforce puts

Shirazi's view on enterprise software is blunt: he'd rather short it. SaaS, in his read, is structurally in jeopardy because building software is now fast and cheap enough that the defensibility most SaaS companies assumed has eroded. Gradient still makes software investments, but only where the company has genuine network effects or data network effects. Without a real moat, he thinks the majority of software companies are in trouble.

The fund's active bets reflect the shift. Recent investments include a robotics company with a hardware component ("I wouldn't touch hardware with a 10-foot pole five years ago"), a tool for managing multiple coding agents simultaneously, a video editing solution, and an AI hedge fund. All are outside the traditional SaaS scope.

Gradient is also deliberately slowing deployment pace until value accrual at the seed stage becomes clearer. Valuations at pre-seed and seed remain manageable, with recent deals at sub-$50M. Series A and beyond, he says, are trading at multiples he finds hard to justify.

The engineer question

Mid-level engineers at scale-up companies are already managing coding agents in English rather than writing code themselves, Shirazi says. Senior engineers still have a role — bad code at volume still needs review — but entry-level positions are under real pressure.

His response to this isn't a new type of CS degree. He thinks the conversation should shift toward trades: welding, plumbing, ceramics, car mechanics. As knowledge work gets displaced by AI, those skills become more valuable, not less. His preferred future looks like a skilled ceramicist whose marketing, photography, finance, and back office are all handled by AI, letting the artisan run a viable lifestyle business without building a unicorn.

He's also direct that everyone should understand how large language models work. He calls the transformer paper "the Magna Carta of this generation" and says it's worth reading regardless of profession.

Solo GP math

Shirazi is constructive on solo GPs, with a clear fund-size ceiling. Below $50M, it's a great business: low competition, strong ability to feed companies to multi-stage funds, and AI tools that now allow a single person to run diligence and back office operations. Above $50M, ownership constraints kick in and the solo GP starts competing directly against established seed funds with larger check sizes.

His advice for solo GPs trying to build a durable franchise: keep the fund consistently small around $30M, compress the deployment cycle, and pre-commit LPs to multiple successive funds on an 18-month cadence rather than raising one larger vehicle. Portfolio modeling at that scale requires 80 to 120 companies per fund to hit a 4 to 5x return in Monte Carlo simulations, which means writing checks every week and staying comfortable with low initial ownership, then using SPVs to build position in follow-on rounds.

LP appetite for solo GPs has cooled since 2021-2022, and Shirazi expects a wave of consolidation, with weaker vintages joining established funds and the genuinely strong performers raising next funds on their track records.