Gokul Rajaram on the SaaSpocalypse: why lightweight SaaS is dead, DoorDash's moat, and 30-40% corporate layoffs coming

Summary

The SaaSpocalypse is real, but not all software dies equally

Gokul Rajaram, a partner at Marathon Management and board member at Coinbase, Pinterest, and The Trade Desk, sees the software industry splitting into two categories. Lightweight, seat-based software faces genuine danger. Zendesk is his example. If you pay for 50 customer support seats, you can reduce to 20 and run AI agents in parallel, comparing their performance without Zendesk's knowledge. You run an A/B test on your own dime. That's a six-to-twelve-month existential problem.

Deeply rooted software like NetSuite, which runs a company's general ledger and core operations, has years to adapt. You cannot run two instances of your accounting system in parallel. NetSuite probably has three to five years before reinvention becomes urgent.

AI model capabilities double every six months. Within the five-to-seven-year horizon VCs use to evaluate companies, models will be 500 to 1,000 times better. Only a few software categories survive: things with physical atoms, regulatory moats, and systems with money flowing through them. Pure workflow software—applications that enable work but don't execute it—becomes indefensible when models like OpenAI's o1 can actually execute the tasks themselves.

DoorDash's moat is not an algorithm

When asked what it would take to replicate DoorDash with code, Rajaram points to two stories that show why AI agents alone cannot replicate managed marketplaces. First, the hypothetical: Claude or a new startup could order your sandwich cheaper and faster, but it arrives cold and the gluten-free bread was forgotten. Who refunds you? Anthropic has no customer service number. DoorDash does. Second, the historical: Google Food Ordering, launched around 2016-2017, drove multiples more traffic to restaurants than DoorDash did at the time. Retention was essentially zero because once you place an order, the physical world creates friction—late arrivals, wrong items, missing instructions—and Google could not take responsibility.

DoorDash's actual moat is managing everything after the order is placed: dasher assignment, real-time tracking, refunds, customer support. Discovery—what an AI agent does well—is just one piece. Google proved that discovery alone is not enough.

The harder, less visible moat is marketplace density. More orders in a given geography at a given time means better batching of deliveries, higher driver utilization, and lower cost per order. The actual delivery cost is $10-15; you pay $2-3 because DoorDash built dense liquidity in every city it operates in. DoorDash spent "a few billion dollars" building that. No startup replicates that with code alone. It requires physical capital, geographic discipline, and years of operational learning.

PE firms buying software companies at discount valuations now realize their traditional playbook—operationalize and cut costs—no longer works. The only play is building a new, AI-native product alongside the legacy business. Podium, a customer communications platform that had grown to $200 million in slow-growth mode, illustrates the shift. Its CEO pivoted to an AI agent product and grew from zero to $100 million in 24 months, with acceleration continuing.

Corporate layoffs are coming, and they're justified

Rajaram expects 30-40% workforce reductions across public companies over the next 18 months. He argues the productivity gains are real. If AI delivers a 30% productivity boost this year and another 30% next year, a company needs 30% fewer people to do the same work. Most public companies have already frozen net new hiring because they can accomplish more with fewer people, better coordinated. Smaller teams are faster and more innovative. Coordination overhead grows exponentially with headcount. Square's cut from 12,000 to 6,000 employees will move faster and drive more productivity than before, he predicts.

On M&A, Rajaram expects significant activity in 2026. Software valuations are compressed, and public companies are trading at depressed multiples. MyFitnessPal's acquisition of Cal AI for $50 million signals the trend. Hyperscalers will acquire next-generation infrastructure companies. Multiple inference cloud platforms are in the $5-10 billion range in value. AWS or Azure cannot sit on the sidelines while inference workloads shift to specialized providers. He anticipates acquisitions in chips, cloud, software, and consumer AI applications, with the current administration's openness to corporate activity accelerating deals.

Board composition is evolving

Rajaram outlines five board archetypes for modern public companies: a finance person for audit, a CEO for advice and confidentiality, a product-and-engineering leader since every company is now a technology company, a customer representative, and a venture investor who backed the company early and retains CEO trust. Examples include Randy Garuti, CEO of Shake Shack, on Square's board; Alfred Lin from Sequoia at DoorDash; Jeff Jordan from Andreessen at Pinterest; and Fred Wilson and Marc Andreessen at Coinbase.

The emerging best practice is a board buddy system in which each executive pairs with a board member in their domain, meeting monthly outside formal board meetings in a judgment-free zone. This gives board members exposure to succession candidates while ensuring the board adds value beyond CEO relationships. It also exposes the board to operational realities, not just C-suite narratives.

Marathon's bets

Rajaram's firm focuses on early-stage software and fintech at the intersection of infrastructure, security, and application. On software, he hunts for durable application-layer companies in a world where lightweight SaaS dies. Security is an active area. Phishing now often comes from AI-generated emails that look human, making detection harder. Infrastructure is the third focus: every infrastructure built for humans over the past decade must be rebuilt for AI agents.

On fintech, he expresses enthusiasm for stablecoins, particularly USDC. Stablecoins are the best real-world crypto application so far, offering people outside the U.S. a way to hold money without suffering local currency inflation. He notes observing 10% daily purchasing-power loss in some markets. Stablecoins solve a genuine global problem rather than speculation or tokenization theater.