CJ Desai: Why the SaaS Bear Thesis Is Overblown

Platforms are rare, products are replaceable, and the panic over software’s terminal value is overstated. That’s the core thesis from CJ Desai, MongoDB’s CEO, delivered at the first-ever live recording of No Priors at MongoDB.local SF.

The Conversation

Sarah Guo sits down with CJ Desai for a wide-ranging conversation about the future of enterprise software in the AI age. Desai, who previously led product at ServiceNow and served as COO at Cloudflare, brings the perspective of someone who has operated across multiple platform-scale enterprise companies. The conversation opens with the hardest question investors are asking: what is software worth when AI can generate it? Desai’s answer is nuanced but firm: the bear case on SaaS terminal value being zero is overblown.

Platforms Are Sticky, Products Are Not

The central framework Desai offers is a clean distinction between platforms and products. Products can always be replaced. As his former boss Frank Slootman at ServiceNow put to it:

“Tools are for fools.”

The implication: if your software is just a tool, you’re one disruption cycle away from irrelevance.

Platforms, by contrast, create compounding stickiness. A customer doesn’t just use one feature; they build multiple applications, integrate with existing systems, pass governance and security audits, and eventually become embedded in the fabric of their infrastructure.

Desai illustrates this with a striking anecdote from a meeting with a London bank’s CTO. The bank runs 300 applications on MongoDB. Desai asked about the denominator. The CTO’s answer: 9,000. “That’s a great opportunity for MongoDB,” Desai said. The CTO’s response: “CJ, don’t worry. We are not going anywhere.”

The deeper point: the number of companies that have crossed $10 billion in pure-play software revenue is in the single digits. The software industry has existed for decades, built by brilliant people. Why so few? Because platforms are rare. Getting from a killer initial use case to a durable platform requires compounding product adoption, deep integrations, and enterprise trust that takes years to build.

Vibe Coding Won’t Solve the Enterprise Problem

Desai acknowledges that vibe coding and AI-assisted development increase app velocity. You can build something fast. But the enterprise gauntlet remains unchanged:

• How do you approach a bank with a go-to-market channel?
• Can you pass regulatory tests?
• Do you support multi-cloud resiliency?
• Can you run in air-gapped networks?

Banks, he notes, talk to regulators far more than they talk to vendors. The enterprise class requirements (governance, security audits, multi-cloud, on-prem) are not things that faster app creation solves. Speed to build is table stakes. Speed to enterprise trust is the actual moat.

Fortune 500 AI Adoption: Still Early

Desai speaks to at least 10 customers per week and shares his pattern-matching. The picture is surprisingly sober:

Office productivity copilots: the feedback is “not great” on value delivered. Customers tried them and found the results underwhelming for Excel and PowerPoint use cases.

Coding assistants: the clear 2024-2025 breakthrough. Very positive feedback across the board, from GitHub Copilot to Anthropic’s tools. This is where real productivity gains have materialized.

Customer support AI: still in the tinkering phase. Large telcos and healthcare companies are exploring AI-native support tools, but the end-to-end customer experience is not there yet for enterprise scale.

The Fortune 500 is still talking about moving X% of applications to cloud (AWS, GCP, Azure). That transition, nearly 20 years after AWS launched, is still ongoing. The AI transition is layered on top, not replacing it.

The “And” vs “Or” Question

One of the most revealing moments comes when Desai describes how enterprise buyers think about AI-native startups relative to their existing systems of record. The key question customers ask him: should I think about this AI-native company as an “and” or an “or”?

Desai’s answer is unambiguous: if a startup comes in and says it can replace the system of record entirely, with cheaper pricing, faster deployment, and better outcomes, that gets his full attention. He would have that conversation “every single day.”

This is surprising from someone who built systems of record at ServiceNow. But Desai practices what he preaches: MongoDB’s own CIO Deepa gets approached by AI-native companies multiple times daily, and Desai’s framework for evaluating them is the same. Can this let us hire fewer people, make existing people more efficient, and transform (not just optimize) our business?

“I want to be AI first organization on behalf of MongoDB to say we are transforming our business, not making just productive.”

Why MongoDB Specifically

Desai walks through his decision framework for joining MongoDB. Three factors:

Durable TAM: The database market has existed for 50 years (Oracle’s 50th anniversary is in a year and a half). MongoDB, founded in 2007, is the only truly disruptive force in that time. It has crossed both $1 billion and $2 billion in revenue, which no other challenger database has achieved.

Accidental AI fit: The founders built MongoDB for document-oriented, unstructured data before AI made that the default data shape. AI applications generate messy, high-velocity data that needs rich search capabilities. MongoDB’s architecture turned out to be accidentally perfect for this.

Dual transition tailwind: Cloud migration is still ongoing. AI adoption is just beginning. MongoDB sits at the data layer, which is required in both transitions. “TAM, must-have layer, no risk of disruption.”

Managing Through Technology Transitions

Desai draws on examples from Nokia and BlackBerry to make a point about change management. BlackBerry was still selling well three to five quarters after the iPhone launched. Disruption doesn’t announce itself with a cliff; it comes as gradual irrelevance.

The pattern he has seen across ServiceNow, Cloudflare, and now MongoDB: when engineering teams are doing well, they resist change. At ServiceNow, when he pushed for AI investment early on, the engineering team’s response was “oh this is just something that’s out there.” His response:

“Not leaning in is not an option.”

The step-function jumps that separate durable platforms from doomed products happen at two moments: going from single-product to multi-product, and navigating technology transitions without losing intellectual honesty. The temptation for large incumbents is to bundle products, relabel existing features as “AI,” and play pricing games to make the numbers work. Desai is blunt: investors see through it. The only way to disprove the bear thesis is to show reacceleration of growth.

A Few Observations

This conversation is most valuable as a framework for thinking about software durability rather than as a MongoDB pitch, though it is obviously both. The key insights worth sitting with:

• The platform vs product distinction is simple but has real explanatory power. Single-digit companies at $10B+ in software revenue is a genuinely striking data point.
• Enterprise buyers are more open to rip-and-replace than conventional wisdom suggests, but only if the replacement offers transformation, not incremental improvement.
• The AI adoption curve among Fortune 500 is far earlier than the hype cycle implies. Coding assistants are the only category with clearly positive feedback.
• Desai’s framing of “and vs or” as the defining question for AI-native startups is useful. Most startups position as “and” (a layer on top of the system of record). The ones that get enterprise attention are betting on “or.”
• The honest admission about MongoDB’s Q3 results (growth was core business, not AI-driven) is refreshing and strategically smart. Over-attributing to AI creates expectations you can’t sustain.