Summary

Desai rejects the market’s “terminal value being zero” view of software: durable value follows companies that build, learn, and pivot fastest through platform shifts. “Speed matters,” whether the transition is internet, mobile, cloud, or AI; falling behind is what prompts customers and investors to question a company’s future.
The wedge may get a software company to $10 million or $100 million, but becoming a platform is what supports billion-dollar scale and retention. Desai notes that only single-digit pure-play software companies exceed $10 billion in revenue because “platforms are rare”: at least two products must work together and integrate deeply into customers’ existing systems.
Vibe coding accelerates application creation without removing enterprise distribution, governance, or resilience barriers. A bank may still demand regulatory approval, security audits, AWS/GCP portability, or an on-premises air-gapped deployment. Faster code therefore does not automatically displace enterprise-grade platforms or confer access to Fortune 500 technology budgets.
Desai sees LLMs as present for the foreseeable future and the data layer as a required component of the emerging AI stack, while everything around them remains contestable. Application vendors must demonstrate vertical expertise, faster time to value, and capabilities old SaaS could not deliver; infrastructure investors should distinguish genuinely “must-have” layers from products exposed to rapid replacement.
Enterprise AI adoption remains sharply uneven: office-productivity copilots have delivered unclear value, while coding assistants became meaningfully useful in 2024 and broke through in 2025. Customers report gains in innovation velocity and security, but end-to-end AI customer support is “not there yet,” leaving buyers to decide whether AI-native tools augment or replace systems of record.
Desai is willing to replace embedded SaaS when an AI-native company is demonstrably cheaper, faster, better, and priced against value. His standard is transformation—not merely productivity—including hiring fewer people or making current teams materially more effective. A European retailer’s decision to build its own ERP on MongoDB after costly failed implementations illustrates that willingness.
Incumbents can refute the AI bear thesis only by converting innovation into renewed sales growth while staying intellectually honest about what caused the numbers. Desai said MongoDB’s Q3 performance was “absolutely not” driven by AI, despite hundreds of AI-native customers; Guo noted that, whether success means $100 million ARR or $1 billion ARR, there are “like 10” such companies today, and Desai framed AI as presently “an and, not an or” to the growing core business.

Deep dive

1. Products win entry; platforms earn durability

Sarah Guo opens with the market’s existential question: what is software worth when a bunch of software can be generated? Desai calls the zero-terminal-value extreme overblown, but accepts that since ChatGPT’s fall 2022 arrival, customers and investors have been questioning the entire stack.
Desai’s governing rule across internet, mobile, cloud, and AI transitions is “speed matters.” Companies must build quickly, learn from the shift, and pivot before the market forces the question, “What is the future of your company?” He concedes that “not every bet will work.”
His sharper distinction is that “platforms are sticky; products are not.” Products can be replaced in a disruptive market; platforms become embedded. Desai says his hiring manager at ServiceNow, Frank Slootman, used to say, “Tools are for fools,” capturing the danger of selling something customers perceive as a disposable utility.
Guo pushes back with startup orthodoxy: ServiceNow itself entered through the service-desk wedge. Desai agrees a killer initial use case is necessary, but argues that easy entry can mean easy exit; a wedge may support the journey from zero to $100 million, while reaching billions requires multiple products and integrations.

2. Enterprise friction survives vibe coding

Desai’s scale evidence is stark: only single-digit pure-play software companies exceed $10 billion in revenue because “platforms are rare.” His minimum definition is “n equals at least two”—multiple products used together, then connected to the surrounding systems of century-old banks, insurers, and healthcare companies.
One bank had roughly 300 critical applications on MongoDB and told him, “We are not going anywhere.” Desai then asked for the denominator: 9,000 applications. His conclusion was both expansion pitch and durability mechanism—the more workloads adopted, the deeper MongoDB sits “in the fabric of their infrastructure.”
Guo’s challenge is that vibe coding could let enterprises create those remaining applications on demand. Desai’s rebuttal: higher app velocity does not supply a go-to-market channel or satisfy regulators, governance reviews, security audits, multicloud resilience across AWS and GCP, and sometimes truly sandboxed, air-gapped on-premises operation.

3. The durable AI stack still needs data

Against the movement of investor dollars toward models, AI infrastructure, and hyperscalers, Desai screened opportunities for a “durable TAM” and a “must-have layer.” MongoDB passed because he found customers running mission-critical e-commerce, commercial-banking, healthcare, and insurance-claims applications on it, while digital and AI natives were also building on it.
Desai traces that conviction to his Oracle experience: Oracle will celebrate its 50th anniversary in a year and a half, and he says the database market has existed for 50 or 60 years. MongoDB, created in 2007 and only “18-ish years” old, is, in his telling, “truly the only disruptive force.”
Cloud migration reinforces the duration of infrastructure transitions. If cloud started with AWS, it is approaching its 20th year, yet Fortune 500 customers still discuss moving percentages of applications across AWS, GCP, Azure, and other environments. “AI transition has just started,” while its messy, unstructured, high-velocity data strengthens the database requirement.
Desai identifies two relatively durable components in the emerging stack: LLMs “will be there for the foreseeable future,” and data “has to be there because you need to store data somewhere.” The surrounding application layer must prove use-case depth—insurance, for example—plus faster value and capabilities that were impossible in old SaaS.

4. Enterprises want transformation, not an AI veneer

Desai considers a week a “total failure” if he does not speak with at least 10 customers. His pattern match: Fortune 500 and Global 2000 adoption remains slow; office-productivity copilots have produced weak or unclear feedback, coding assistance took off meaningfully in 2024 and broke through in 2025, while end-to-end customer support is “not there yet,” though some initial use cases work in certain industries.
Buyers consequently ask whether AI-native support is an “and or an or” beside Salesforce and other systems of record. Desai will have a conversation every day with a vendor promising full replacement that is “cheaper,” “faster,” “better,” and disruptively priced according to delivered value—an openness Guo finds surprising given implementation sunk costs.
His internal standard is similarly demanding: MongoDB should be “AI first,” using AI to transform the business rather than merely make people productive. If it allows MongoDB to hire fewer people and make its people more efficient, he says, the organization will use that budget. A European retailer that decided to build its own ERP after expensive, failed implementations elicited his response: “You had me at hello.”

5. Incumbents must prove the transition in revenue

Desai credits John Thompson at Symantec around 2005 with teaching him that product leaders must continuously visit customers, ask about adjacent pain, and “see around the corner.” “As you build, they will come” does not happen; customer intimacy reveals deployment time, expected value, pricing, and crisis expectations.
At NRF, a European retailer’s CTO said e-commerce generated 20% of revenue and ran happily on MongoDB, but did not know MongoDB also offered search and vector search. For Desai, that exchange shows how direct customer contact improves both product judgment and platform expansion.
Technology transitions are largely change-management problems. Desai told a skeptical ServiceNow engineering team that “not leaning in is not an option”; AI might mature in two years or four, but the organization had to engage. BlackBerry still sold well for, he hedges, “three or five quarters” after the iPhone before disruption arrived.
Guo warns that incumbents can bundle products that may or may not be working for customers, relabel a piece “cloud” or “AI,” and use pricing maneuvers to manufacture results; organizations need guardrails between customer reality and Wall Street. She also notes that, whether success is defined as $100 million ARR or $1 billion ARR, there are “like 10” such companies today. Desai agrees there are not many successful companies and therefore not much data, though some use MongoDB.
Desai’s answer is reacceleration with candor: he said MongoDB’s Q3 results reflected its core, not AI, while AI-native demand remains additive—“an and, not an or.”

Europe's Free Speech Failure and the "Censorship Industrial Complex" with Sarah B. Rogers Elon Musk vs Sam Altman | The Implosion of Thinking Machines | Can VC Survive Public Pricing?