Dev Ittycheria - The Database Evolution

Value first accrues at the bottom layer of the stack (e.g., infrastructure/compute like Cisco in 2000 or Nvidia today). This matters for AI because early outsized returns show up in compute providers, while more durable, widespread returns later come from applications that deliver customer value.

He says today's AI apps (simple chatbots, summarizers) are like the first trivial iPhone apps (flashlight, iBeer). Over time, AI apps will resemble Uber/Airbnb-level transformations: deeply embedded, business-critical applications leveraging real-time data.

While early value accrues to compute/infrastructure, he believes long-term returns will come from applications that (1) create great customer experiences, (2) automate and cut costs, and (3) enable new or disruptive business models using AI.

Because every dev team already has a large backlog, and historically each tech transition lowers the cost of building apps, leading to more applications and more data. AI will increase developer productivity and thus the number of applications built, not eliminate developers.

The primary customer is the developer. AI changes their workflow by increasing productivity and enabling them to build more, richer applications faster; MongoDB is focused on how the developer workflow evolves in an AI-first world.

He points to thousands of startups already building AI applications on MongoDB's flexible, scalable platform, which he treats as a "crystal ball" for future large-enterprise adoption patterns.

Launched in 2016 (pre-IPO), cloud was ~2% of revenue in 2017 and grew to ~68%. Despite skepticism about "partnering and competing" with hyperscalers, MongoDB bet that customers wanted a managed service, not to operate the database themselves—this bet became a major growth driver.

MongoDB uses a document-based model instead of rigid tables. Data is stored in documents that align with how developers think in code, avoiding the cognitive load of decomposing entities into many related tables.

He says any application (shopping, travel, gaming, mobile, enterprise) depends on persistent data; about 70% of developer time is spent working with data—presenting the right data, at the right time, to the right user in the right context.

Once applications are built on a database, migrating them is very hard and expensive. This stickiness explains why many organizations still run Oracle despite disliking it.

If LLMs store user history (queries, actions, context) in a persistent data store, switching between models becomes harder because that historical data is valuable and bound to the database, increasing switching costs and the importance of data stores.

He believes leaders must be "incredibly judgmental"—willing to make hard, extreme decisions on people, product, and go-to-market. Most leaders avoid painful conversations and accountability, leading to average or mediocre companies.

He describes it as people nodding in meetings but privately doubting decisions. He views it as a form of duplicity and combats it by building a culture of intellectual honesty and fierce, direct conversations about what's working and what's not.

Raising only $6M right before 9/11 at BladeLogic, during a nuclear winter for tech, forced him to be ruthless about people and decisions—he couldn't afford to be patient with underperformance, which built his "deal with problems head on" muscle.

Excess capital reduces pressure to make hard decisions quickly, encouraging tolerance of mediocrity and delayed action on people or strategy problems, which can compound and hurt the business.

He notes almost everyone fires "too late" rather than too early, because they introduce hope and give underperformers more time, even when the evidence shows they won't work out.

Instead of assuming adults will just do what's asked, leaders must frequently and deeply inspect work. When people know performance is regularly inspected, they raise their game and issues are surfaced early.

It meant constant inspection and transparency: no place to hide poor performance or problems because metrics, deals, and execution were continuously reviewed and discussed.

1) Why does the customer want to do anything? (pain/problem)

2) Why MongoDB/our solution? (best option vs alternatives)

3) Why now? (compelling event to avoid decision deferral)

Because customers can accept the problem and your solution but still postpone action; many deals are lost to "no decision," not competitors. Creating urgency is often the hardest part.

A champion sells for you when you're not in the room, has power and credibility, and is a change agent. A coach wants you to win and gives intel but won't take personal risk or drive the decision.

Broad-based performance across reps, not an 80/20 pattern where a small minority carries the majority of results, plus consistent execution over quarters instead of yo-yo results.

Recruiting brings in hungry, smart talent, but without development—on product, deal prosecution, qualification, forecasting—those people won't reach their potential or execute consistently.

In high growth, revenue/bookings forecasts are proxies for expense planning. Poor forecasting can lead to overspending and burning cash too quickly.

1) Set very clear expectations.

2) If the person first misses, assume you (the manager) weren't clear and re-explain the why.

3) If they miss again, hold them directly accountable—no ambiguity or hiding.

He asks: Do they have the skill to do the job and the will (drive/motivation)? Failures usually stem from one or both being low, guiding whether to coach, reassign, or exit someone.

Investigate why their motivation dropped (e.g., denied promotion, personal issues, misalignment with role) and, if will doesn't return, consider replacing them despite their skills.

He watches whether they are losing their team—high attrition of good people and declining quality of new hires suggest the leader isn't scaling and is likely failing.

According to Dev, A's hire A's, B's hire C's, and C's hire F's. Watch the quality of people each leader brings in as a leading indicator of their own caliber.

Frontline employees see problems first and clearly; as bad news moves upward, it's filtered and softened, so by the time it reaches the CEO, it's delayed and understated unless leadership actively digs in.

He immediately digs in, assuming he's the last to know and that the situation is worse than presented. This prevents being blindsided later by escalated problems (e.g., churned customers).

Dev says: "If you see a problem and don't act on it, then you're the problem."

He probes what makes people tick—their upbringing, hardest experiences, motivations, long-term orientation, and how thoughtfully they made past life and career decisions—to assess grit and values.

"What's the most difficult thing you've ever endured?"

Listen for real adversity, how they responded, what they learned, and whether it created durable grit and drive that will sustain them through startup challenges.

Ask it openly, listen for real adversity, how they responded, what they learned, and whether it created durable grit and drive that will sustain them through startup challenges.

1) A big market,

2) A defensible technology advantage with proof points,

3) A CEO who is smart and coachable, with the right cultural/values fit.

High NDR (e.g., 120%+) shows customers are expanding spend net of churn, indicating the product is delivering ongoing value and justifying more go-to-market investment.

Month-over-month growth of existing customers' spend of around 5%—with relatively junior sales reps—showed easy adoption and rapid expansion.

Their discipline in prioritizing building great businesses over optics, and Roelof's habit of listening carefully and speaking with high signal-to-noise instead of leveraging ownership to dominate board discussions.

They must be incredibly self-aware, know how to recruit well, and know how to hold people accountable.