Sep 10, 2026 · 1h 9m · neon-show
Why Startups Fails Even After Finding PMF | Prukalpa Sankar, Atlan
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
Atlan co-founder Prukalpa Sankar examines why enterprise context has become the defining moat in artificial intelligence, detailing how startups can achieve true product-market fit through rigorous customer discovery. She breaks down the technical and organizational transitions required to scale enterprise data infrastructure and build high-performance, AI-native companies.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Siddhartha holds 9.5% of the talking time here. How this is scored →
Siddhartha as informed peer 2.7 Guest teaching 4.9 Guest disagreement 1.2 Siddhartha pushing back 0.9
speaking balance: gold is Siddhartha, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 1:03:54 Rejection of the non-technical sales rep premise
When Siddhartha argues sales reps don't need to be builders, Prukalpa directly shuts down the premise, insisting sellers to frontier companies must understand gateways, harnesses, and memory systems.
Hardest push from Siddhartha ▶ 17:58 Host pushes back on rebuild contradictionSiddhartha directly presses Prukalpa on an apparent inconsistency between claiming to build from scratch while previously stating internal tools formed the core product.
Biggest teaching moment ▶ 8:35 Correcting host on infra transitions from servicesWhen Siddhartha claims transitioning from data services to infra is globally unheard of without 3 years of building, Prukalpa schools him with direct counterexamples like DBT/Fishtown and Alteryx.
Siddhartha holds their own ▶ 12:35 Host challenges Atlan's architectural positionSiddhartha demonstrates data stack domain knowledge by asking where Atlan fits if an enterprise already operates data warehouses and data lakes.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Siddhartha as informed peer | Guest teaching | Guest disagreement | Siddhartha pushing back | Why |
|---|---|---|---|---|---|---|
| The Evolution of Context in Data and AI | 3 | 5 | 1 | 0 | Siddhartha opens by flattering Prukalpa on coining context in 2019 versus Bill Gates's 'content is king' quote. Prukalpa graciously corrects the timeline to 2020 and gives a foundational masterclass on how Atlan evolved from SocialCops into an AI context layer. | |
| Scaling Through Operational Pain and Finding PMF | 2 | 4 | 0 | 0 | Siddhartha asks about getting 200 customers in 50 countries within a year. Prukalpa explains that internal operational pain at SocialCops created genuine user demand rather than software built just to be sold. | |
| Searching for Truth Rather Than Traction | 4 | 6 | 4 | 2 | Siddhartha claims it is unheard of globally for a data services company to pivot to an infra company quickly without three years of building. Prukalpa immediately corrects his premise, citing DBT (Fishtown) and Alteryx, and delivers her core thesis that PMF is a search for truth rather than traction. | |
| Validating User Pain and Discovering the ICP | 3 | 5 | 1 | 2 | Siddhartha challenges how Atlan fits into the modern data stack alongside warehouses and lakes. Prukalpa clarifies user demand versus buying budgets, detailing their structured outbound discovery lab. | |
| Rebuilding Infrastructure and Data Lineage Profiles | 3 | 5 | 2 | 3 | Siddhartha probes for inconsistency, noting Prukalpa previously mentioned reusing internal tools but then claimed they rebuilt from scratch. Prukalpa clarifies that rebuilding tech is fungible and the easiest part, while user flow and lineage map insights are the real product. | |
| Scaling to AI Agents: Knowledge, Expertise, and Norms | 2 | 6 | 0 | 0 | Siddhartha asks what the product looks like today. Prukalpa breaks down the tripartite framework required for AI agents: knowledge, expertise, and norms, walking through a concrete customer support refund scenario. | |
| Defining the AI Context Platform Category | 2 | 5 | 1 | 0 | Siddhartha asks what companies Atlan replaces and who else occupies the space. Prukalpa articulates the architecture (orchestration frameworks, control planes, systems of record) and notes Gartner's recent elevation of the category. | |
| Executive AI Workflows and Hands-On Building | 2 | 3 | 0 | 1 | Siddhartha inquires about customer counts and Prukalpa's personal bot-building habits. Prukalpa shares concrete tactical details on their internal Marketing OS and personal context-engineered autonomous email agents. | |
| Why Context is the Real Moat for CIOs | 3 | 6 | 1 | 0 | Siddhartha asks why context suddenly became a CIO-level line item. Prukalpa gives a sharp, analytical monologue explaining that intelligence is commoditized and human job performance relies over 90% on context rather than raw intelligence. | |
| Evolving GTM into Strategic Context Advisory | 3 | 5 | 2 | 1 | Siddhartha asks if sales processes have gotten easier. Prukalpa explains the nuances of selling strategic advisory versus commodity software where Atlan held an 85% competitive win rate. | |
| Structuring the AI-Native Frontier Enterprise | 3 | 5 | 1 | 1 | Siddhartha challenges the buzzword 'AI-native'. Prukalpa provides a rigorous definition of a frontier enterprise that re-evaluates org charts, single-player vs multiplayer workflows, and agent accountability. | |
| Cultivating Human Taste, Judgment, and Org Learning | 2 | 6 | 1 | 0 | Siddhartha asks about frontier labs' org structures. Prukalpa explains how AI shifts the premium to human taste, judgment, and being a force of nature, highlighting their open-source initiative Becoming Frontier. | |
| The Inbound Fundraising Strategy and Seed Journey | 2 | 4 | 0 | 0 | Siddhartha asks how Prukalpa navigated inbound fundraising from India. She explains her network-building tactic of asking investors to introduce the smartest people they knew and forcing them to earn their seat on Atlan's cap table through customer intros. | |
| Growth Capital: Peak XV, Insight Partners, and Beyond | 3 | 3 | 0 | 0 | Siddhartha walks through subsequent rounds with Peak XV, Insight Partners, GIC, and Meritech. Prukalpa recounts the origin stories and quick preemption during the Series A. | |
| Founder Mindset: Global Ambition and Hands-On Building | 4 | 5 | 4 | 3 | Siddhartha asserts that while CEOs must build, sales reps do not need to be builders. Prukalpa flatly disagrees, arguing anyone selling to the frontier must understand model gateways and harnesses, before outlining core advice for Indian founders going global. |
Statements from this episode (31)
Sankar: Agentic AI use cases took off after reasoning models dropped
“All kinds of agentic use cases, which really started becoming a thing after the reasoning model dropped.”
Sankar: AI needs context to achieve agentic outcomes
“Every AI use case, if you want to go towards agentic outcomes is going to need context.”
Sankar: Product-market fit is about repeatability, not novelty
“Product market fit is about repeatability, not novelty.”
Sankar: Product-market fit is a search for truth, not traction
“Product market fit is a search for truths, not traction. And that's a really strange thing to internalize as an entrepreneur.”
Sankar: Selling to personal networks creates fake traction because people want to help
“I believe, most people love founding stories. Most people want to help. And most people will find a way if you speak with them and tell them this is the problem you're trying to solve, they'll find a way to help you. And that doesn't mean that they are going t…”
Sankar: Horizontal platforms are loved by VCs but pose horrible GTM challenges
“We were a horizontal platform, which was great. You know, VCs love it. It's actually a horrible go-to-market problem because you know, where do you start? It's about repeatability, right?”
Sankar: Atlan would not exist today if discovery was delayed three months
“If we had been three months late we would not have existed today. We would not be alive.”
Sankar: Software is fungible, but building products people use is hard
“Rebuilding technology is actually the easiest part of this problem, right? Like software itself is fungible. It has always been fungible. Like, I don't know why people now in AI suddenly are like, oh my God, it was always fungible. It was never that hard to Bu…”
Sankar: Analysts spend 30% to 40% of time discovering data
“The challenge for the analysts was they used to spend, in fact, about 30, 40% of their time just discovering data and playing around with data before they could get to analysis.”
Sankar: Data engineers spend roughly 80% of time debugging
“In fact, like, 80% of time of data engineers were spent on debugging things.”
Sankar: 50% of Atlan's platform users are AI agents
“Actually I, I'd say 50% of our end users are now agents and the rest 50% are humans.”
Sankar: Agents require three context layers: knowledge, expertise, and norms
“This is the context that any agent, actually human or AI agent needs to be able to solve this problem for the customer. Knowledge, expertise, and norms. So if you had to build this in a business, you need to be able to bring together first the map of the busin…”
Sankar: Legacy data platforms and knowledge graphs are rebranding as context platforms
“So you'll see, you know, semantically or platforms are rebranding as context platforms and ontologies and knowledge graphs are rebranding as context platforms and data platforms are rebranding as context platforms. The model layers and the agent platforms are …”
Sankar: Enterprise context will be the defining moat and proprietary IP in AI
“It is going to be company IP. It is going to be the true differentiator or the moat in this new world.”
Sankar: Atlan built an internal marketing OS powering 16 agents
“So we've built something called a marketing OS, which powers all our agents, and we have about 16 agents, and all the humans in our marketing team.”
Sankar: Less than 10% of human job performance stems from cognitive intelligence
“And in fact, less than 10% of human job performance is explained through cognitive intelligence. Everything else is Context and how you learn on the job and how you act in real life.”
Sankar: Building AI agents is easy, but managing context is hard
“Investing in the
Orchestration layer or the model layer, like that's the easier problem to solve. Building an agent is easy. How do you actually make it accurate? How do you actually manage your context? That's going to be the hard problem.”
Sankar: The AI market currently has more sellers than buyers
“I think there are more sellers than there are buyers right now in the market, right? It's like we're in the middle of this AI. Everything is AI. Everything's trying to sell something AI.”
Sankar: Atlan maintains an 85% competitive win rate against rivals
“At Athlan, we, this probably a public number our competitive win rate is 85%.”
Sankar: AI-native companies must foundationally redesign hierarchies and processes
“You have to like leveling structures, like you have to foundationally reimagine how you operate as a company to be truly AI native or on the frontier versus I think The other option of it is you could say, hey, this is how I do things today. I'm gonna go slap …”
Sankar: Humans will always remain at the center of AI development
“One of our biggest realizations has been that humans will always be the center of AI. And how humans expand their creativity and their thinking through this is what is going to actually drive the next wave of this.”
Sankar: AI remains a single-player game instead of multiplayer
“There's still AI is a very single player game. It's not a multiplayer game.”
Sankar: AI adoption makes human judgment, taste, and agency far more important
“Actually, as AI nativity increases on one side, like if you think about the axis of impact certain human skills also become much more important. So when AI raises the bar judgment, Taste. Being a force of nature. These are all human skills that actually become…”
Sankar: Corporate AI discussions are theater because companies hide failed experiments
“And in fact, the core purpose is there's right now a lot of AI theater going on. So people only share the good things about AI. Nobody shares the experiments that went wrong.”
Sankar: Every institutional VC that backed Atlan helped close customer deals first
“Every round that we closed when we raised from an institution, the institution had helped us close some customers.”
Sankar: Settling for average PMF stalls startups at $5M to $7M
“You either never find product market fit or you find product market fit, but it's an average product market fit. And so you peter out at five million or seven million because you can't scale beyond that.”
Sankar: Frontier tech salespeople must be builders to hold customer credibility
“Well, I actually think they do. I think if you're trying to sell to the frontier what, how are you going to have a conversation with a customer who's talking to you about, you know, the difference between model gateways and three harnesses and memory systems, …”
Sankar: Modern leaders have no excuse not to build directly
“In today's world, you can't be a leader And not be pushing your team. I truly believe like you have to be in the details and the bar is so low. There's no excuse. It's so easy. Like there's no excuse to not set up and build yourself today.”
Sankar: Demand generation, not sales execution, bottlenecks enterprise scaling
“The bottleneck is not sales, it's demand generation and pipeline.”
Sankar: India lacks the Bay Area's generational legacy of enterprise learning
“I think the challenge that India has is that India doesn't have the ecosystem that the Bay Area has, right? In the Bay Area, you had Adobe that passed to Google, who passed to Facebook, who passed to, you know, like, there's just this, like, legacy of learning…”