By playing, you agree to uListen's Terms and Privacy Policy and YouTube's Terms of Service.

Why HubSpot’s Co-Founder Is Betting Against the SaaSpocalypse

As frontier AI labs expand into applications, where should startups build? a16z Partner Joe Schmidt joins HubSpot co-founder and CTO Dharmesh Shah to explore why the software opportunity is growing, and how founders can stay off the…

Dharmesh ShahguestJoe Schmidthost

Nov 10, 202659mWatch on YouTube

Summaries, chapters, transcripts, and FAQs on this page are made by uListen, not YouTube.

Great conversations run long. Their best ideas don't have to.

uListen is free on iPhone, iPad, Android, Chrome, and Claude.

CHAPTERS

Jump to publisher timestamps

Chapter summaries

ChatGPT as the new OS: setting the stakes for SaaS and CRM
Dharmesh Shah frames ChatGPT and Claude as a new operating layer for how people access software. Joe Schmidt sets up the conversation as a continuation of his “Yellow Brick Road” idea: there’s room to build durable software outside the path frontier labs are racing down.
- •Dharmesh says ChatGPT and Claude are effectively “the new OS.”
- •Joe explains the discussion started from his “Yellow Brick Road” essay about avoiding direct competition with frontier labs.
- •Dharmesh emphasizes that value comes from the work happening, with interface as a means to an end.
- •The episode’s focus is established: what’s worth building and what established software companies must change as AI becomes the customer interface.
Best time to be a builder: market uncertainty vs. renewed mission
Joe asks what it feels like to run a successful public software company amid a pessimistic AI narrative. Dharmesh describes it as both uncertain and energizing, drawing a parallel between the early internet era and today’s AI moment for SMBs.
- •Dharmesh calls it a “best of times, worst of times” period with excitement alongside uncertainty.
- •He compares HubSpot’s origin—helping SMBs benefit from the internet—to today’s goal of helping SMBs benefit from AI.
- •He says focusing on customers helps cut through “fog of war” in the market.
- •Dharmesh says this is the most fun he’s had building software in 30+ years.
AI as the next jump in abstraction—and why iteration speed matters
Dharmesh explains software progress as a march toward higher abstraction, and he sees AI as another step that expands what builders can do. He argues the biggest practical change is faster iteration: learning cycles compress dramatically.
- •Dharmesh describes software history as increasing abstraction from machine language through higher-level languages.
- •He frames AI as a toolset that lets builders work at higher abstraction and iterate faster.
- •He notes the “scope of who can be a builder” expands with new development tools.
- •He says faster iteration enables quicker customer testing and cheaper learning loops.
How to actually pick a startup idea: test quickly, then narrow distribution
Dharmesh challenges the idea that founders can spreadsheet their way to the best idea. He argues the only reliable test is shipping and learning, while warning that distribution is still difficult and often harder as building gets easier.
- •Dharmesh says rigorous upfront analysis rarely predicts which ideas will work; testing is what reveals quality.
- •He notes successful outcomes often don’t resemble the original idea founders started with.
- •He emphasizes narrowing focus early to improve odds of learning and breaking through distribution noise.
- •He critiques founders who lead with huge TAM narratives instead of what they’ll build in the first 30 days.
- •He highlights winning as having the shortest, cheapest iteration loop.
From LegalSpot to HubSpot: choosing SMB and abandoning the law-firm ERP path
Joe asks about HubSpot’s early formation and the rumored legal product. Dharmesh recounts how he and his cofounder targeted SMBs for the “best of enterprise and consumer,” initially narrowed to ERP for law firms, then pivoted when customer interest centered on marketing and growth.
- •Dharmesh says SMB appealed because it supports charging money without enterprise-style revenue concentration.
- •Their original broad concept was “an Oracle for small business,” narrowed to ERP for law firms as a test (LegalSpot).
- •Customer conversations revealed law firms cared most about marketing and getting more ideal clients.
- •Dharmesh says they nearly could have flipped a coin, but chose marketing and avoided selling deeply into law firms.
- •He jokes lawyers are very hard to sell software to—second only to software developers.
The 60% without a CRM: why HubSpot built a free CRM and became a platform
Dharmesh explains HubSpot didn’t start with a CRM platform vision; it emerged as a second act. A customer survey showed most paying HubSpot customers weren’t using any CRM, motivating a product that removed intimidation and cost—then evolved into a platform to compete in CRM long-term.
- •Dharmesh says the move to CRM wasn’t planned at “T equals zero.”
- •A customer survey found 20% used Salesforce, 20% used another CRM, and 60% used no CRM at all.
- •He says customers were intimidated and believed CRMs were too expensive, so HubSpot focused on ease and made the CRM free (not freemium).
- •Dharmesh describes a philosophy of creating customer value before extracting value.
- •He explains that to win in CRM, the product had to become a platform, and HubSpot often broadens scope to solve the whole customer problem (reducing fragmentation).
Building an ecosystem developers trust: openness, clarity, and “time to joy”
Joe asks how HubSpot built its ecosystem, and Dharmesh describes patterns that break developer trust when platforms later “stomp over” partners. He outlines HubSpot’s approach: transparency about intent, avoiding hostile shifts, and treating developers as customers with their own time-to-value.
- •Dharmesh warns platforms often attract developers, then later close off access or compete with them, eroding trust.
- •He references early Twitter as an example where an open ecosystem later became developer-unfriendly.
- •He says HubSpot tries to be the platform they wished existed: friendly, transparent, and pragmatic about mutual profit.
- •Dharmesh describes sharing guidance on where HubSpot sees white space and what it’s unlikely to build.
- •He frames ecosystem building like product building, optimizing “time to joy” and speed to a developer’s first working integration and first dollar.
Staying off the Yellow Brick Road: avoiding ‘yet’ bets and product-vs-company thinking
Dharmesh expands on Joe’s “Yellow Brick Road” metaphor, arguing founders should avoid building directly in the path frontier labs will inevitably widen into. He flags two common traps: competing with a feature gap that can be solved “yet,” and comparing against a product rather than the full capabilities of the company behind it.
- •Dharmesh agrees frontier labs will “barrel down” the obvious path and that the road is widening.
- •He warns against startup ideas that rely on “ChatGPT can’t do this”—because adding “yet” often makes it risky.
- •He distinguishes between what a product can do vs. what the company behind it can decide to build.
- •He suggests choosing opportunities where frontier labs face structural constraints (e.g., regulatory) or incentive constraints (unlikely to focus).
- •He uses the idea of a frontier company targeting CRM revenue as theoretically possible but practically unlikely given their broader ambitions.
Why models are becoming infrastructure: fungibility, routing, and less fear for builders
Dharmesh argues frontier labs must monetize massive CapEx by selling inference, but that doesn’t require owning every application experience. He says models are trending toward infrastructure: interchangeable enough to route across, with multiple commercial and open options reducing single-supplier risk.
- •Dharmesh describes frontier labs’ need to monetize by selling inference/tokens to consumers and enterprises.
- •He says an alternative path is SaaS companies acting as domain-aware sellers/resellers of inference-enabled value.
- •He compares AI infrastructure dynamics to cloud infrastructure adoption 15 years ago.
- •He says it’s less scary now than in 2023 because models are becoming more fungible and swappable.
- •He references use case decomposition and model routing within SaaS, plus the existence of multiple model options (including open source).
ChatGPT as the new browser: interface vs. real value and when to avoid inference
Joe asks whether plugging into ChatGPT erodes HubSpot’s value; Dharmesh argues most business apps derive value from the work and underlying systems, not the UI. He frames ChatGPT/Claude as a new access layer and emphasizes using deterministic systems where they fit rather than forcing everything through LLM inference.
- •Dharmesh says concerns about AI interfaces diminishing SaaS value misunderstand where value resides in many business apps.
- •He compares businesses like delivery apps where the network/operations matter more than the UI, arguing B2B often similarly values the underlying work.
- •He calls ChatGPT/Claude the new OS or new browser layer through which customers consume value.
- •He says marketplaces and plug-in architectures inside AI products reinforce that applications will plug in rather than disappear.
- •Dharmesh argues anything solvable deterministically with a structured query should be solved that way, not via expensive inference.
Rethinking HubSpot’s primitives for AI: system of record plus system of context
Joe asks how HubSpot’s original architecture—built for humans navigating structured information—changes in an AI world. Dharmesh describes evolving the platform’s primitives to pair structured “system of record” data with unstructured “system of context,” potentially organized as a graph to power inference, agents, and outcomes.
- •Dharmesh describes building large systems around reusable primitives that can be remixed over time.
- •He argues structured storage remains necessary for known facts and deterministic operations.
- •He introduces the idea of adding a “system of context” for unstructured data that was previously hard to use at scale.
- •He suggests organizing context as a graph (a “context graph”) to support richer inference and agentic workflows.
- •He frames the goal as updating the framework for the next 20 years of HubSpot in an AI world.
Replaying 20 years of code with AI: alternate architectures, evals, and big rewrites
Dharmesh describes a new possibility for incumbents: using agentic coding to explore large-scale rewrites and architectural alternatives that were previously too expensive. He explains how existing tests and evaluation systems can validate whether alternative implementations preserve outcomes while trading off quality, latency, and cost.
- •Dharmesh says AI doesn’t just speed up feature development; it enables deeper, larger projects that teams might not have attempted.
- •He imagines “playing back” different architectural choices over an existing codebase by spending tokens to regenerate variants.
- •He notes incumbents have assets like extensive tests (and what he calls modern “evals”) to measure viability of changes.
- •He describes evaluating trade-offs across models or implementations (quality points vs. latency and economics).
- •He frames this as exploring alternate “branches” of what the product could have been with different early choices.
HubSpot Next, uSpot, and a startup within a startup: learning fast without the baggage
Dharmesh explains HubSpot Next as an innovation lab designed to move with startup-level freedom inside a larger company. He details uSpot—a one-person CRM for solopreneurs—as a 30-day-old experiment to test product ideas like a “second brain,” along with early monetization and a consumption-based model signal.
- •Dharmesh describes HubSpot Next as a “startup within a startup” with fewer constraints on stack, brand, and rules.
- •He says uSpot targets one-person businesses and helps narrow focus for faster learning and experimentation.
- •He explains how projects test new categories, markets, or technologies, including ideas like a “second brain” and context ingestion.
- •Dharmesh says uSpot monetized early to learn pricing and value perception, and he was surprised by willingness to pay via consumption-based pricing.
- •He shares market signals: record new company formations and a growing share expected to remain one-person companies, motivating the product direction.
AI is still just software: rebutting the “SaaSpocalypse” with new problem space
Dharmesh argues AI doesn’t end software—it expands what software can solve, even if some companies fail to adapt. He revisits HubSpot’s early fundraising skepticism about SMB and marketing software and explains how internet-driven changes made marketing more software-suited—drawing a parallel to AI unlocking a new class of solvable problems.
- •Dharmesh says AI is software—“really smart software”—not a new energy source.
- •He recalls VCs saying there weren’t billion-dollar marketing software companies and that SMB was hard to monetize.
- •He explains HubSpot’s thesis that the internet would make marketing resemble problems software was designed to solve.
- •He argues AI unlocks a massive new set of problems software couldn’t previously address effectively.
- •He concludes the aggregate opportunity is larger than ever, even if some software companies don’t survive the transition.
Lightning round and closing reflections: pragmatism, joy of building, and best days ahead
Joe wraps with a lightning round and asks Dharmesh to reflect on HubSpot’s culture and his personal experience. Dharmesh emphasizes customer-focused pragmatism, explains he operates without direct reports as an individual contributor, and says his most enjoyable period at HubSpot is right now.
- •Dharmesh answers rapid-fire personal and cultural questions, including HubSpot’s “most Boston” traits.
- •He describes HubSpot as combining West Coast ambition with East Coast pragmatism and grind.
- •Dharmesh says a recent week was his best day because ideas and execution started falling into place.
- •He notes he has no direct reports and can focus as an individual contributor while still making long-term bets.
- •He closes by saying the best time in HubSpot’s 20-year history—for him—has been the present.

Publisher timestamps

Get more out of YouTube videos.

Read along on your phone, search it beside the video, or ask your favorite AI app what was said.