Skip to content
YC Root AccessYC Root Access

PostHog: Pivots Were The Real Lesson In Building A Startup

In this fireside from Startup School Paris, PostHog CEO James Hawkins talks about surviving five failed ideas, why going open-source nearly trapped the company in enterprise on-prem debugging hell, and how a hackathon side project by one engineer turned into a multi-product strategy neither founder had planned. Apply to Y Combinator: https://www.ycombinator.com/apply Work at a startup: https://www.ycombinator.com/jobs Chapters: 00:00 — Intro 00:46 — The Rule: Sleep on It Once 02:13 — Why Pivots Were the Real Education 02:31 — Birth of Open Source Analytics 05:14 — How They Knew This Idea Was Different 05:49 — Competing Against Amplitude & Heap 07:17 — The On-Prem Nightmare 08:57 — Going Cloud & Killing the Friction 10:33 — The Engineer Who Forced Multi-Product 13:06 — When Ambition Exploded 15:45 — The Shift to Self-Driving Software 17:18 — The Agent That Beat All Other UX 19:29 — Fixing Bugs While You Sleep 21:15 — Half of PRs Are Getting Merged 23:02 — Betting on Where the Puck Is Going 24:27 — Elephant Carpaccio 29:06 — The Founder Who Became a PM 31:49 — Why Going Wild Gets Easier 34:41 — Free Up Everything That Isn't This

James Hawkinsguest
Aug 19, 202638mWatch on YouTube ↗

CHAPTERS

  1. 0:05 – 1:45

    From five pivots to PostHog: treating early ideas as a learning machine

    Jared and James set the stage with PostHog’s origin story: it was the team’s sixth attempt after a rapid sequence of pivots. James explains how they optimized for learning speed rather than debating, using quick real-world feedback to decide whether to continue or restart.

    • PostHog was the 6th idea after ~7–8 months of rapid pivots
    • Early goal: build a “machine that can learn,” not maximize revenue immediately
    • Common failure modes: no one to talk to, no one uses it, no one pays
    • Bias toward action over prolonged internal deliberation
    • Learning product skills by shipping instead of over-intellectualizing
  2. 1:45 – 2:31

    The one-night rule: deciding to pivot without spiraling

    James details their simple operational rule for making hard decisions: sleep on it once, then commit to restarting if it still feels wrong. This prevented endless second-guessing and kept momentum high despite repeated resets.

    • Rule: if it feels not working, sleep once, decide next day
    • Avoids weeks of existential debate and indecision
    • Each idea was pursued intensely (“burn the ships”) before dropping it
    • They knew they were inexperienced at product and leaned into fast iteration
    • Getting software into users’ hands produced the most reliable signal
  3. 2:31 – 5:20

    Why open-source product analytics: frustration with buying and implementing tools

    PostHog emerged from the founders’ pain repeatedly needing analytics while building new products and hating the purchasing/implementation friction of existing vendors. James contrasts his enterprise-sales background with the desire to build a bottom-up, developer-friendly product.

    • Constantly rebuilding sites/logins made analytics a repeated need
    • Existing tools were frustrating to buy and integrate (sales-led friction)
    • Developer-first expectations: clear pricing, self-serve signup, credit card flow
    • James’s enterprise-sales past motivated building the opposite motion
    • Belief in bottom-up adoption spreading inside companies
  4. 5:20 – 6:22

    Standing out vs Amplitude/Heap/Mixpanel: GitLab-style open source and on-prem promise

    With major incumbents already successful, PostHog’s early differentiation was open source plus self-hosting to keep data in the customer’s infrastructure. James describes the initial hypothesis and the immediate realization that it risked pulling them back into enterprise selling.

    • Competitive landscape: Amplitude, Heap, Mixpanel already established
    • Differentiation: open source + run in customer infrastructure (data stays local)
    • Initial analogy: GitLab vs GitHub-style positioning
    • Early traction and revenue appeared—but skewed toward conservative buyers
    • Concern: this model leads back to heavyweight enterprise motion
  5. 6:22 – 7:28

    The on-prem nightmare: debugging customers’ infrastructure at scale

    Self-hosting created a hidden support burden: analytics workloads are far heavier than typical production databases, and customers’ DevOps environments varied wildly. James shares vivid examples of operational chaos that made on-prem support unsustainable.

    • Analytics data volume dwarfs typical production DBs (clicks/pageviews/etc.)
    • Running on customer infra exceeded many teams’ operational competence
    • Support became “debugging other people’s Kubernetes over Slack”
    • Security/permission constraints made troubleshooting painful and slow
    • Operational drag threatened focus and scalability
  6. 7:28 – 9:13

    Going cloud: removing friction with self-serve, big free tier, and cheaper cloud than self-host

    PostHog shifted to cloud—initially to help customers who were self-hosting ‘for no reason’ and struggling. They leaned into self-serve onboarding, usage-based pricing, and a generous free tier to reduce adoption friction, even if it meant under-monetizing early value.

    • Cloud was built to solve the ‘hosting is painful’ problem for users
    • Cloud product initially worse than competitors, but choice became: PostHog OSS vs PostHog Cloud
    • Motivation: control experience and stop ‘buggy’ perception driven by self-host issues
    • Strategy: self-serve onboarding, usage-based pricing, large free tier
    • Tradeoff: intentionally not capturing full value early; win via volume and ease
  7. 9:13 – 10:30

    Open source’s new role: trust, community, and an irreversible commitment

    As cloud became dominant, open source shifted from core go-to-market wedge to a durable brand/community asset. James explains that most users moved to cloud (aided by compliance work), and self-hosting became a small minority of usage with minimal commercial focus.

    • Open source remained, but became less central to day-to-day adoption
    • Estimated <5% of users self-host; a few large companies still do
    • Compliance (HIPAA, SOC 2) enabled many ‘never cloud’ customers to migrate
    • ~95% of paying self-hosted users moved to cloud despite initial resistance
    • Self-hosted users largely kept for free with limited support expectations
  8. 10:30 – 11:41

    An engineer forced the multi-product strategy: Session Replay as the second wedge

    Multi-product expansion wasn’t initially planned—it was triggered by an engineer shipping Session Replay after user requests. The unexpected success reframed PostHog’s competitive strategy: instead of being a point solution, they could out-ship incumbents by building an integrated suite.

    • Session Replay built in a hackathon from user demand—despite founders’ skepticism
    • Immediate popularity validated multi-product expansion
    • Thesis: competitors are point solutions; PostHog can build ‘all their products’
    • Perceived advantage: faster shipping velocity, amplified by open source building blocks
    • Suite strategy strengthened cloud differentiation against incumbents
  9. 11:41 – 13:06

    Building ‘full-on’ products like a data warehouse: scaling with customers’ data teams

    James describes how deeper customer maturity (bigger spend, dedicated data teams) pushed PostHog into warehousing and data-combining workflows. The team’s pattern—pressure-testing ‘impossible’ ideas until a clear build path emerged—enabled ambitious product jumps.

    • Data warehouse was initially viewed as too complex—even internally
    • Reframing exercise (‘gun to your head’) helped uncover a feasible approach quickly
    • Customers over ~$100k/year begin integrating product data with systems like Salesforce
    • Shared infrastructure and shared context across tools creates compounding value
    • Multi-product started early (within ~1 year of pushing cloud) due to competition
  10. 13:06 – 15:31

    Ambition explodes: competing where PMF already exists, and outworking the market

    James argues that competing in established markets was a shortcut to product-market fit compared to their earlier ‘no competition’ ideas. With clearer benchmarks, the team could focus on differentiation, execution speed, and shipping an unusually broad surface area even pre-AI.

    • Counterintuitive lesson: they did better against big competitors than in empty markets
    • Competition signals real demand; the task becomes standing out and branding
    • Clear target feature set made execution straightforward—then they “outworked everyone”
    • Early PostHog shipping sprint: product, website, SDKs, docs in weeks
    • Founder roles: Tim coded; James handled users/marketing and removed blockers
  11. 15:31 – 17:54

    From ‘too many buttons’ to agent-first UX: the path to self-driving software

    As the product suite grew, some customers found it overwhelming—AI became the mechanism to unify complexity into a single interface. PostHog built an agent with an iterative query loop, and once it could try/assess/retry, it became the dominant user experience and the main driver of positive feedback.

    • Customer feedback: multi-product suite felt overwhelming despite ‘magic box’ benefits
    • 2025 shift: AI as a way to use all products together on shared data
    • Key technical unlock: iterative ‘single query loop’ for agent experimentation
    • Agent quality jump: from mediocre to better-than-human outcomes (though still slow)
    • Brand monitoring showed ~95% of positive feedback tied to the agent experience
  12. 17:54 – 22:30

    Self-driving mode: prioritize problems, generate fixes, and merge real PRs

    James explains ‘self-driving mode’: pipelines and “scouts” that build context and memory to prioritize what matters, then move from surfacing issues to actually fixing them. Early rollout shows strong adoption and a meaningful PR merge rate, with ongoing work on review/testing and safer deployments behind flags.

    • Goal shift: from ‘data company’ to ‘doing company’—solve, don’t just surface
    • Scouts read across errors/logs/replays/flags to prioritize like a human PM would
    • Proactive fixing while you sleep (P0 issues), leveraging continuous monitoring advantage
    • Early launch: emailed ~7,000 users; reached a couple million run-rate within ~24 hours
    • Quality metric: >50% of generated PRs merged; building review, agentic testing, and flag-based rollout
  13. 22:30 – 29:05

    Betting where the puck is going: harnesses for intent, and scaling AI across the company

    To keep up with AI-native competitors, PostHog is investing not only in improving today’s agent workflows but also in the next layer: systems that extract goals/requirements and guide longer-running agents. James connects this to broader organizational change—roles converging as non-engineers ship code and agents begin to automate end-to-end jobs (including go-to-market).

    • New reality: must bet on direction of travel, not just ship fast where demand already is
    • Next bottleneck: translating human intent into requirements as PR volume scales massively
    • Build a ‘harness’ that behaves like a strong product engineer—asking why, challenging assumptions
    • Internal shift: support/PM/marketing increasingly ship code; many PRs via Slack are from non-engineers
    • ‘Elephant carpaccio’ idea: automate whole tasks end-to-end, not just narrow analytics slices
  14. 29:05 – 38:53

    Founder as PM and change leader: going ‘wild’ gets easier, but requires ruthless focus

    James describes taking direct ownership of the AI initiative despite a sales background, driven by feedback, upside asymmetry, and the need for founder-level commitment. He closes with tactical advice: free up time aggressively, narrow responsibility, start small with a few people, and manage the human side of change as legacy UX and habits are disrupted.

    • James leads AI/product shift; feedback rituals (360 dinners) helped him ‘back himself’
    • Thesis: ambitious products are easier to recruit for, market, fundraise for, and energize teams
    • Founders can’t outsource a seismic AI transformation; product excellence is the lever
    • Operational advice: ruthlessly drop non-core work (interviews, investor work, 1:1s) to focus
    • Change is messy: protect existing users, consider separate surfaces (e.g., new desktop app), and guide teams through disruption

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.