YC Root AccessRebuilding Customer Support for the AI Era
CHAPTERS
What Pylon is today: AI-native B2B customer support platform (scale, customers, funding)
Harj opens by introducing Pylon and the founders, then quickly establishes what the product does and the company’s current traction. The team shares headcount, ARR growth, notable customers, and funding history to frame the rest of the conversation.
- •Pylon replaces tools like Zendesk/Intercom/Service Cloud with an AI-native platform built for B2B support
- •Current scale: ~79 employees, 8-digit ARR, 5.35x YoY growth, 1,000+ customers
- •Examples of customers: Honeycomb, Linear, Incident.io, Applied Intuition, Deel
- •Funding path: YC → Seed (General Catalyst) → Series A (a16z) → Series B (BCV), $51M total
Founder backstories: internships, early startup pull, and the “can’t not do it” mindset
Each founder shares the experiences that pulled them toward startups—school projects, Bay Area exposure, and dissatisfaction with large-company work. A common theme emerges: the urge to build felt inevitable.
- •Advith & Robert met at Caltech, ran a hackathon together (team-building muscle)
- •Internships at DoorDash, Slack, Airbnb shaped views on impact and pace
- •Advith quit a Slack internship early due to feeling underutilized
- •Marty’s early inspiration came from tech-founder movies and self-teaching programming
- •Shared belief: starting a company felt like the only path that would be fulfilling
Starting while employed: COVID-era routines, customer calls, and “Grind Time”
They describe the mechanics of exploring ideas while holding full-time jobs, enabled by remote work during COVID. Their process evolved from weekly brainstorming to daily customer discovery, filling any free slot with calls.
- •Remote work made side calls feasible; mornings and nights became startup time
- •Progression from weekly ideation → daily cadence → heavy discovery-call volume
- •Used structured discovery practices inspired by The Mom Test
- •Created explicit calendar blocks like “Grind Time” to maintain consistency
- •Early lesson: speed and repetition matter more than perfect planning
Early pivot mistakes: building too soon and choosing a weak initial market (edtech alumni portal)
Their first serious attempt targeted a school-admin use case: an alumni portal for career connections. They built prematurely and learned classic pivot-hell lessons, including market difficulty and insufficient validation.
- •Initial user-discovery target: school administrators (accessible network)
- •Built an alumni portal concept for career center connections
- •Key error: built before validating real demand
- •Recognized edtech as a tough space for early-stage distribution and sales
- •Each failed attempt refined their validation process for the next iteration
Marty’s parallel search: many side projects and a structured ‘best vs worst outcome’ decision to quit
Marty recounts trying multiple projects with different cofounder configurations while at Airbnb, then leaving to pursue startups full-time. A written best/worst-case analysis helped reduce fear and commit.
- •Tried a wide range of ideas (dentistry app, 3D clothing manufacturing, health tech)
- •Realized he wouldn’t trade places with career ladder roles at Airbnb
- •Learned that making progress often required leaving full-time employment
- •Used a best-case/worst-case document to rationally assess risk
- •Concluded opportunity cost was manageable; unhappiness staying was certain
How the three cofounders formed (and initially didn’t): timing, reluctance, and finally committing
The founders explain how they eventually teamed up and why it didn’t happen immediately. Early on, they felt adding more people before an idea existed could slow things down, and they also learned hard lessons about cofounder fit.
- •Marty approached Advith; Advith/Robert initially resisted teaming up pre-idea
- •Early coordination issues: ambiguity and slow follow-through
- •They explored some dead-end concepts (e.g., metaverse/digital land)
- •Close calls with adding ‘domain sales veterans’ in logistics, but trust/equity didn’t work
- •Core learning: pick cofounders for long-term working fit, not for a single idea
Debunking ‘passion’ and ‘solve your own problem’: optimizing for fun + big outcomes + ability to pivot
They discuss why they didn’t anchor on personal passion or being the end-user, and how that enabled pivots. Instead, they aligned on motivations (fun/adventure), ambition (big company), and working style (live together, grind).
- •Cofounders tied too tightly to a domain can prevent needed pivots
- •Passion for a space can blind you to building the biggest business
- •You can ‘speed-run’ user insight through deep curiosity and repeated conversations
- •Alignment dimensions: why (fun/adventure), what (big fast-growing company), how (tight, high-intensity collaboration)
- •Preference for B2B because buying decisions and ROI are more legible
Market-size discipline: bottoms-up analysis, $10B SaaS study, and comparing ideas head-to-head
To choose what to build, they explicitly worked backwards from a $10B+ outcome and analyzed public B2B SaaS winners. They compared competing internal ideas using bottoms-up models and discovered how much market choice caps company potential.
- •Defined ‘big’ as ~$1B revenue / ~$10B company and studied large public B2B SaaS
- •Built spreadsheets to compare fintech vs logistics vs a ‘horizontal SaaS’ model
- •Realized some ideas capped at tens of millions vs hundreds of millions vs orders-of-magnitude larger markets
- •Adopted a framework: huge existing category + an emerging ‘why now’ trend within it
- •Identified customer support as a massive category (e.g., Salesforce Service Cloud scale)
Arriving at the core insight: B2B companies increasingly support customers in shared Slack channels
A friend’s complaint sparked the key ‘why now’: inter-company communication shifting into Slack, breaking traditional support workflows. They rapidly validated it across their startup network and saw clear, urgent demand.
- •Shift: Slack evolving from internal chat to cross-company customer communication
- •Visceral pain: broadcasting outages and managing many customer channels doesn’t scale
- •Validation pattern: people actively searching, building in-house, or dissatisfied with existing tools
- •Persona focus: post-sales roles (CSMs, solutions engineers) as a growing, modern function
- •‘Why now’ framing becomes central: post-pandemic workflows + new communication channels
High-velocity discovery: LinkedIn DM machine, call structure, and daily iteration after quitting jobs
They describe an aggressive outreach engine: each founder sent maximum personalized LinkedIn requests daily and ran discovery calls with a repeatable script. Quitting jobs increased iteration speed, allowing questions and hypotheses to update daily.
- •Daily habit: ~40 personalized LinkedIn DMs per founder (low response rate but high volume)
- •Discovery call template: role, tools, pains, metrics, what boss cares about
- •Balance calls: ~80% probe a hypothesis, ~20% open-ended for unexpected insights
- •Key leverage from quitting: daily iteration vs weekly ‘same headspace’ loops
- •Early positioning evolved from ‘we’re exploring’ to ‘we’re building Slack support tooling’
De-risking ‘others tried this’: Hightouch’s warning, first customer fast, and first-principles confidence
They sought feedback from a founder who had tried the idea and was told not to pursue it—yet they immediately found demand. Their takeaway: prior failures often reflect execution or timing, not idea invalidity; you need a reason it will work for you.
- •Immediate validation: Hightouch actively needed a solution right then
- •Hightouch had tried the same concept; even their name tied to ‘high-touch support’
- •Lesson: many startups die from execution/cofounder issues, not idea quality
- •In YC W23, many ‘AI support’ startups died despite the category later working
- •Their edge: timing (trend accelerating) + execution + belief in multi-channel future (Slack/Teams/Discord/WhatsApp/Telegram)
Getting into YC: using a 2-week deadline to close a customer and sharpen the pitch
Their YC path wasn’t a standard application-only route: an intro led to partner conversation and an option to interview immediately or later. They chose two weeks to close an early customer and show real traction, using the deadline to focus execution.
- •YC interaction began via a connection and ‘pre-idea’ style conversations
- •They delayed the interview by two weeks to close Hightouch and build MVP
- •Scrappy selling with no brand—just a Gmail account and rapid product building
- •YC interview prep included rapid-fire question practice (‘PG simulator’)
- •Decision rationale: YC value is community and learning speed, not just capital
Product evolution: from Slack→Zendesk integration to full B2B support platform replacing incumbents
After a year focused on the Slack-ticketing integration wedge, customers pulled them toward building the full system of record. Signals included requests for core ticketing features, teams wanting to use only Pylon, and dissatisfaction with incumbents as AI reshaped expectations.
- •Year 1 wedge: route Slack shared-channel conversations into Zendesk/Intercom
- •Reached ~400K ARR on the integration-first product
- •PMF signals: customers demanded adjacent ‘core platform’ features beyond the wedge
- •Market shift: Zendesk PE acquisition + slower innovation created an opening
- •Decision: build a B2B-first support platform (AI-native) rather than remain an integration
Becoming AI-native without chasing hype: workflow sequencing, B2B accuracy constraints, and leveraging conversational data
They explain how skepticism helped them avoid shallow AI features and instead apply AI where it truly solved customer pain. In B2B support, error tolerance is low and relationships matter, so AI must be sequenced carefully—often as human+AI collaboration rather than full automation.
- •Early skepticism: didn’t add AI just because customers asked for ‘AI stuff’
- •B2B constraint: you can’t answer high-value customers incorrectly; tone must remain personal
- •AI is best applied by workflow: some tasks agentic, many human-in-the-loop, some purely human
- •Core advantage: support systems contain conversational text—ideal substrate for LLM structuring and automation
- •Advice to SaaS incumbents: map core data + workflows, pick low-risk/high-value AI insertions, and bet on tech improving over time
What’s next: expanding from ticketing to CRM-like post-sales workflows, global offices, and founder growth lessons
They close with the roadmap: expanding beyond ticketing into customer success/account management and cross-team workflows that use customer conversation data. On the company side, they discuss international expansion, building management systems intentionally, and personal growth in leading through people.
- •Product direction: ticketing → broader post-sales (CS/account management) → CRM-like system around the customer
- •Enable cross-functional workflows (feature requests to product, customer lists to marketing, etc.)
- •Long-term framing: ‘less like AI Zendesk, more like AI Salesforce’
- •Company plans: multi-region (NY, Europe), more layered leadership, intentional operations vs copying big-company process
- •Founder reflections: people are the main leverage; founders must steer because no one else has full context
