The Twenty Minute VCEoghan McCabe, CEO @Intercom: Freedom of Speech, Censorship and Government Control | E1213
CHAPTERS
- 0:00 – 2:28
Intercom vs. AI disruption: why customer support isn’t “dead next year”
Eoghan responds to the claim that AI will quickly wipe out customer support and therefore threaten Intercom. He argues that while AGI could eventually obsolete today’s software, real-world adoption and productization take much longer than people expect.
- •AI will eventually reshape most work, but timelines are often overstated
- •Intercom isn’t building as if it will be “crushed in 1–2 years”
- •Tech cycles reward timing; companies must reinvent before obsolescence hits
- •Customer support is a major AI target, but rollout and adoption are slower than hype
- 2:28 – 5:59
How Intercom’s AI agent “Fin” works and why domain AI is hard
Eoghan explains why Intercom’s bot resolves a high share of tickets and why that performance isn’t just “wrapping OpenAI.” He emphasizes the integration of historical data, many experiments, multi-model orchestration, and purpose-built components on top of LLMs.
- •Fin’s resolution rate improved from ~high-20% to ~50%
- •Performance comes from historical conversation data + extensive system design
- •Over 100 experiments, patented components, multiple LLMs, and homegrown AI
- •Domain-specific AI requires significant engineering beyond base models
- •Application-layer companies can win because labs won’t build every domain solution
- 5:59 – 9:07
Overestimating AI’s near-term impact vs. underestimating today’s value
They debate whether AI progress is overhyped. Eoghan argues the world-change narrative is inflated in the short term, but the market still underappreciates how many AI-first companies are already delivering real value and scaling quickly.
- •Raw capability may arrive before broad societal impact due to adoption friction
- •AI-first companies are hitting revenue milestones faster than SaaS historically
- •There’s heavy noise and hype—investors must separate signal from excitement
- •Retention and usage are key to identifying durable AI revenue vs. pilots/POCs
- 9:07 – 10:35
Will companies build their own AI tools? The Klarna narrative challenged
Harry raises the idea that companies will abandon SaaS vendors and build in-house AI systems. Eoghan calls this a repeat of “build your own workflow tools”—possible in the short term, but usually inferior to specialized, deeply engineered products.
- •Building a full AI agent platform in-house is deeper than most expect
- •Purpose-built systems require sizable ML and product engineering investment
- •Short-term prototypes can look exciting but may not be robust long term
- •Eoghan predicts many will return to professional vendors over time
- 10:35 – 12:13
AI economics: deflationary costs, but rising “AI consumption”
They explore whether AI will force big price increases or drive costs down. Eoghan argues technology is fundamentally deflationary, and that savings from reduced human labor will be reinvested into delivering better, more expansive customer experiences.
- •Compute/model costs trend downward over time (deflationary tech pattern)
- •AI agents can become cheaper than humans while improving service quality
- •Budgets may shift from labor to heavier AI usage, creating over-served customers
- •As AI features commoditize, pricing pressure increases over time
- 12:13 – 14:26
Jobs and team size: hiring slows, work shifts upward, not instant mass layoffs
Eoghan discusses how AI affects org size and labor. He argues businesses will need fewer people over time, but changes often show up as slower hiring and role evolution rather than immediate layoffs—using Intercom’s support org as an example.
- •Technology historically enables high enterprise value with fewer employees
- •Intercom used Fin for ~2 years without layoffs, but stopped support team growth
- •AI handles simpler work first; humans move to higher-level/creative support
- •Some companies can automate more if their inbound issues are highly concentrated
- 14:26 – 17:31
Will foundation model companies swallow the app layer? Plus 10-year “crazy” predictions
Harry asks whether OpenAI/Anthropic will subsume applications like customer support. Eoghan says maybe in the far future, but builders must solve practical adoption and product problems; he also predicts major growth in robotics and continued automation of “dangerous, demeaning” work.
- •AGI-like capability could eventually collapse app categories, but not soon
- •Visionary narratives often ignore messy realities of business adoption
- •Robotics likely expands dramatically, impacting blue-collar work
- •Automation historically replaces dangerous work while overall prosperity rises
- 17:31 – 20:50
‘Thin layers’ and startup defensibility: moats are overrated, teams matter
They discuss YC-style AI apps that look like “model stitching” and whether that’s a problem. Eoghan argues early products often look non-defensible; execution speed and quality matter more, and durable defensibility often comes later via networks/ecosystems/data.
- •Many great companies start as simple products with weak-looking moats
- •Defensibility often = ability to build faster/better and drive adoption
- •Even giants (e.g., Apple) face fast-follow competition despite patents
- •Enduring moats can emerge later: networks, ecosystems, proprietary data
- 20:50 – 23:22
Why Intercom isn’t public: mission, time horizon, and freedom to make big swings
Eoghan explains why he doesn’t prioritize an IPO despite Intercom being in the “hundreds of millions” revenue range. He argues public-market compliance and quarterly pressures inhibit bold reinvention—especially in periods of major technological change.
- •IPO benefits are often framed around investor needs (liquidity, analyst rigor)
- •Founder priorities: customers, revenue, impact—not where shares trade
- •Public-company constraints can limit large bets during volatile transitions
- •Intercom nearly filed in 2022, but the market collapse halted plans
- 23:22 – 24:40
Does the VC model still work? Liquidity bottlenecks and early-stage commoditization
They broaden from Intercom’s IPO choice to the venture ecosystem. Eoghan notes the extended private-company timeline creates tension for VC expectations, and suggests early-stage capital has become commoditized, making returns harder for many funds.
- •Extended time-to-liquidity challenges traditional VC timelines
- •Private liquidity, M&A, and PE may reopen but the market is “awkward” now
- •Early-stage capital abundance suggests weak differentiation among many VCs
- •Big value creation may happen “behind closed doors” during change cycles
- 24:40 – 25:45
Re-accelerating Intercom: commercialization mistakes and the SMB-to-mid-market reset
Eoghan details why growth plateaued and what changed. He points to forcing more sales involvement, greedy pricing/contracting, and shrinking customer count—then describes a return to healthier growth with a clearer segmentation strategy.
- •Forcing SMB users into sales increased ACV but hurt top-of-funnel growth
- •SMB customers were the “lifeblood” that later matured into larger accounts
- •Pricing and contracting strategy became too aggressive and reduced expansion
- •Intercom remains SMB-heavy while moving more intentionally into mid-market
- 25:45 – 30:10
‘Project 52’ and culture reinvention: intensity, founder mode, and hard-work values
Eoghan describes internal changes to “restart the startup,” including narrowing focus to customer service and increasing operating intensity. They debate work ethic, work/life balance, office presence, and top-down decision-making versus consensus leadership.
- •Clear focus choice: stop “doing all the things,” commit to customer service
- •Project 52: 52-week reinvention with weekly company communication
- •Performance management aligned to values; mismatched employees exited
- •Eoghan argues hard work beats “work smarter” narratives; founder-led decisiveness
- •Office policy: mandated two days/week; he’d choose five if restarting
- 30:10 – 36:42
Public political views: why he speaks out, pro-Trump framing, and politics at work
Harry challenges Eoghan on bringing politics into public discourse. Eoghan frames his stance as pro-liberty and anti-war, argues tech leaders often agree privately, and supports keeping politics out of the workplace while defending employees’ rights to personal expression.
- •Motivation: protecting individual liberty, free speech, and peace
- •He supports candidates he believes maximize liberty (claims Trump/Vance do)
- •Many in tech won’t admit views publicly due to backlash dynamics
- •Workplace stance: no politics inside Intercom; company shouldn’t police employees’ speech
- •Authenticity in leadership: letting people opt in/out based on real beliefs
- 36:42 – 43:11
Freedom of speech vs. misinformation: who decides truth, and where limits exist
They argue through censorship, misinformation, and the dangers of institutional control. Eoghan asserts that centralized authorities can’t be trusted to arbitrate truth, while acknowledging narrow constraints around direct harm (e.g., incitement, doxxing, “fire in a theater”).
- •Censorship is portrayed as a pathway to dangerous state power
- •Misinformation policing raises the core question: who gets to decide?
- •Reputation, criticism, and social consequences as alternatives to censorship
- •Limited carve-outs: direct incitement/coordination to harm crosses into illegality
- •Cultural context: US vs. UK/Europe attitudes toward sovereignty and speech norms
- 43:11 – 50:50
Leadership evolution, SF Freedom Club, and quick-fire on nuclear risk & reading
Eoghan reflects on personal hardship, ego, authenticity, and what drives founders. He explains the San Francisco Freedom Club concept and closes with quick-fire answers touching on nuclear war risk, books, and how his view of LLMs has shifted.
- •Returning as CEO after illness/press attacks led to greater authenticity
- •Admits a “chip on shoulder” and desire to prove himself as a founder trait
- •SF Freedom Club: building community/parties around “freedom” values; strong demand
- •Quick-fire: nuclear war risk, ‘1984,’ preference for podcasts, notable guests
- •Changed mind: LLMs won’t “eat software” as soon as many think
