CHAPTERS
- 0:00 – 2:29
From Plaid CTO to an unexpected new startup: why Town caught a16z’s attention
Alex recounts being confused by Jean-Denis leaving Plaid to build an SMB tax product—then being blown away when Ben Horowitz praised “Town.” They tee up the core theme: why a seemingly “working” business can be the wrong bet in the age of AI, and how Town reframes what AI at work should look like.
- •Ben Horowitz’s endorsement creates curiosity and contrast with the earlier “SMB tax” idea
- •Jean-Denis’s early thesis: AI must deliver value beyond a simple text box
- •Town’s ambition: AI embedded into everyday knowledge-work workflows
- •Set-up for the founder journey from Plaid → tax → Town
- 2:29 – 6:19
The SMB tax bet: why it looked promising and how AI made it less venture-scale
Jean-Denis explains why SMB business tax prep was an appealing early AI market: text-heavy workflows, structured inputs, and clear automation potential. But after building for a year and reaching meaningful early revenue, the team realized the go-to-market constraints and incumbent advantages limited venture-scale upside.
- •Tax is well-suited to LLMs: heavy text plus structured documents
- •Technical feasibility was validated; product worked
- •Go-to-market to reach millions of SMBs was harder than expected
- •In an AI world, $1–3M revenue trajectories can be insufficient for venture scale
- •Incumbents can automate faster using existing distribution
- 6:19 – 10:28
Walking away from a working business: the courage and the timing constraints of a cyclical market
They unpack the emotional and ethical difficulty of pivoting when the business isn’t failing—just capped. Jean-Denis describes founders’ motivations (impact, honesty with team), the reality of employee opportunity cost, and why tax’s annual cycle forced a decisive pivot window.
- •Founders must periodically re-justify “why am I doing this?”
- •If conviction in a great outcome is gone, recruiting and leadership become morally hard
- •Pivoting impacts employees; mitigation includes severance and support
- •Tax’s annual cadence creates a narrow decision window (post-season)
- •Cyclical businesses force clearer go/no-go decisions than continuous markets
- 10:28 – 13:35
How Town was born: the overlooked agent living in email + calendar
Exploring several ideas, the team focused on a deceptively simple question: why hasn’t anyone built a great agent that works from email and calendar? Customer feedback from the tax product and improving agentic models helped Town reach fast product pull, revealing a huge horizontal surface area across roles and geographies.
- •Origin insight: meeting prep, follow-ups, repeat email responses are universal pain points
- •Tax customers themselves requested adjacent automation help
- •Agentic model improvements made the concept viable at the right time
- •PMF showed up as strong user pull—TAM analysis came later
- •Email/calendar are near-universal business interfaces worldwide
- 13:35 – 16:16
PLG-first and consumer-grade UX: don’t let enterprise check-writers ruin the product
Alex and Jean-Denis discuss why Town prioritized product-led growth and user delight over premature enterprise compliance-driven design. Drawing on Dropbox lessons, they argue it’s feasible to add enterprise controls later—but extremely hard to retrofit a great UX into a compliance-first tool.
- •Risk: building for CISOs/admins can degrade the end-user experience
- •Dropbox lesson: great consumer UX can be enterprise-ified; reverse is far harder
- •Thesis: executives will demand consumer-grade AI at work (like iPhone adoption)
- •Strategy: keep improving core UX while ensuring strong security foundations
- •Now shifting to scale enterprise deployment once core UX is proven
- 16:16 – 21:27
The “iPhone moment” for AI at work: adoption happens at human speed, then accelerates
Using BlackBerry vs iPhone (and Good Technologies) as analogies, they frame AI work adoption as inevitable but gated by trust, data access, and enterprise permissioning. Town’s wedge is immediate value from email/calendar, then earning the right to connect deeper systems of record with proper controls.
- •Analogy: consumer exposure drives enterprise mandates (iPhone dynamic)
- •Cautionary tale: ‘enterprise wrapper’ apps can still lose on quality (Good)
- •Agents need secure access to business data/APIs to deliver full value
- •Land-and-expand journey: start with email/calendar → later connect CRM and more
- •Dual-focus: SMB self-serve growth + enterprise security/compliance feature set
- 21:27 – 29:10
Town as a token-allocation machine: finding ROI in the long tail of work
They redefine the market beyond “assistants,” arguing cheaper tokens unlock tasks humans/EA’s would never do at scale. Town’s job is to continuously surface high-ROI opportunities—like meeting prep or vendor price monitoring—while making the value concrete and automated, not prompt-driven.
- •Market isn’t ‘EA replacement’—it’s continuous AI deployment to improve work
- •As tokens get cheaper, new use cases become economical (Jevons paradox)
- •Examples: prep notes for every meeting; long-tail research at near-zero cost
- •SMB example: analyze wholesaler price sheets monthly to save real money
- •Core UX requirement: show value and do it for users, not just offer a text box
- 29:10 – 34:03
Horizontal vs vertical: personalization via routines, power-user discovery, and an AI-driven flywheel
Jean-Denis argues most AI tools will converge in capability, making differentiation about personalization and deployment, not features. Town starts horizontal to learn from diverse users, then builds role-specific “routines” seeded from power users—moving toward AI-assisted discovery that scales beyond human interviews.
- •Capability parity is rising; durable advantage shifts away from feature leads
- •Town bets on deep personalization by role/industry during onboarding
- •Uses aggregate metrics + power-user interviews to uncover common automations
- •Builds and ships ‘routines’ templates (e.g., restaurant workflows) as seeds
- •Contrasts with FDE-heavy approaches: human custom work scales poorly long-term
- 34:03 – 45:46
Why Town over OpenAI/Anthropic: neutrality, data ownership, and a more agent-native future
They tackle the risk of competing with model suppliers and explain Town’s differentiation as a neutral router across models and modalities. Town emphasizes customer ownership of context/memory, non-training posture, and more aggressive product bets around reducing human-in-the-loop routing inside organizations.
- •Supplier competition is risky; Town’s answer is customer-centric neutrality
- •Neutral routing: pick best model by task (text/voice/video/slides) over time
- •Data/context ownership: make switching easy; avoid vendor lock-in via FDEs
- •Training stance: don’t train on customer business processes
- •Product philosophy: reduce ‘human router’ latency via agent-mediated internal requests
- 45:46 – 52:33
Why AI is underhyped outside the Valley: the ‘tinkerer’ gap and the limits of the chat box
Jean-Denis explains that surprising AI deployments appear across the economy, but only where a “tinkerer” exists to operationalize them. The bigger barrier isn’t intelligence—it’s productization, integrations, and guided deployment so non-tinkerers can get the same value without custom work.
- •Unexpected real-world use cases emerge (e.g., surfboard damage/deposit workflows)
- •Diffusion depends on whether someone has time/skill to tinker
- •Most businesses still use AI like ‘better Google’ due to UX constraints
- •Key insight: the experience can’t be a text box; you must show and automate value
- •Goal: deliver out-of-the-box, personalized magic as soon as systems connect
- 52:33 – 1:01:58
Neutral agents, distribution shifts, and the broken ad covenant: adversarial vs synergistic futures
They explore how agents disrupt the traditional “free content for eyeballs” bargain, pressuring ad-driven platforms and supply aggregators. The discussion contrasts friction humans enjoy (shopping as experience) versus friction they don’t (replenishing staples), and argues users will demand neutral agents whose only incentive is the user’s wallet.
- •Agent traffic is rising; the classic ad covenant (content for eyeballs) breaks
- •Concern: agents tied to ad businesses may bias outcomes (distribution distortion)
- •Preference for user-aligned, wallet-funded neutral agents
- •Not all friction should be removed—some shopping is experiential
- •Supply aggregation becomes less valuable as attention can be automated; platforms may need their own agents
