"From Copilots to Agents: Rebuilding the Company Around AI"
The a16z Show | Host: Andreessen Horowitz
Airdate: February 18, 2026
Guests: Carlos Garcia Otati (Founder & CEO, Kavak), Angela Strange, Gabriel Vasquez
Episode Overview
In this episode, the a16z Show explores how Kavak, Latin America’s leading used-car marketplace, underwent a radical transformation by embracing AI agents at the core of its operations. Carlos Garcia Otati, Kavak’s founder and CEO, recounts not only the technological journey from traditional processes to “copilots” and now fully autonomous agents, but also the personal and organizational challenges in navigating this generational shift. This episode is a candid masterclass on scaling tech in emerging markets, cultural adaptation, crisis leadership, and AI strategy.
Key Discussion Points & Insights
Background: Kavak’s Mission & Scale
- Kavak is an end-to-end marketplace managing a car’s lifetime—buying, selling, reconditioning, financing, logistics, and more—primarily across Latin America and recently, the Middle East.
- Handles 10,000+ monthly transactions, with 3,500+ employees after a major recent reorganization.
“We manage the end to end process of a user's journey while they own their car... we buy, recondition, sell, finance, manage warranties, services, pay their tickets...”
—Carlos Garcia Otati (01:59)
The Genesis: Personal Experience Drives Purpose
- Carlos’s entrepreneurial journey began as a teenager and traces a path from Venezuela to Oxford, Amazon, then back to Latin America.
- The concept for Kavak was sparked by personal fraud: twice defrauded in car transactions, revealing wide-scale informality and risk in the region’s used car market.
“...This is not the lemon issue that you find in the US. This is literally getting kidnapped, killed, being defrauded... the real question is, how do you show up for your users when that happens?”
—Carlos (13:51-15:22)
Building for Latin America: Multiplying Complexity & Opportunity
- Success in Latin America meant vertically integrating and building multiple businesses under one roof, as basic infrastructure—fintech rails, delivery, logistics—was often lacking.
- Kavak’s approach: treat cars as a gateway to the middle class, seeing themselves as a mix of “Spotify meets Amazon meets Toyota meets Citibank.”
“If you're going to build a business in Latin America, you need to build like 10 different businesses to make your business work.”
—Carlos (18:02-18:23)
International Expansion Philosophy
- Expansion was driven by density of opportunity (city-by-city), size of the market, and where inefficiency was greatest; success in one city (Mexico City) became the springboard for others.
“The ambition is to solve this global issue. And we're trying to pace ourselves into that.”
—Carlos (26:42)
AI Journey: From Copilots to Agents
Early Vision (2017): Robot Pipeline
- Carlos anticipated the importance of scalable automation in 2017, before generative AI became mainstream.
- Sent a company-wide memo proposing a “robot pipeline”—virtual VPs managing critical functions across sourcing, customer satisfaction, cross-selling, etc.
"2017, the transformer paper had just come out. It was still five plus years until… ChatGPT."
—Angela Strange (27:11)
“We needed to build from day one the right ontology around all of our different businesses or business lines for them to be able to communicate together.”
—Carlos (29:21)
Data, Ontology & Edge Cases
- Kavak’s vertical integration generated rich proprietary data, crucial for AI—but the real challenge was mastering edge cases (unique customer problems).
- The solution was building a framework for all business lines to communicate and exchange data—a precondition for effective AI agent orchestration.
“We're the master of edge cases. And when you're dealing with an edge case… it requires a lot of information and communication from the company to be able to solve that in a way that the user is satisfied."
—Carlos (28:49)
Copilots Fail, Agents Succeed (2022-2024)
- Initial strategy: “Copilot” tools to assist employees. These failed due to low adoption and minimal workflow improvement.
- Pivoted to end-to-end AI “agents” running entire workflows and customer interactions; transitioned gradually by focusing first on the hardest issues (e.g., underwriting, post-sale failures), not just customer service.
“We made the typical mistake that everybody's making, that is, you know, we build this copilot tools... and we realized very quickly they didn't adopt them. And when we saw that, we knew ... they were not going to do it.”
—Carlos (35:09)
The Orchestration Model
- Adopted an agent/human hybrid approach, with an orchestrator deciding which tasks are handled by agents and which by people—ramped up progressively by funnel.
- The transition required enduring a year of flat growth and some initial service deterioration, betting on eventual agent parity and then outperformance.
“We had a year... the business was growing 300%. Like 2022 was growing 100%, and 2023, we were flat... We didn't have a plan B. The plan A was we need to make this work.”
—Carlos (39:53)
Notable Strategy: Building for the Future
- “We don’t build for ChatGPT-4. We build for ChatGPT-7.”
—Carlos (41:19)
A future-proofing philosophy that ensured infrastructure would be ready to leverage advances as models improved.
The Tipping Point & Current State
- Now, over 90% of customer interactions are handled by AI agents.
- Backend processes (pricing, underwriting, invoicing) are increasingly agent-run, not just ML-assisted.
- Each new funnel followed a pattern: underperform, reach parity, surge past humans—then repeat with the next workflow.
"We just have to have the patience of it. But it took us, you know, like it was a year of flat. A lot of... under a lot of pressure."
—Carlos (41:29)
Crisis Leadership: Pressure, Culture, & Change Management
The Market Crash & Forced Bets
- Faced an auto market that “imploded”—down 99.9% during 2020-2021; at its peak, Kavak had 10,000 employees and was burning $1M/day.
- With funding evaporated, the choice was innovate or die: full steam ahead on AI, while simultaneously shrinking to profitability and scaling impact.
“It was like taking off on a plane... the worst time to slow down. You need to get yourself to like 10,000ft... we needed to slow down while we were sort of like going through that takeoff process."
—Carlos (45:01)
Culture-Building & Handling Organizational Stress
- Emphasized the existential necessity—and opportunity—of the AI “streaming transition,” using the Netflix/Blockbuster metaphor to frame urgency.
“You need to make sure that everybody understands that we're going to go streaming or die. Or die.”
—Carlos (43:24)
- The sense of building something people need, not just want, powered resilience.
“40% of our customers are buying their car for the first time in their life... we knew that we had something special.”
—Carlos (47:04)
Founder Wisdom: Building in Emerging Markets
Build for Need, Not Want
“If you're building in Latam, you have to think about building products that people actually need, not that people necessarily want.”
—Carlos (49:14)
Relentless Friction Removal & Compounding
- Focus on removing user friction 1% at a time, every single week—this compounds into generational, multi-billion dollar outcomes.
“...97% of their value gets created after year 15… making your business 1% better, making your user's friction 1% lower, that compounds over time.”
—Carlos (50:45-51:11)
Be Agnostic, Think Big
- Don’t let regional boundaries restrict ambition; go where the biggest, most meaningful opportunities are.
- Treat problems as “protein bars”—the core activity and nourishment of a founder’s journey.
“Enjoy the micro of it... and treat problems as a protein bar. Our job is literally to solve problems.”
—Carlos (52:38)
Memorable Quotes & Moments
-
Why AI agents, and why now:
“When you're building AI, the first thing that you need to build is the brakes of the system. It's the understanding on where you're going to deploy it and how you're going to deploy it.”
—Carlos (00:00; reiterated at 34:56) -
On stubborn resilience:
“We didn't have a plan B. The plan A was we need to make this work.”
—Carlos (39:53) -
On the founder's job:
“My job is cleaning cars. At the end of it, we're building all of these AI... but at the end of the day, it's delivering a car that's clean.”
—Carlos (51:39) -
Personal introspection as CEO:
“Every year or so I fire myself... I lay out if you're going to hire a CEO to build that vision... and then I ask myself: can I hire myself back?”
—Carlos (54:09) -
Changing identities as a founder:
“You sort of have to get into a character... in this moment in time, you have to get away from that character and realize what's the new Persona that is needed for this new phase, and let go of... certain beliefs.”
—Carlos (54:50)
Timestamps for Key Segments
- 01:59 — Kavak’s scale and all-in-one model explained
- 03:26 — Influences from early life, resilience, and family
- 07:51 — Carlos’s entrepreneurial journey (the lawnmowing business at age 12)
- 13:31 — Origin/founding story and uncovering a massive market gap
- 17:18 — Why Kavak is more than a marketplace: “Spotify meets Amazon meets Toyota meets Citibank”
- 27:11 — Using a “Robot Pipeline” memo in 2017: Early vision for AI
- 34:37 — Pivot from AI copilots to agent orchestration
- 39:34 — The pain and confidence during the agent transition
- 42:52 — Leading culture through a year of flat growth & high pressure
- 48:46 — Advice for founders in emerging markets
- 54:05 — The annual “fire myself” CEO self-assessment exercise
Takeaways for Listeners New to the Episode
- Carlos Otati’s journey is both a startup playbook and a tale of radical self-awareness: from hustle entrepreneurship to world-scale AI orchestration—always grounded in solving real, existential problems.
- Kavak stands as a live experiment in not just adding AI tools, but rebuilding the entire operating model around interconnected autonomous agents, with the patience, resilience, and sacrifice that comes with such a paradigm shift.
- The episode offers rich, founder-to-founder advice for startup builders everywhere, especially those tackling fundamental needs in emerging, high-complexity markets, with the message: solving true pain points, compounding improvements, and embracing change is what creates enduring value.