Aug 9, 2026 · 36m · a16z
Kavak's Playbook for Rebuilding a Company Around AI
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of The a16z Show, Ale Massa, Head of AI at Kavak, breaks down how the pre-owned car giant completely restructured its architecture, workforce, and business model to become an AI-native organization.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. How this is scored →
The host as informed peer 4.8 Guest teaching 5.5 Guest disagreement 1.5 The host pushing back 1.2
speaking balance: gold is the host, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 25:50 Rejecting bottom-up hackathons
Ale directly dismisses the common corporate playbook of hackathons and bottom-up innovation, insisting transformation must be strictly top-down and vertically directed like a military command.
Hardest push from the host ▶ 13:30 Challenging AI limits in regulated financeAngela presses Ale on industry skepticism regarding AI's ability to execute fully compliant, end-to-end regulated financial underwriting without human brokers.
Biggest teaching moment ▶ 32:20 Historical electrification and redesignAle educates the hosts using the historical parallel of early dynamos and flat factory architecture to explain why surface-level AI adoption only yields marginal gains.
The host holds their own ▶ 13:40 Detailing thin-file underwriting nuancesAngela articulates the complex technical requirements of pricing, servicing, and underwriting thin-file borrowers, demonstrating deep fintech domain expertise.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The host as informed peer | Guest teaching | Guest disagreement | The host pushing back | Why |
|---|---|---|---|---|---|---|
| Ale Massa's Journey from Early ML to Kavak | 4 | 4 | 1 | 1 | Hosts Gabriel and Angela set the stage by asking about Ale's background in early ML and Kavak's operational scope. Ale explains the historical context and architecture, framing the move from early ML to transformers. | |
| Redesigning the Company Around AI and Lifetime Value | 5 | 6 | 2 | 2 | Angela highlights the risk Kavak took downsizing and rebuilding for a year, prompting Ale to detail the three strategic pillars. Ale explains why standard enterprise adoption fails without ground-up re-architecture around superhuman agents. | |
| The Importance of Evals in Scaling AI Agents | 5 | 5 | 1 | 1 | Gabriel asks about the reality of evals over agent demos at 98% operational scale. Ale outlines Kavak's eval framework and explains how evals act as the essential brakes enabling extreme execution speed. | |
| AI Agents as Superhuman Sales Representatives | 4 | 5 | 1 | 1 | Gabriel raises industry skepticism regarding whether AI agents can actually execute high-stakes sales. Ale presents data showing agents converting at more than double the human baseline across complex multi-variable transactions. | |
| AI-Driven Financial Services and Loan Underwriting | 6 | 5 | 1 | 2 | Angela demonstrates domain knowledge regarding regulated fintech, thin-file underwriting, and pricing. Ale breaks down how vertical integration and real-time customer data enable instant three-minute loan approvals. | |
| Testing an AI CEO and Human-Agent Collaboration | 4 | 7 | 2 | 1 | Ale shares surprising experimental results running an AI CEO in Cuernavaca that lifted profits by 50%. He also illustrates the physical boundary of AI by describing how mechanics use an AI sidekick for inspection guidance. | |
| Retraining Staff via Kavak's Jedi Academy | 4 | 6 | 2 | 1 | Gabriel prompts a discussion on the future of labor in AI-centric companies. Ale explains the mandatory Jedi Academy training program and the company-wide ultimatum to master agent tooling or exit. | |
| Flattening the Org Chart and Closing Feedback Loops | 5 | 5 | 2 | 2 | Angela asks how human-in-the-loop functions without traditional middle management. Ale critiques standard tiered support handoffs and explains Kavak's closed feedback loops where agents pull in human specialists via APIs. | |
| Top-Down Strategy and Measuring Token ROI | 5 | 6 | 2 | 1 | Angela addresses the organizational roadblocks legacy leaders face when deploying AI. Ale rejects bottom-up hackathon approaches and lays out a three-tier token ROI classification model. | |
| Micro Virtual Machines and Self-Improving Organizations | 6 | 6 | 2 | 1 | Angela prompts Ale on provisioning dedicated micro virtual machines per customer instead of single tasks. Ale discusses discarding their initial two-year architecture to build an autonomous self-improving organization. | |
| Creative Destruction and AI-Native Disruption | 5 | 7 | 2 | 1 | Gabriel invites Ale to elaborate on disruptive startup opportunities. Ale invokes Schumpeter's creative destruction and draws an analogy to historical electrification and factory redesign to explain why incumbents struggle. | |
| Advice for Founders and Final Thoughts | 4 | 4 | 0 | 0 | Angela closes with a call for founder advice. Ale delivers an optimistic concluding outlook on building natively around advancing AI trajectories. |
Statements from this episode (24)
Ale Massa: Kavak Had to Vertically Build Latin America's Car Infrastructure
“But to do that, we also had to build a fintech and a logistics company and the Carfax and like Basically, all the infrastructure for this to work didn't exist in, in LATAM, so we had to build everything vertically so we could serve our customers the right way.”
Kavak Spawns a Dedicated Agent with Its Own VM per Customer
“So when a customer comes in right now agent will get spawned specifically for this customer with its own virtual machine. It will remember years of interaction of these customers with Kavak, what they visited in the webpage or like Call they had two years ago.”
Massa: Long-Running Agents with Hard Goals Outperform Basic Workflows
“We realized to bet that long running agents with hard goals, not just workflows could maximize our customers satisfaction and obviously their lifetime value.”
Massa: Handing LLM tools to unchanged teams yields zero efficiency gains
“The first, and this is where I think many companies are stuck right now, is the first instinct is, okay, let's adopt AI. And you basically leave your structure as it is and just give ChatGPT or Claude to your team. And then there's no efficiencies. Your custom…”
Kavak Bet AI Agents Could Outperform Its Best Human Hires
“And this is the second bet that we made that, that we could build superhuman agents. This means that by every dimension that matters, like conversion, lifetime value customer experience, our agents would outperform the best human we had ever hired.”
Massa: Kavak assigns AI agents to most of its 10M customers
“And we moved to a relational company where now I have ten million customers in my database and I have agents assigned to most of them with the task of maximizing their lifetime value.”
Kavak AI Agents Handle 96% of Interactions and 95% of Transactions
“90, like 96% of all interactions are handled by agents. So, so no humans there. Like 95% of all transactions are completely handled by agents.”
Kavak instantiates between 100,000 and 200,000 AI agents daily
“Every day between a 102 100,000 agents get instantiated in a day. They wake up, they work sometimes for three minutes, sometimes for eight hours, sometimes for three days, and they like set an alarm clock for their next task and to go back to sleep.”
Kavak Spends Equal Time and Resources on Evals as on Agents
“We spend about the same amount of time, engineer time, tokens, and money on building the evals than building the agents.”
Massa: AI agent evals should measure conversion, not call volume or duration
“Like I see companies like measuring number of calls or minutes during the call or some like superficial KPIs that give you some information, but that doesn't really work. Like the important thing is Did this customer convert? Is it bringing value to the custom…”
Massa: Deploying AI sales agents tripled Kavak NPS and CSAT scores
“We tripled NPS and customer satisfaction score by putting the agent in front of the customer.”
Massa: Kavak AI sales agents convert 2.1x better than human teams
“And it, at first it converted, like, 50% more than our human team, and now it's converting over that, like, 2.1 x more.”
Massa: Kavak approves car loans in under three minutes
“And usually in, in Mexico and in some emerging markets, it'll get like two months or more to get a car loan approved. We usually approve it in under three minutes which is like pretty cool because we have all this data around the customer and the car.”
Massa: AI will likely be capable of CEO roles by 2035
“Is, is AI going to be able to do this job, like even the CEO job or jobs where the leadership is? And the answer honestly is probably yes. Like in 2035 with a rate of improvement, it will be able to do so.”
Kavak: Experimental AI Branch Manager Increased Profits by 50%
“The goal of the first month was to double the profits of Cuernavaca. It didn't reach it, but it was 1.5 X, like 50% more profits just by managing the city, which is, it's crazy, right?”
Kavak: Mechanic AI sidekick reduced warranty claims by 20-26%
“Warranties came down around like 20, 26% since we launched and customer satisfaction again went up.”
Massa: Kavak Mechanics and Finance Staff Ship Production AI Agents Within Weeks
“So we train everyone, and after six weeks, they launch state-of-the-art agents, AA agents to production. And it's mechanics and finance guys and engineers, like, everyone can do it.”
Massa: Kavak Uses AI Agents as Managers Over Human Workers
“And then if you look at Kavak now, any process, it's really a collaboration of, Agents and humans, and sometimes like agents are the bosses or of humans, and sometimes humans are designing the agents.”
Massa: Kavak's org consists of teams building, supporting, or executing for AI
“The way it looks like now is very flat teams, very senior teams, super empowered. If you look at a team, you'll have engineering, AI, like operations, like everything. And they're either building the agents, working for the agents, or being in the physical wor…”
Handing AI Failures to Tier-Two Support Fails Without Closed Feedback
“Most of these agentic systems in production right now, like large scale agentic systems, usually if an agent hits a wall or can't perform anymore, it'll like send the, this case or this customer to a tier two support and forget about it. That doesn't really wo…”
Massa: Enterprise AI transformation must be top-down, not bottom-up hackathons
“The first is it has to be top down because of this. Like, if you just get adoption, it won't go anywhere because it's hard to generate this taste or strategy for people to bottom up, decide what to build and whatnot, and come up with something that works for t…”
Kavak measures the specific ROI of individual tokens used by AI agents
“Level tier three tokens, the most valuable. Are this agents where you can get the ROI of each specific token? And I can do that now. That's great news for me because I'm growing. And because I know the ROI of each token, because it goes to agents that are perf…”
Massa: Multi-Agent Graph Harnesses Constrain Advanced Model Intelligence
“The intelligence now doesn't need like the graph and the multi-agent lattice work and harness because it will constrain this level of intelligence.”
Kavak Scrapped Two Years of AI Infrastructure for Newer Models
“So we decided to like destroy everything we had been building for two years that was working, that brought us to profitability, that brought us amazing growth. And start over with a harness that we thought would be robust and scalable and leverage recursive se…”