Deep Dive · Xiaohu's Take

Kavak rebuilt its entire company around AI. Now 96% of customer interactions are handled by AI agents.

Kavak didn't bolt AI onto its existing business. It asked what the company would look like if built from scratch today, then tore down its profitable operation to rebuild it that way.
The 60-Second Version
  • Most companies ask how to add AI to their organization. This one asked what the company should look like if built from scratch today.
  • Agents handle 96% of customer interactions and 95% of transactions, with 100,000 to 200,000 agents spun up daily.
  • A profitable, two-year-old architecture was scrapped and rebuilt because of a single new model release.
⚑ Content is from a16z's own podcast; the interviewee is Alejandro Maza, Kavak's Chief Product & AI Officer. One relationship to note upfront: a16z led Kavak's $300 million funding round in February 2026, and the show's disclaimer says its affiliates may hold investments in the companies discussed. All performance figures below are company-reported and have not been independently verified.
The Starting Point

No tinkering at the edges—rebuild the org structure itself

Most companies adopt AI the same way: the org chart stays frozen, everyone gets a ChatGPT or Claude account, and they wait for productivity to magically appear.

The usual outcome: no real efficiency gains, customers still face the same problems, nothing changes. It's easy to see why. Processes, interfaces, and KPIs were all designed for human execution. Drop a new tool in, and it can only help with the occasional task in the margins.

Kavak asked a different question: If it's 2035, and AI is as powerful as it's going to be, what would this company look like if we started from zero?

Following that logic, the resulting company bore almost no resemblance to the one they'd already built. Different processes, different team structures, a completely different customer journey from first click to final payment. So they made an uncommon decision—rather than incrementally changing the existing company, they rebuilt it to match that vision.

The Standard Question

How do we add AI to our existing org?

→ Keep the structure, hand out tools, wait

Kavak's Question

What would this company look like if built from scratch today?

→ Rebuild the current company to match that vision

Quick context: who is Kavak?

Kavak is Latin America's largest used-car platform. Founded in 2016, it was Mexico's first unicorn, surpassing a $1.15 billion valuation in October 2020 and peaking at $8.7 billion in 2021, making it briefly the most valuable startup in the region. It operates in Mexico, Argentina, Brazil, Chile, Colombia, Peru, Turkey, and the UAE.

But Kavak isn't just a car-listing website. Because Latin America lacked ready-made infrastructure, it built its own financing, logistics, and vehicle history databases, similar to Carfax in the US. It's a vertically integrated company. This matters later: because they own the data and the entire chain, they can do things most competitors can't.

96%
Customer interactions handled by agents, with no human in between
95%
Transactions fully completed by agents (though a human still hands you the keys at pickup)
100K–200K
Agents launched daily, each with its own virtual machine
Full video of this interview, 36.5 minutes. Source: a16z's official channel.
The big picture

Six takeaways and five stealable frameworks

Before diving deeper, here are the most valuable insights from this episode. Six takeaways:

1
Don't hand out tools—rebuild the entire company around agents. This includes rewriting most APIs and systems so agents can actually use them.
2
The architecture assigns one agent per customer, not one agent per task. Each agent has its own VM, memory, and long-term goal, with 100,000 to 200,000 launched daily.
3
A profitable, two-year-old architecture was scrapped because of a new model release. The reasoning: the orchestration layer was capping what a sufficiently smart model could do.
4
They invest as much in evals as they do in building agents. Engineer time, tokens, and money are split roughly 50/50.
5
The AI sales team converts at 2.1x the rate of the human team; an AI acting as a city's CEO increased monthly profit by 50% in the first month.
6
Six-week training for everyone, from executives to mechanics. The final project is getting a current state-of-the-art agent deployed to production.

Five frameworks you can copy verbatim

Eval vs. Dev resource ratio The three-metric eval rubric The three-tier token framework The four-part long-running agent setup The two prerequisites for org transformation

Details below. The ordering matters: the first few points are about building the foundation, the later ones about what happens when you apply that foundation to business and org structure.

The Core Decisions

Three foundational decisions: rewriting APIs, setting superhuman standards, changing KPIs

Everything stems from three decisions. None are about tech stack choices; they all answer the same question: What are we trying to become?

1. Redesign the company, don't just hand out tools

On an engineering level, this means rewriting most APIs and systems so agents can use them to get things done. It was a necessary step because interfaces people can tolerate are often unusable for agents. A workflow that requires a person to guess, click around, and copy-paste between three systems simply can't be executed by an agent. If they hadn't changed the interfaces, agents would've been stuck on the sidelines.

The next step was building the data and feedback loops needed to train these agents. How? By releasing them to face real customers, gathering data, getting eval results, and then training.

2. Bet on building superhuman agents.

"Superhuman" has a concrete definition here: on the dimensions that matter—conversion rate, customer lifetime value, customer experience—the agent must outperform the best human the company has ever hired. And it needs to be pointed at the hardest problems, not the easy scraps.

3. Shift from a transactional company to a relationship company.

Kavak used to be a transactional business, measured by cars bought, cars sold, and brake pads procured. Now it's a relationship company. There are 10 million customers in its database, most with an individual agent whose goal is to maximize that customer's lifetime value.

This shift works for Kavak because of two simple calculations. First, the money: they sell high-ticket items like cars and large personal loans. Activating just 1% of those 10 million people correctly is worth hundreds of millions of dollars. Second, the nature of the industry: buying a used car requires trust before a customer will spend. And trust is built by actually knowing the customer and managing the relationship long-term—which is precisely what these agents are good at.