The AI Throughput Illusion: Why Expensive Models Won't Ship Your Code

The AI Throughput Illusion: Why Expensive Models Won't Ship Your Code

Noah Labhart | Technical Founder & Startup Mentor

Related Episode

S12 E28: The AI Throughput Illusion: Why Splurging on Expensive Models Fails to Ship Code and How to Measure Real Engineering Output with Emilie Schario, Co-Founder & Head of Product & Engineering at Kilo Code

There’s a dangerous myth spreading through startup engineering teams right now: if you just pay for the most expensive, most powerful AI model available, you’ll ship code faster and better. Emilie Schario is here to tell you that’s not how it works.

When Emilie joined Kilo about a year and a half ago as Co-Founder and Head of Product & Engineering, the company was already working on an ambitious mission—building the fastest AI coding harness on the market. But the more she worked with teams trying to use AI coding assistants, the more she realized the real bottleneck wasn’t raw model power. It was flexibility, measurement, and honest conversations about what actually ships code.

The Model Freedom Revolution

Here’s what makes Kilo different from tools like GitHub Copilot or Claude Code: you’re not locked into one AI provider’s ecosystem.

“When you use a tool like Copilot or Cloud Code, you’re using software that’s tied to a particular lab,” Emilie explains. “If you’re using Cloud Code, you can only use Anthropic models. If you’re using Copilot, you’re stuck with OpenAI models.”

Kilo flips this on its head. The platform gives developers access to over 500 different models—OpenAI, Anthropic, DeepSeek, Grok, and dozens more. This isn’t just about choice for choice’s sake. It’s about choosing the right tool for the right job.

Think about it: not every coding task needs the most expensive, most capable model. Sometimes you need speed. Sometimes you need cost efficiency. Sometimes you need a model that’s particularly good at a specific type of problem. The teams using Kilo understand this intuitively, and they’re shipping code faster because of it.

The MVP That Keeps Moving

One of the most honest things Emilie says in the episode is this: “Every day at Kilo is about figuring out what the MVP for coding harnesses is—because the measuring stick is definitely moving.”

This is the reality of building in the AI space right now. The baseline expectations change constantly. Two months ago, a viral tweet from Peter Steinberger shifted the entire industry’s thinking: “If you’re writing prompts, you’re doing it wrong. We’re all about loops now.” Within a week, every serious coding tool had to support loops. That’s the pace of change we’re dealing with.

For Emilie and the team, the first version of Kilo was simple: Can you take a prompt and have it actually change code? But now? Now they’re constantly iterating to include the features that make it possible for people to use Kilo every single day to do most of their work. It’s a moving target, and they’re sprinting just to keep up.

The Hard Choices: When to Build and When to Kill

Here’s where Emilie’s pragmatism really shines through. In the second half of last year, Kilo was trying to broaden its reach—not just serving developers, but all knowledge workers. It sounds ambitious. It also sounds like a recipe for spreading yourself too thin.

“We had to confront our limited resources against that goal,” Emilie admits. “We only have so many engineers. We can only tackle so many projects if we’re going to do them well.”

One casualty: Kilo App Builder, a feature that let users go from prompt to hosted website. It worked. People built incredible things with it. But it wasn’t getting the investment it deserved, and Emilie made a tough call: either keep it alive as a mediocre experience or be honest about sunsetting it.

They chose honesty. They removed it from marketing pages, hid it in the app for new users, and started planning a graceful migration for existing users. It’s not the sexiest product decision, but it’s the right one. And it’s the kind of decision that separates companies that scale from companies that implode under the weight of half-finished features.

This is a lesson every founder needs to hear: your resources are finite. Your energy is finite. You have to be willing to kill things, even good things, if you can’t do them justice.

Product Management in the Age of AI

Emilie’s philosophy on building the roadmap is refreshingly different from traditional product management. There’s no six-month plan. No five-year vision cast in stone. Instead, she’s juggling:

  • What enterprise customers are asking for
  • What would unblock deals in the pipeline
  • What the community is asking for
  • How the team is feeling about the work ahead

But here’s the kicker: she’s also rethinking what product management even means now that engineers can do more than they’ve ever done before.

“Engineering is no longer the bottleneck,” she says. “That means engineers’ jobs are going to change, and product managers’ jobs are also going to change.”

The new priority? Understanding what features will actually build a business. What will people pay for? That’s the question that matters now. Everything else is secondary.

Building a Team That Doesn’t Slow You Down

Kilo is about 35 people, globally distributed across North America, South America, Europe, and Africa. And Emilie is unapologetic about having a high bar for hiring.

“There’s not politics. There’s not any of the things that create you down like the things that create you down in a company with B players,” she says.

The structure is lean: product and engineering led by Emilie, go-to-market led by co-founder Scott. Everyone operates with the same rhythm—daily standups, ambitious goals, and product engineers who have the autonomy to move fast.

The best part? The power users of Kilo are the people building Kilo. They’re using their own product every single day. That’s how you catch problems early. That’s how you stay honest about what actually works.

The Real Measure of Engineering Output

Throughout this conversation, what becomes clear is that Emilie is obsessed with measurement. Not vanity metrics. Not lines of code written. But real, tangible output: code shipped, features launched, customers served.

The AI throughput illusion—the belief that expensive models equal faster shipping—falls apart under this kind of scrutiny. What matters is choosing the right model, building the right features, and having a team aligned on what actually needs to get done.

For technical founders building in the AI space right now, this episode is essential listening. Emilie’s approach to product decisions, team building, and navigating a market that’s changing faster than anyone expected is a masterclass in startup pragmatism.

Ready to hear the full story of how Emilie and the team at Kilo are redefining what’s possible with AI coding agents? Listen to the full episode now to dive deeper into the engineering decisions, the pivots, and the philosophy that’s driving one of the most innovative tools in the AI coding space.