Skip to main content

Larridin raises $30M to continue to help enterprise organizations understand the impact of their AI investment   Read the Press Release      Watch the Video

A senior engineering leader told our CTO, Ameya Kanitkar, that all work on her team stopped one day. The reason surprised him: they'd run out of AI tokens.

He explained his thinking on LinkedIn:

Read Ameya's post on LinkedIn

Ameya Kanitkar, co-founder and CTO of Larridin, on LinkedIn:

A senior leader told me why all work on her engineering team stopped one day. I wasn’t expecting the reason.

They had run out of AI tokens.

I would have expected things to slow down. Engineers would go back to working manually until access returned.

Instead, she told me, everything stopped.

That conversation made me think about how much we now rely on agents beyond just writing code.

Many engineers now regularly ask coding agents to locate a feature, trace a request, explain a dependency, and figure out where a change belongs. The agent is becoming how they navigate the codebase.
These engineers still know how to write code. But without an agent, they may not know where in the codebase to make the change.

I discussed this with 30 or 40 senior engineering leaders, and several recognized the pattern in their teams. Engineers were increasingly relying on agents to understand the system as well as modify it. There is a tradeoff here that is easy to miss. When you search through files, follow a call chain, and debug something yourself, you gradually build a mental map of the system.

When an agent does that exploration, you can finish the task without developing the same understanding.

That can be a very productive way to work. It also means familiarity with the codebase may stop growing alongside the code.

Losing access to AI then starts to feel like losing electricity or Wi-Fi. Even figuring out what to do next depends on the service coming back.

The conversations made me wonder how many engineering organizations have developed this dependency without explicitly deciding to.

We planned for AI to become part of how engineers work. I don't think we planned for running out of tokens to mean all work stops.

Why This Matters if You Run an Engineering Team

You'd expect losing AI access to slow a team down. Engineers would grumble, then fall back on doing things by hand until access came back. Work stopping entirely is a different story, and it says a lot about how engineers work now.

Agents Have Become the Map of the Codebase

Plenty of engineers now ask coding agents to find a feature, trace a request, explain a dependency, and figure out where a change belongs. The agent has become their way of navigating the codebase.

These engineers still know how to write code. Without the agent, though, they may not know where to make the change in the codebase. Ameya raised this with 30 or 40 senior engineering leaders, and several recognized the pattern on their own teams.

The Tradeoff That’s Easy to Miss

When you search through files, follow a call chain, and debug something yourself, you slowly build a mental map of the system. When an agent explores the codebase for you, you can finish the task without developing your own understanding of how the system works.

That can increase productivity. It also means familiarity with the codebase may stop growing alongside the code itself. And once that happens, losing AI access starts to feel like losing electricity or Wi-Fi.

"Claude Told Me This"

In another post, Ameya shared a story from a leader who asked an engineer why there was a bug in the code. The engineer's answer: "Claude told me this."

As Ameya put it, there's a fine line between using AI to speed up your thinking and using it to replace your thinking. The code may work and the explanation may sound plausible. But when the system does something unexpected, someone still has to understand what's happening and why.

That leader's team started experimenting with "No AI Hours." They brainstorm without AI, develop their own hypotheses, and challenge each other, then bring AI back in to test their thinking and take it further. The goal is to keep using AI while making sure it sharpens human judgment.

There's an accountability angle here too. Engineering leaders promote people, reward them, and occasionally let them go. None of that applies to a model. AI can contribute to a decision, but the human is still where the buck stops.

Closing Thought

Ameya wonders how many engineering orgs have built this dependency without ever deciding to. A few questions worth asking your team:

  • What would happen if we lost AI access for a day?
  • Can engineers explain why the code they shipped works?
  • Do we have any deliberate practice, like No AI Hours, that keeps people's mental map of the system growing?

Ameya shares this kind of practical, thought-provoking stuff on LinkedIn pretty regularly. If you want more of it in your feed, follow Ameya Kanitkar on LinkedIn. He's co-founder and CTO of Larridin.

On larridin.com: