The tech industry is betting on AI coding assistants to boost developer productivity or even replace coders entirely. Non-technical leaders assume that engineering teams waste hours slowly developing code, debugging, testing and rewriting. So, if AI can write that code in a fraction of the time, it seems like a no-brainer.
The reality is that developers don’t lose time to writing code. They actually lose it to information gathering. I worked at Uber Eats as an engineer for years, and that job was more detective work than anything else. I was constantly asking questions: Which version of the API is in production? Why did this pipeline fail last week? Where is the documentation for this legacy service? I often found myself sifting through outdated documentation or hunting down answers from colleagues. The actual time spent coding, when I could manage it, was the most productive part of my day — and the part I enjoyed the most.
3 Things AI Assistants Could Do to Actually Help Engineers
- Improving documentation.
- Building better knowledge-sharing platforms.
- Creating tools that map out system dependencies.
The Problem With AI Code Assistants
AI coding assistants might introduce gains in productivity, but they don’t solve the core problem developers face. The average engineer only codes for about an hour a day. The rest of the time is consumed by information discovery — finding relevant documentation, understanding system dependencies and resolving technical roadblocks. Every time an engineer’s coding is interrupted to track down a missing piece of information, their focus breaks, and delays ensue.
Take a developer trying to implement a feature. They might spend 15 minutes writing code, only to spend hours researching how their changes could impact other systems, reading outdated documentation, searching for the owner of previous features and writing new documentation. This fragmentation means that, even when developers are technically coding, their overall productivity is still low. The mental energy spent piecing together context often leads to more time spent on tasks that could have been completed faster had the right information been available upfront.
The problem compounds with modern architecture. Each microservice adds exponential complexity to the information maze. At Uber, a simple feature deployment meant understanding dependencies across dozens of services, each with its own documentation scattered across multiple tools and teams.
Where AI for Devs Can Actually Make a Difference
AI can play a role in improving developer productivity, but focusing on speeding up coding alone isn’t the answer. The real opportunity lies in changing the way we approach information discovery. Rather than fixating on how quickly code can be written, we should focus on tools that make it easier to find, understand and share information. Improving documentation, building better knowledge-sharing platforms, and creating tools that map out system dependencies will go a long way toward reducing the time spent searching for answers.
Imagine AI as a seasoned technical advisor who knows your entire system inside and out. Rather than generating code, it could map know what services need your attention, surface relevant documentation during code reviews, and proactively flag potential issues before deployment.
For example, when you ask "How will this change impact our payment system?", it would instantly provide an interactive architecture diagram highlighting the affected components, list recent incidents in the payment flow with root causes, show deployment windows with lowest transaction volumes, identify potential race conditions with concurrent payment processing jobs, and estimate throughput impact based on benchmark data from similar changes. This analysis creates a unified view where knowledge that was previously scattered across Jira tickets, Slack conversations, wiki pages and tribal knowledge becomes instantly accessible through a single interface, dramatically reducing cognitive overhead and accelerating development cycles.
AI Should Be an Assistant, Not a Coder
AI will transform engineering productivity, but it won’t be through automated code writing alone. These tools will be most effective when developers already have a clear understanding of what they’re building and how it fits into the overall system. Without this context, AI-generated code may only add to the confusion, increasing technical debt and contributing to more complexity.
To unlock real productivity gains, the industry needs to shift its focus from simply accelerating coding time to improving the way engineers access and manage the information they rely on to do their jobs. That’s where AI can realize its true potential.
