x
The landscape of software development is undergoing a seismic shift, and crypto giant Coinbase is leading the charge. In a move that highlights the staggering speed of artificial intelligence adoption, the company has revealed that the vast majority of its software code is now generated with the assistance of Large Language Models (LLMs).
This transition represents far more than just a minor upgrade in developer tools. It signals a fundamental restructuring of how tech companies build, test, and scale their platforms. Let’s dive into how Coinbase achieved this, what it means for the future of engineering, and why "human judgement" has become the most valuable asset in the tech industry.
Only a few months ago, in February, Coinbase estimated that roughly 40% of its code was being written using AI. By July, that figure skyrocketed.
According to Rob Witoff, Coinbase’s Head of Platform, AI has permeated every single layer of the company’s daily operations. Speaking on the rapid integration of these tools, Witoff noted:
"Effectively, 100% of our employees are using AI on a daily basis here. And close to 100% of our code, probably somewhere between 95% and 100%, is written by or with LLMs today."
This rapid evolution aligns with a broader mandate from Coinbase CEO Brian Armstrong. Following a decision to reduce the company's workforce by 14% (approximately 700 staff members), Armstrong emphasised the need to leverage technology to regain agility. He expressed a desire to "return to the speed and focus of our startup founding, with AI at our core."
While a headline declaring that "95% of code is AI-written" might conjure up images of autonomous machines running a financial platform unsupervised, the reality is highly nuanced. Coinbase employs a strategic spectrum of AI integration, depending entirely on the sensitivity of the task at hand.
1. High-Security Cryptography (Heavy Human Input)
When it comes to the core cryptographic systems that secure billions of pounds in user assets, human precision remains irreplaceable. AI acts as a secondary verification tool rather than the creator. Witoff explained:
"For example, when we’re writing core cryptography, we have industry-leading cryptographers that are meticulously researching and reviewing one line at a time."
He added that while AI is heavily utilised to check for vulnerabilities, test edge cases, and verify mathematics, this area remains a highly manual, human-driven process.
2. Core Infrastructure (Balanced Collaboration)
For middle-tier systems, developers work hand-in-hand with AI agents. The AI suggests structures, drafts functions, and automates boilerplate code, while senior engineers orchestrate the logic and guide the architecture.
3. Internal Prototyping (Fully Automated)
If an engineer wants to test a new feature or spin up an internal concept, they can let the AI run wild. This stage of development has transitioned to a nearly frictionless, automated pipeline where ideas are translated into functioning code in minutes.
Perhaps the most disruptive consequence of this technological leap is how it alters the composition of engineering teams. Historically, software teams required large cohorts of junior developers to handle basic coding tasks, bug fixing, and documentation. Today, those entry-level tasks are easily swallowed by LLMs.
Coinbase’s shift has allowed them to reorganise around much smaller, flatter, and more senior teams. Today, a tight-knit group of two or three senior engineers can accomplish what once required a team of ten or more. However, this model relies entirely on having highly experienced leaders at the helm.
Witoff highlighted this shift, stating:
"There were a lot of junior development roles that were impacted... For those smaller teams to work, for people to have the taste, the judgement, I think a lot about people having the battle scars so they know how to point agents in the right direction."
In this new paradigm, the role of a software engineer shifts from being a manual writer of code to a "tastemaker" — an editor, architect, and strategist who possesses the deep industry experience required to guide AI agents effectively. Currently, Coinbase engineers manage between five and ten AI agents simultaneously. Collectively, these digital agents are doing the equivalent coding work of roughly 1,200 employees. Looking forward to 2030, Witoff predicts that AI agents could perform the equivalent work of 100,000 employees.
Coinbase is far from alone in this transition. The broader technology and Web3 sectors are experiencing a massive wave of restructuring as companies pivot to flatter, AI-driven models.
Earlier this year, Block CEO Jack Dorsey announced significant workforce reductions of up to 40% to streamline operations, sharing on X:
"We’re already seeing that the intelligence tools we’re creating and using, paired with smaller and flatter teams, are enabling a new way of working which fundamentally changes what it means to build and run a company."
Other prominent industry players, including Crypto.com, Kraken, Gemini, Messari, and Dune, have similarly cited AI-driven efficiencies during their respective corporate restructurings.
The Coinbase revelation makes one thing incredibly clear: the ability to simply "write code" is quickly becoming commoditised. The future belongs to those who possess the strategic vision, critical thinking, and domain expertise to orchestrate AI tools safely and creatively.
As software engineering evolves, the developers who thrive won't be the ones who write code the fastest, but the "tastemakers" with the battle scars to guide the machines.
For more details on this story, you can read the original coverage on Cointelegraph:
👉 Over 95% of Coinbase’s code is now written with help of AI
Disclaimer: This article is provided for informational purposes only, mistakes may be made, and it's not offered or intended to be used as legal, tax, investment, financial, or any other advice.
