Over 95% of Coinbase’s code is now written with help of AI

We have reached a weird milestone in the crypto stack. Coinbase recently went public with a statistic that should make every junior developer sweat: roughly 95 percent of the code running at the exchange is now being written with the help of artificial intelligence. According to their leadership, the transition isn't just a pilot program; it is the new standard operating procedure. This isn't about robots taking over the board room, but it is about the death of the manual labor era in software engineering.

The End of Manual Toil

For those of us who have been around the block, we know that building an exchange is less about fancy algorithms and more about the grueling work of managing state, handling edge cases in transaction logic, and ensuring that dozens of APIs stay in sync. It is often boring, repetitive, and prone to human fat finger errors. Coinbase is leaning into AI to handle that heavy lifting. By automating the bulk of the syntax and rote implementation, they are essentially turning their engineers into editors rather than writers.

This shift matters because Coinbase isn't some small DeFi protocol running a few smart contracts. They are a massive, publicly traded entity with thousands of moving parts. If they can trust LLMs to draft 95 percent of their codebase, it means the threshold for what we consider an entry-level technical task has moved permanently. The bar for entry isn't just knowing how to code anymore; it is knowing how to direct the machine that codes.

High Agency over Syntax

Rob Witoff, a key leader at Coinbase, made an important distinction when discussing this transition. He noted that while the AI handles the volume, the company still relies heavily on high-agency human workers for judgment and strategy. This is the part most people miss when they see a 95 percent figure. It does not mean they fired 95 percent of the staff. It means the staff is now spending their time on the 5 percent of decisions that actually determine if a product succeeds or fails.

In a builder's world, agency is everything. A high-agency engineer is someone who sees a problem, understands the business context, and finds a solution without needing a roadmap. AI is the exact opposite; it needs a very specific map to go anywhere. If you are a founder, you should be looking at this as a blueprint for your own team. Stop hiring for language proficiency and start hiring for architectural foresight.

What This Means for Founders

If you are building a startup in the current climate, trying to compete with a legacy-style engineering team is a suicide mission. Coinbase is proving that a lean team using modern tools can outpace a massive department relying on manual PR reviews and hand-written boilerplate. Here is how I see this affecting the startup ecosystem:

• Speed is the Only Moat: If everyone has access to the same AI coding assistants, the advantage goes to whoever can ship and iterate the fastest. The time it takes to go from a whitepaper to a working beta has shrunk from months to weeks.
• Security Shifts Left: With AI generating code at scale, the risk of systemic vulnerabilities increases if the human oversight is lazy. We are going to see a surge in AI-driven security auditing tools to keep pace with AI-driven development.
• The Death of the Specialist: We no longer need the guy who is just a 'JavaScript expert.' We need generalists who understand how the frontend interacts with the smart contract and the database, and who can use AI to bridge the gaps in their own knowledge.

The Trust Gap

There is a massive elephant in the room here: trust. Even as Coinbase embraces this technology, they aren't blindly pushing AI code to production without human fingerprints. In the crypto world, an error isn't just a bug; it is often an irreversible loss of funds. This creates a friction point. How do you maintain the decentralized, trustless ethos of crypto while using a black-box AI to write the foundational code?

I am skeptical of anyone who says AI is foolproof. We have all seen LLMs hallucinate functions that don't exist. The real work at Coinbase right now isn't writing the code, it is the rigorous testing and validation of that code. Builders need to realize that more code does not equal a better product. In fact, more code usually means a larger attack surface. The goal should be to use AI to write more concise, efficient solutions, not just to flood the repository with more lines of text.

The New Reality for Builders

If you are currently learning to code or building a dev team, you have to accept that the old ways are dead. The 95 percent figure from Coinbase should serve as a wake-up call. We are moving toward a future where the human's primary role is to act as an architect and a safety protocol. The machine provides the bricks; the human provides the blueprint.

The value of a developer is no longer measured by their ability to write code, but by their ability to solve problems. Code is just a commodity now.

We are going to see a lot of noise about how AI is replacing jobs. In the short term, that might be true for people who only know how to perform basic tasks. But for founders and high-agency builders, this is the greatest leverage we have ever had. It allows a three-person team to act like a thirty-person team. That should be exciting, not terrifying.

Coinbase is just the first major player to admit it. Within the next eighteen months, every major tech company in the space will be reporting similar numbers. The question isn't whether or not the code is AI-generated; the question is whether the humans in charge have enough sense to know when the AI is wrong.

Read the original at Cointelegraph →