**Subscribe to stay ahead of technology trends.** **Never miss future editions.**
Rob Witoff, CTO of Coinbase, joins us in his first interview since returning to the crypto company. He explains why 98–99% of Coinbase’s code is now written by agents, how his engineers went from writing code by hand to supervising 3–10 agents each in a single year, and why a Friday-morning “spicy feedback” agent may be giving Coinbase bigger productivity gains than AI in product development.
In our in-depth conversation, we discuss:
\* Why Coinbase engineers are producing 100x the code they did a year ago, and why agent managers of agent managers are coming next
\* Why it’s now cheaper to build a prototype than to ask for permission, and why a picture is worth a thousand words but a prototype is worth a million
\* The weekly AI feedback agent every Coinbase employee receives, and the “spicy” section that tells Rob to get out of the weeds
\* Why crypto is so volatile, and why "it's never as good or as bad as it seems
\* Will Coinbase become the financial super app?
***Watch or listen now across*** *YouTube****,*** *Apple Podcasts****,*** *Spotify****, and*** *X*
***Download the transcript 👇***
Timestamps
(0:00) Meet Rob Witoff(2:30) Crypto's Understated Opportunity(4:09) Coinbase's Next Opportunity(5:30) Inside Coinbase's CTO Role(8:10) Managing a Third of the Company Under Engineering(10:28) Taste & Judgment in the Modern Era of Engineering(13:33) Inside Coinbase's Engineering Team: Zero to One(15:35) How to Set Up Engineers for Success(18:28) How Rob Sets Up His Agents(20:17) How Disruptive Are Model Updates(22:51) Would Coinbase Ever Launch Its Own AI Model?(26:12) Agentic Trading(26:52) Why Are Markets So Volatile?(29:02) What Is the Super App Opportunity?(31:20) Why Now Is the Moment for Agentic Financial Services(34:38) The Fundamentals of Security(35:38) The Agentic Consumer Experience(41:34) Prediction Markets(43:52) Stablecoins(45:45) Will Stablecoins Be Country-Owned?(47:35) The New Financial Infrastructure(49:52) How Coinbase Structures Internal Feedback(54:16) Ollie Joins as Rob's Chief of Staff(57:41) Has 2026 Been the Inflection Point for AI?
Lessons from this episode with Rob
**1. Why 98–99% of Coinbase’s code is now written by agents, and why engineers are becoming supervisors of loops and systems**
A year ago, Coinbase engineers were writing code by hand. Today almost all of it is written by agents, and Rob says he didn’t predict it would change this fast.
\* Most engineers now supervise three to five agents, and sometimes 10, throughout the day. Rob calls that a very different muscle to exercise.
\* The next step is supervising whole loops and systems, which shifts the job to deciding what problems should be solved in the first place.
\* Coinbase is building a “Coinbase brain” that holds its product and architecture strategy and directs its agents and loops.
**2. The weekly AI feedback agent that tells Rob to get out of the weeds**
Every Friday at 9am, everyone at Coinbase gets a full assessment of how they’re working, built from their Slack messages and transcribed meetings. It covers what they’re doing well and where their gaps are.
\* Rob’s favorite part is the “spicy feedback” section, which asks: what is the really hard feedback you need to hear?
\* People might not want to say “Rob, get out of my way,” but his agent isn’t afraid to say it.
\* Lessons that used to take him six months to learn, he now iterates on every week.
**3. Why it’s now cheaper to build a prototype than to ask for permission**
Rob doesn’t want people coming to him or their lead with an idea. He wants them to go and build it, and that applies to 100% of the company, including non-engineers.
\* During one tax season, people were debating how to fix a confusing tax form. A non-engineer built a fully working prototype, and it immediately became the thing they built.
\* A picture is worth a thousand words, but a prototype is worth a million.
\* Once people can see, touch and feel something, it gets far more attention and momentum.
**4. How Coinbase builds taste and judgment in engineers**
Rob says taste in modern engineering means thinking like an entrepreneur about what you’re building and why. That starts with being a daily user of your own product.
\* Employees can now direct deposit their pay into the Coinbase app, a small feature that makes it easy for everyone to use the product and find its rough edges.
\* Judgment comes from battle scars: you have to make mistakes to learn the big lessons.
\* Large companies can’t let staff make mistakes that hit customers, so Coinbase builds safe environments where engineers can break things without customers being affected.
**5. Why Coinbase flattened its layers so the engineer doing the work is in the room**
Large companies have traditionally had many layers, from manager to director to senior director and up. Rob says that is good for coordinating but slow for executing.
\* Over the last year, Coinbase deliberately removed layers between decision-makers and the people doing the work.
\* His rule of thumb is that for any big architectural decision, the engineer who will execute it is in the room when it’s discussed.
\* This removes the games of telephone that have historically made it hard to execute at scale.
**6. How to trust an AI agent with your money: read-only first, hard limits second, human sign-off third**
Rob describes a path that applies to almost any use of agents. Agents start with read-only access and earn more control over time.
\* Even read-only access can show you how your portfolio is allocated and whether it still matches your risk tolerance.
\* Once agents can trade on your behalf, they need hard limits on which funds they can move.
\* Escalating to a human has to be a core part of the workflow, so the agent never makes an assumption you wouldn’t make.
Where to find and connect with us
Follow Ollie on X: https://x.com/ollieforsyth
Follow Rob on X: https://x.com/rwitoff
Visit Coinbase: https://coinbase.com/
Our partner for today's episode is Harmonic, your go-to startup database: https://harmonic.ai
Previous episodes include
See all previous episodes here 👉
If you enjoyed this episode, help sustain our work by clicking ❤️ and 🔄 at the top of this post.
Get full access to NEW ECONOMIES at www.neweconomies.co/subscribe