Dhanji R. Prasanna on Block, AI-Native Engineering, and the Code Quality Myth
Source: Lenny’s Podcast Speaker: Dhanji R. Prasanna Date: ~2024 Link: Episode
Dhanji R. Prasanna is CTO at Block (parent of Square, Cash App, Afterpay, TIDAL, and Bitcoin). Previously head of engineering at Cash App, scaling it from 10 to 200+ engineers and ~10–20M users; prior to that, worked on Google Wave and Gmail. This conversation covers how Block restructured its engineering organisation for the AI era, the open-source Goose agent, and two counterintuitive lessons from a career spanning Google and Block.
Key ideas
-
Conway’s Law as competitive weapon. Block’s decisive move was a Jobs-style restructuring: all engineers under a single engineering leader, eliminating silos between Square, Cash App, Afterpay, and TIDAL. The result — 20–25% manual hours saved company-wide — exceeded the contribution of AI tooling alone. “You ship your org structure”: Conway’s Law can be imposed deliberately to produce the product architecture you want, not just the one your org accidentally creates.
-
Goose: open-source AI agent for engineering. Built on MCP (Model Context Protocol), Goose started as one engineer’s side project and is now fully open-source (any company can adopt it). AI-forward engineers report 8–10 hours saved per week; non-technical workers (enterprise risk, legal) are the biggest beneficiaries, building their own tools without engineering involvement. Median session runs ~5 minutes; Block is pushing toward hour-long autonomous sessions. 60% autonomous feature completion rate.
-
Code quality and product success have nothing to do with each other. YouTube: videos stored as blobs in MySQL with a Python stack. Google Video: superior C++/Java architecture. YouTube won. Code quality is a red herring; the only questions that matter are why you are building something and for whom. Engineers who conflate craft with product success spend their careers on the wrong problem.
-
Start small as a forcing function. Bitcoin on Cash App: Jack Dorsey + Dhanji + one engineer, hackathon. Cash App itself: hack week project. Google Wave failure: started with 70–80 engineers before a single user. The pattern is consistent — small beginnings allow fast feedback and genuine discovery; large beginnings optimise for the wrong things.
-
LLMs should work overnight and at weekends. The productivity floor is rising: “this is the worst it will ever be.” Block runs experiments from wishlists overnight while engineers sleep. The eventual vision is rm -rf and rebuild from scratch with AI. Current AI still underperforms humans on architecture, race conditions, and orchestration — human taste is required to prevent “AI slop.”
Context
Block employs thousands of engineers across multiple product lines. Dhanji has been at Block/Square since the early Cash App days, spending 4.5 years as “Mad Scientist” part-time at Square before joining full-time. The Goose agent is available publicly; Block’s AI-native transition is documented via internal engineering blogs and this episode.
Related
- Agentic Engineering — Goose and the overnight-work model are direct expressions of this concept
- Conway's Law — Block’s functional restructuring is the canonical deliberate application
- quotes — see quotes page
- Asha Sharma on AI's Impact on Product Management — adjacent view on agentic product orgs
- Amjad Masad on Replit and the Future of Software Development — complementary view on AI-native engineering
- Crystal Widjaja on Growth, Data, and Building Smarter — org structure and product velocity parallels