Keep pulling the thread on Dhanji R. Prasanna.
At Block, engineering teams that are highly AI-forward report saving 8 to 10 hours per week by using tools like the company's internal agent, Goose.
Danji Prasana asserts that there is no correlation between code quality and building a successful product, citing YouTube's success despite its initially poor codebase.
Block's transition from a General Manager (GM) organizational structure to a functional org structure was a key enabler of its transformation into an AI-native company.
Block's number one priority is to "automate Block" by implementing AI-driven automation across the entire company.
Block estimates it is trending towards saving 20-25% of manual work hours across the entire company, including support, legal, and risk teams, due to AI adoption.
Block's AI agent, Goose, is an entirely open-source project that other companies can download and extend.
Danji Prasana is pushing his teams at Block to explore a future development paradigm where applications are completely rewritten from scratch for every new release, enabled by AI.
An engineer at Block has an experimental setup where the AI agent Goose proactively builds features discussed in Slack or email and opens pull requests on Git.
Block utilizes background AI processes that run 24/7 in its CI pipeline to analyze vulnerabilities and attempt to build patches for bugs while engineers are asleep.
At Block, the most surprising and impactful use of AI tools has been by non-technical employees, such as the Enterprise Risk Management team, building their own software.
Block has developed a tool called Gosling, an AI agent for mobile that operates the Android OS at a native level via the accessibility API to automate UI tests.
Danji Prasana believes that AI currently underperforms senior human engineers in complex tasks like system architecture, design, and handling race conditions.