Interview, Fireside Chat
Block CTO Dhanji Prasanna: Building the AI-First Enterprise with Goose, their Open Source Agent
- The "Goose" system is expected to identify non-intuitive human workflows, execute tasks more quickly than humans, and operate with caution regarding tool usage while gradually increasing efficiency.
- Specific efficiency targets include saving 25% of manual company hours by the end of the year, with engineers currently reporting eight to ten hours saved weekly that are projected to rise further.
- The speaker envisions Goose eventually building competent software without user coding knowledge and rewriting its own code 100% from scratch for future releases.
- Future agent capabilities are predicted to rely on swarm intelligence comprising 50 to 1,000 instances of agents, where the combined utility of many small open-source models surpasses that of single large language models.
- The open-source model "Qwen" is anticipated to improve rapidly, and the agent middleware layer is expected to continue unlocking value over time.
- Block is positioned to benefit from AI integration, though risks exist if the company becomes passive; the utility phase of LLMs is considered to be ahead of the current moment, with the next evolution of agents expected in three years.
- A "trough of disillusionment" for AI is anticipated in 2026, with a small chance that current improvement rates for LLMs will continue, while the market is expected to experience positive surprises in AI impact by 2030.
- Companies prioritizing core customer value are forecast to succeed, whereas those chasing hype risk being left behind, alongside concerns that AI could be utilized for nefarious purposes despite its significant potential to do good.
- The future outlook includes the possibility of large-scale "flock" deployments of agents and a strategic focus on unlocking AI utility as the primary industry conversation for the coming year.