Descriptions:
Angie Jones, head of AI enablement at Block, recounts the company’s two-year effort to transform its 3,500-person engineering organization into what she calls an “agentic engineering org” — one where AI agents serve as primary collaborators rather than productivity add-ons. Block’s journey began early: the company built Goose, an internal coding agent, before LLMs even supported tool calling, and worked with Anthropic as a design partner on MCP’s initial release, with Goose becoming the reference MCP client implementation.
Despite 90% of engineers adopting tools like Goose and Claude Code, Block’s CEO observed no shipping acceleration — a gap Jones traces to engineers using AI passively inside their IDEs rather than delegating meaningful work. Her response was a five-stage maturity model measuring how deeply engineers delegate to agents: from autocomplete use (stage 1) through running parallel agents (stage 4) to fully hands-off task delegation producing shippable output (stage 5). Applying the 1-9-90 community participation rule, Jones built an “AI Champions” program spanning Square, Cash App, Afterpay, and Tidal — a small group of power users who developed and distributed patterns rather than requiring every engineer to self-educate.
Practical outputs include a standard set of repo assets for making codebases agent-ready: agents.md and claude.md context files, rules files for guardrails, slash commands, AI code reviewers with custom instructions, and PR-level AI attribution. Jones details how monorepos, mobile apps, and web services each required distinct approaches, and how bottom-up pattern discovery — rather than top-down mandates — drove genuine adoption.
📺 Source: AI Engineer · Published June 28, 2026
🏷️ Format: Keynote Launch
