A walkthrough of a real session where two developers and their AI agent teams collaborated on a complex architectural change — going
from a rough spec to 8 PRs, 4,500 lines of code, and a clean task board in a single day.

The talk covers how we used git worktrees as a shared context surface between two developers' Claude Code agents, spawned specialist
subagents (including a Redis expert that caught two critical production bugs during design review), ran parallel implementation
across multiple branches, and coordinated via a dedicated agent that polled for commits and monitored Slack and CI.

The focus is on the collaboration model — not the technical feature itself — and what the human's role becomes when AI agents can
read, write, test, and coordinate autonomously.

Chat interfaces are everywhere, but what challenges are there to overcome to create a truly joyful user experience? In this talk, we’ll explore the key problems and solutions across the entire stack: from the tools you can give your agent to better respond to real human behaviour, to the underlying infrastructure and optimisations required to handle client disconnects, cancellations, and multi-device usage gracefully.

Format: Technical talk with live demos, code and prompt examples

Description:

The role of the developer is fundamentally changing. We're moving from being executors who write every line of code to becoming orchestrators who conduct AI agents to build complex systems. This talk, based on my essay about transitioning from traditional coding to AI orchestration, shares practical insights from a year of experimenting with multi-agent development workflows. 🔗 Essay linked here: https://pivotech.substack.com/p/from-executor-to-orchestrator-my

Through real code examples and live demonstrations, I'll walk through my evolution from using ChatGPT for learning CS50 concepts to orchestrating Claude, Gemini CLI, and NotebookLM to build complete products. You'll discover the three distinct schools of AI development I've identified through hands-on experimentation: the One-Shot method, the Incremental approach, and my hybrid Layering technique.

I'll share the workflows I use to go from customer discovery sessions to deployed applications, including the mistakes, frustrations, and breakthroughs that shaped my approach. We'll explore the "Legacy Codebase Problem" that emerges from AI-generated code, the "Hyper-specificity Paradox" of detailed prompting, and the new skill set required to become an effective AI orchestrator.

Key Takeaways:

Three proven patterns for AI-assisted development and when to use each

Practical orchestration workflows for complex projects

The emerging skillset of the developer-orchestrator

How to maintain technical depth while leveraging AI efficiency

Real-world pitfalls and how to navigate them

Target Audience: Developers looking to evolve their practice in the age of intelligent agents. Minimal level of AI development experience required e.g. prompting Claude

Multi-agent AI frameworks are everywhere right now — OpenAI Swarm, CrewAI, Langflow — but most of them are built Python-first. Meanwhile, a huge chunk of AI startups (60–70% in YC, by some counts) are building in TypeScript. And honestly, spinning up a whole Python backend just to leverage some of these frameworks didn’t feel worth it.

In this talk, we’ll walk through how we approached building a multi-agent AI system in TypeScript — by focusing on the core principles behind these frameworks and building something similar that fits the JS stack.