Presented at the AI Native Dev meetup in Stockholm, Patrick Debois traces the rapid evolution of AI coding tools — from simple autocomplete copilots through chat interfaces, codebase-aware editors, multi-file agents, and now headless CLI-based agent swarms running asynchronously in the background. He frames this progression against the concept of “AI native” development: not just sprinkling AI on top of existing workflows, but fundamentally rethinking how developers work, much as cloud native meant more than lifting VMs into the cloud.
The talk introduces four patterns that capture how developer roles are shifting. The first pattern, from producer to manager, addresses the reality that as AI generates more code, developers spend less time writing and more time reviewing. Patrick explores how tools are evolving to reduce cognitive load — condensed diff views, step-by-step review flows, visual diagram diffs, and the concept of moldable development environments that adapt the IDE to the review context at hand. He highlights Aider’s approach of auto-committing by default and letting developers revert, flipping the traditional approval model. Practical concerns like setting file-level permissions for agents and monitoring agent costs (citing a CTO spending $80/day on seven concurrent coding agents) underscore the operational reality of managing AI workers.
The second pattern, from implementation to intent, describes how developers increasingly focus on specifying what they want built rather than how to build it. Patrick shows how this evolved from simple spec.md files added to prompts, through GitHub’s task-oriented planning features, to fully specification-centric tools where the code is secondary to the product requirements document. He cautions that specifications carry their own challenges — political dynamics, conflicting requirements between security and business, and the risk of repeating waterfall-era mistakes with upfront design documents.
The third pattern, from delivery to discovery, positions the developer as a product explorer. Patrick reframes vibe coding not as careless development but as exploratory coding — rapid prototyping to understand what is technically possible and what users actually need, then discarding the prototype and rebuilding with proper specifications and accumulated knowledge. He shares his personal workflow of vibe coding for two days to build up understanding, then starting fresh with a well-informed spec. He also floats the provocative idea of letting customers vibe code the interfaces they want on top of your product, treating it as AB testing on steroids.
The fourth pattern, from content creation to knowledge, focuses on capturing and preserving what teams learn. Patrick discusses tools that bridge the traditionally siloed worlds of coding editors and observability platforms, bringing production context like call frequency and incident history into the development environment. He highlights Devon’s approach of proactively suggesting that certain insights be saved as persistent knowledge, creating an inline knowledge management loop beneficial to both humans and AI agents over time. The talk closes by framing these four patterns as an expansion of the developer role — touching operations, QA, product ownership, and data engineering — and notes that the curated AI Native Dev landscape now tracks roughly 300 tools in the space.
Watch on YouTube — available on the jedi4ever channel
This summary was generated using AI based on the auto-generated transcript.