Mike Krieger acknowledges a dual emotion among top engineers: immense excitement about their new leverage with AI, but also a genuine sadness for the disappearing craft of coding. The experience of dreaming about a code problem and waking up with an elegant solution is fading.
Krieger highlights a new paradigm where he delegates multi-hour tasks to the AI before bed. The model autonomously handles obstacles, like a service outage, by building temporary scaffolds and then completing the original task when the service returns.
For tasks too complex for a single prompt, Krieger uses "dynamic workflows." This involves designing a multi-step plan (e.g., understand, spec, translate, test) which the AI executes autonomously. This allowed Fable to port a complex Python project to TypeScript over a weekend.
Despite access to the powerful Fable model, Mike Krieger finds it's "overkill" for simple queries like sports scores. He deliberately uses the faster, less "thoughtful" Sonnet model on his phone, highlighting the need for a "model fleet" approach for different tasks.
A key advancement in Fable is its ability to exercise judgment. When receiving feedback from a human or another AI, it can analyze the suggestion and disagree, explaining why its original approach is better for the given context, thus mimicking a senior collaborator.
Advanced AI models are closing the gap between intent and execution for non-coders. Mike Krieger cites a recruiter at Anthropic who, for the first time, could build a tool from her imagination, then iterate on and deploy it to her entire organization without engineering support.
Krieger demonstrates an "agent-native architecture" where the AI isn't just a feature but can directly modify the application's source code. A long-press on a chat button allows him to request features, which the AI then implements, builds, and deploys.
Instagram Co-founder Mike Krieger says the engineer's role is evolving. The satisfaction of elegantly solving a problem in code is being replaced by the ability to manage AI teammates, define high-level architecture, verify outputs, and handle production incidents—skills that are more about orchestration than implementation.
To trust AI-generated code, Krieger’s team requires pull requests to include visual proof, such as a "full screenshot gallery of the full UI." This allows human reviewers to quickly spot issues in error states or animations that code review alone would miss, tightening the development loop.