I’m getting less and less confident in my ability to predict what AI will and won’t be able to do in the future in the scope of coding and product development.
In mid- to late-2025 my mental model was:
- Role of human: Architect and project manager
- Role of AI: Productive junior developer
For most of the development work I did in 2025, it felt like AI could effectively fill the role of a very productive junior developer, where you could hand off well-scoped and clearly defined tasks to the AI, and a human in the loop was required for any higher-level architectural decisions.
Over the last several months, it feels like the new models have gotten much better at the architecture work. They’re able to plan out the organization of multiple services, work across multiple layers of abstraction, and understand how different pieces need to work together. I don’t have enough confidence yet in the AI tools to make any significant architecture decisions without in-depth review, but it seems like we’re heading in the direction of this problem being solved.
My 2026 mental model of the role of humans vs AI in product development is starting to shift more towards the delineation of defining business logic vs encoding this business logic into software:
- Role of human: Define the core business logic of a system.
- Role of AI: Encode the core business logic into software.
For example, if you’re building TurboTax, you’re going to need to define the core business logic for how a consumer tax return service should work. This may include defining core customer segments, triaging different tax prep use cases, and outlining high-level processes to file taxes for different types of customer segments. All of these things require strong domain knowledge, require context of how the internal business units and services are structured, understanding which problems are easy vs hard, and judgment on product design (less frontend/UI design and more the canonical definition of design of building a product that serves a specific functional need very well).
Once the core business logic for a product is clearly defined, it seems like AI is quickly becoming good at building the product end-to-end, including everything from system design, writing the code, writing tests, CI/CD processes, setting up infrastructure.
In the future, I think even some of the more PM type work is going to be automated with AI. Many tech companies already have a large amount of data capture on internal & external business operations, where meetings are auto-recorded and internal processes are documented in tools like Google Drive, Confluence, Notion.
My soft predictions for what this means for builders in 2026 are:
- EQ is going to matter more for builders: Engineering roles are becoming more generalized, and overlapping more with PM work. Essentially, every IC engineer is moving up one level in the org chart and managing a team of AI devs, requiring more focus on alignment and outcomes.
- Being proactive about learning is very important: There’s a huge gap between experimenting with AI coding tools over a few weekends and immersing yourself in this new way of building. I think this gap is going to continue expanding very quickly.
- Documentation will be king: Written documentation is the required precursor to support automation/acceleration with AI. For companies that have strong cultural standards around documentation (especially in business units outside of engineering like product, customer success, sales), this will enable high-leverage use cases of AI.