What I'm building as an AI Product Engineer in 2026

A few months of intensive AI-agent development. The shift is real — implementation got dramatically easier.

I kept asking "should I write this myself?" Turns out, no. Give Claude Code the context, point it in a direction, and the code appears. Component splitting, API integration, state management — the typing is no longer your job.

This isn't about coding faster. The role itself is changing.

Small scale: feature designer. Large scale: app and systems architect. From someone who writes code to someone who designs the structure code gets built from.

To evolve, build a lot first

Design skills don't come from books. You only see certain things after actually building.

Build features, apps, whole systems. Auth, payments, notifications, analytics — experience all of it inside a single project. That's when better structure starts revealing itself on the next one. I learned this building an ERP system from scratch.

But change how you build.

Don't type it yourself. Have AI do it. My job is deciding "what to build" and "what structure to use" — implementation is the agent's job. When AI goes off track, the root cause is almost always my design. Better design, better AI output. Proven repeatedly.

Repeat enough and you'll see the patterns

Delegating to AI repeatedly — patterns start showing up.

Same context explained every time. Same rules repeated. Same task order requested. That's the signal. That's where abstraction starts.

Claude Code Skills, Custom Agents — they're all just workflow abstractions.

Best way to discover your workflow: do a lot of work in a short focused sprint. Run varied projects quickly and the repeating patterns surface on their own.

First thing I'm abstracting — RN AI Agent

Multiple projects in, and I started seeing what to abstract. First thing I'm building: the RN AI Agent.

Every React Native app has the same repetitive initial setup. Navigation structure, state management, API layer, auth flow. Before, I solved this with boilerplate.

That's not enough anymore. Boilerplate is a code template. The RN AI Agent is a colleague with actual implementation skills. Iron Man suit analogy: put it on and you can fly. Equip the RN AI Agent and you ship more apps, faster.

But flying faster isn't the point.

Cut implementation time dramatically with the suit, and the AI Product Engineer finally has time for what actually matters: defining the problem. Observing, empathizing, discovering what users are actually struggling with. Then abstracting that discovery process itself into a workflow.

Implementation goes to the agent. More time for problem discovery and definition. And in that process, I'm starting to see repeating patterns there too — things that might eventually be abstracted as well.

Why apps specifically

Web is getting saturated with vibe-coded projects. Landing pages, dashboards — fast to spin up with the right tools.

Apps have a higher barrier. Still.

Native build environments, App Store review, device compatibility, background processing, push notifications — hurdles that pure vibe-coding can't clear. That's the gap. Still open for developers who know what they're doing.

More apps, more problems solved. That's the only reason the suit matters.

The real goal: find your workflow

Building a lot isn't the goal.

Discovering your own workflow through the process of building — that's the goal. And that workflow needs to be found above the implementation layer.

Observe → empathize → discover → hypothesize → experiment → validate. AI can automate from "experiment" onward. "Why this problem matters" is still the human job.

Parallel to that: keep studying agent usage for implementation. Skills architecture, Custom Agent design, MCP server integration — deeper tool understanding means higher abstraction levels.

Laid out like this, everything I'm trying in H1 2026 connects into one thing.

Build more. Change how I build. Discover my workflow in the process. Implementation to the Iron Man suit (RN AI Agent), direction from the human. Abstracting the direction-setting process itself — that's the AI Product Engineer's job.

I'll keep sharing what I find. If you want to follow this journey and pick up insights along the way, sign up.