Philosophy · an essay
How I evaluate an AI-First developer.
Eight required skills for developers who use AI coding agents across planning, implementation, testing, debugging, and documentation.
In this essay, AI-First has a narrow meaning. The developer uses AI coding agents throughout the software development lifecycle and remains accountable for every result.
The agent may draft plans, code, tests, fixes, and documentation. The developer provides context, chooses tools, reviews the output, runs the checks, and decides what can ship.
I call my own practice an AI-exclusive workflow. I use models such as Claude, Gemini, and GPT for planning, architecture, coding, testing, debugging, and documentation. That choice raises the standard for context, review, and verification because generated work is never accepted on authorship alone.
Treat AI-generated code as untrusted until you can explain it, test it, and observe the result. The developer owns the output.
01 · Selection
The AI-First selection funnel.
Start with five filters. Look for sustained agent use, clear context, critical review, broad technical fundamentals, and precise communication. Industry experience comes later.
Figure 01, selection funnel
02 · Qualifications
Eight required qualifications.
The attributes below are what I look for. Must-haves are non-negotiable given the AI-exclusive approach. Nice-to-haves strengthen a candidate but aren't prerequisites.
Figure 02, qualifications table
Must-have
Non-negotiable skills.
If a candidate lacks any of these, they aren't a fit for an AI-exclusive role.
AI Development Commitment
- Uses models such as Claude, Gemini, or GPT as the primary tool for planning, architecture, coding, testing, debugging, and documentation.
- Can explain where AI helps in each phase, where it fails, and which checks catch those failures.
AI Collaboration & Context Engineering
- Context engineering structures system prompts, tool descriptions, and file context so the model has the evidence needed for the task. The developer knows when to add context and when to remove it.
- Agentic workflow design decomposes work into plan, execute, verify, and iterate loops. The developer can choose between one model call, one agent, or several specialized agents.
- Iterative prompting uses screenshots, stack traces, test output, and reviewer feedback instead of repeating the same request.
- Able to decompose large problems into smaller tasks suitable for AI-driven generation.
Tool & Context Integration
- AI coding toolchain means practical use of at least one coding agent such as Claude Code, Cursor, Windsurf, Cline, Aider, or Copilot. The developer can compare terminal and editor workflows, autonomy, and context handling.
- Model Context Protocol (MCP) connects agents to external tools and data. The developer can choose among an MCP server, an inline tool definition, and a direct API call.
- Rechecks tool choices when models, constraints, or project needs change.
Critical Thinking & Output Validation
- Critically evaluates AI-generated code, designs, tests, and docs for correctness, efficiency, security, maintainability.
- Uses reproducible failures, logs, and tests to guide the agent through a fix.
- Understands how to test and verify solutions where the AI was the primary author. Evaluator is half the job.
Foundational Tech Knowledge
- Core programming: data structures, algorithms, control flow, patterns common across languages (OOP, functional).
- Web fundamentals: HTML, CSS, JavaScript/ES6+, DOM, client-server, REST, request/response.
- Backend: server logic, API design, database interaction, Python/Java/Node/Go basics.
- Architecture: design patterns, modular design, microservices, event-driven, API best practices.
- Databases: relational and NoSQL data, schema design, querying, and transactions.
Communication & Problem Articulation
- Writes technical requirements, user stories, constraints, and acceptance criteria in specification-grade language.
- Effectively communicates with non-technical stakeholders about progress and AI-driven methodology.
Adaptability & Learning Agility
- Verifies unfamiliar tools and techniques instead of accepting a model recommendation from memory.
SDLC & Agile Principles
- Knows all SDLC phases even when AI executes many tasks. Manages the process end-to-end through AI guidance, adaptable across methodologies.
Nice-to-have
Desirable but not required.
If the must-haves are met, these strengthen the case but don't gatekeep.
Hands-on experience in key technologies
- Prior professional experience in common stacks: .NET/C#, Java/Spring, Python/Django/Flask, Node/Express, Ruby on Rails, Go, Rust.
- Specific frontend frameworks: Vue 3, React, Angular, Svelte.
- Mobile platforms: Swift/iOS, Kotlin/Android, React Native, Flutter.
- Cloud (AWS/Azure/GCP) and containerization (Docker/Kubernetes).
Domain-specific knowledge
- Experience in the specific industry or domain can reduce the learning curve but is not a substitute for the required skills.
- Familiarity with relevant protocols or specialized systems for the project domain.
Advanced software engineering practices
- Deep understanding of advanced design patterns and architecture.
- Experience with TDD, BDD, and applying them effectively in an AI-assisted workflow.
- Spec-driven development uses precise intent specifications that an AI implements and tests end to end.
- AI as PR reviewer uses automated review as one CI signal without treating it as approval.
- Applies threat modeling and secure defaults to AI-generated code.
Multi-agent orchestration
- Designs workflows where specialized agents handle planning, implementation, review, or testing.
- Knows when to orchestrate multiple agents vs. when a single strong agent is sufficient. Over-engineering agentic systems is a common anti-pattern.
CI/CD pipelines and DevOps
- Understands CI/CD and DevOps, and how AI integrates into these pipelines across app types.
Version control fluency
- Uses branching strategies and resolves complex merge conflicts even when AI handles routine version control tasks.
Project / technical leadership aptitude
- Experience leading small projects or technical initiatives in diverse environments.
- Translates business needs from any sector into technical specs suitable for AI development.
Closing notes
Four hiring rules.
The AI-centric must-haves are the primary filter. Resistance or inability to fully embrace an AI-exclusive workflow is a non-starter.
The role needs both an orchestrator and evaluator. Broad technical knowledge helps the developer direct the agent and review its output across a stack.
Human oversight remains essential. Frontier models are capable but not context-complete. The developer's critical thinking is paramount across every endeavor.
The stack a candidate names matters less than the method. Ask for a real task, the context provided, the output produced, the checks run, and the corrections made.