banatie-content/assets/ai-coding-methodologies-beyond-vibe-coding/perplexity-terminology-research.md

AI-Assisted Development: Кластеризованная терминология и подходы

Source: Perplexity Research, January 2026 Original file: /research/perplexity-chats/AI-Assisted Development_ Кластеризованная терминол.md

---

Domain 1: Experimental & Low-Quality Approaches

Vibe Coding

Authority Rank: 1 | Perception: Negative

Sources:

1. Andrej Karpathy (OpenAI co-founder, Tesla AI Director) — Wikipedia, Feb 2025
2. Collins English Dictionary — Word of the Year 2025
3. SonarSource (Code Quality Platform) — code quality analysis

Description: Term coined by Andrej Karpathy in February 2025, quickly became cultural phenomenon — Collins English Dictionary named it Word of Year 2025. Approach where developer describes task in natural language, AI generates code, but key distinction: developer does NOT review code, only looks at execution results.

Simon Willison quote: "If LLM wrote every line of your code, but you reviewed, tested and understood it all — that's not vibe coding, that's using LLM as typing assistant". Key characteristic: accepting AI-generated code without understanding it.

Critics point to lack of accountability, maintainability problems, increased security vulnerability risk. May 2025: Swedish app Lovable (using vibe coding) had security vulnerabilities in 170 of 1,645 created web apps. Fast Company September 2025 reported "vibe coding hangover" — senior engineers cite "development hell" working with such code.

Suitable for "throwaway weekend projects" as Karpathy originally intended, but risky for production systems.

Links:

• https://en.wikipedia.org/wiki/Vibe_coding
• https://www.sonarsource.com/resources/library/vibe-coding/

---

Domain 2: Enterprise & Production-Grade Methodologies

AI-Driven Development Life Cycle (AI-DLC)

Authority Rank: 1 | Perception: Positive - Enterprise

Sources:

1. AWS — Raja SP, Principal Solutions Architect, July 2025
2. Amazon Q Developer & Kiro — official AWS platform

Description: Presented by AWS July 2025 as transformative enterprise methodology. Raja SP created AI-DLC with team after working with 100+ large customers.

AI as central collaborator throughout SDLC with two dimensions:

1. AI-Powered Execution with Human Oversight — AI creates detailed work plans, actively requests clarifications, defers critical decisions to humans
2. Dynamic Team Collaboration — while AI handles routine tasks, teams unite for real-time problem solving

Three phases: Inception (Mob Elaboration), Construction (Mob Construction), Operations

Terminology: "sprints" → "bolts" (hours/days instead of weeks); Epics → Units of Work

Link: https://aws.amazon.com/blogs/devops/ai-driven-development-life-cycle/

Spec-Driven Development (SDD)

Authority Rank: 2 | Perception: Positive - Systematic

Sources:

1. GitHub Engineering — Den Delimarsky, Spec Kit toolkit, September 2025
2. ThoughtWorks Technology Radar — November 2025
3. Red Hat Developers — October 2025

Description: Emerged 2025 as direct response to "vibe coding" problems. ThoughtWorks included in Technology Radar. GitHub open-sourced Spec Kit September 2025, supports Claude Code, GitHub Copilot, Gemini CLI.

Key principle: specification becomes source of truth, not code.

Spec Kit Workflow:

1. Constitution — immutable high-level principles (rules file)
2. /specify — create specification from high-level prompt
3. /plan — technical planning based on specification
4. /tasks — break down into manageable phased parts

Three interpretations: Spec-first, Spec-anchored, Spec-as-source

Tools: Amazon Kiro, GitHub Spec Kit, Tessl Framework

Links:

• https://github.blog/ai-and-ml/generative-ai/spec-driven-development-with-ai-get-started-with-a-new-open-source-toolkit/
• https://www.thoughtworks.com/radar/techniques/spec-driven-development
• https://martinfowler.com/articles/exploring-gen-ai/sdd-3-tools.html
• https://developer.microsoft.com/blog/spec-driven-development-spec-kit

Architecture-First AI Development

Authority Rank: 3 | Perception: Positive - Professional/Mature

Sources:

1. WaveMaker — Vikram Srivats (CCO), Prashant Reddy (Head of AI Product Engineering), January 2026
2. ITBrief Industry Analysis — January 2026

Description: 2026 industry shift. Quote Vikram Srivats: "Second coming of AI coding tools must be all about Architectural Intelligence — just Artificial Intelligence no longer fits."

Shift from "vibe coding" experiments to governance, architecture alignment, long-term maintainability.

Key characteristics:

• System design before implementation
• AI agents with clear roles: Architect, Builder, Guardian
• Coding architectural rules, enforcement review processes
• Working from formal specifications
• Respect for internal organizational standards

Links:

• https://itbrief.co.uk/story/ai-coding-tools-face-2026-reset-towards-architecture
• https://itbrief.news/story/ai-coding-tools-face-2026-reset-towards-architecture

---

Domain 3: Quality & Validation-Focused Approaches

Test-Driven Development with AI (TDD-AI)

Authority Rank: 1 | Perception: Positive - Quality-Focused

Sources:

1. Galileo AI Research — August 2025
2. Builder.io Engineering — August 2025

Description: Traditional TDD adapted for AI systems. Tests written first → AI generates code to pass tests → verify → refactor.

Statistical testing for non-deterministic AI outputs — critical distinction from traditional TDD.

Links:

• https://galileo.ai/blog/tdd-ai-system-architecture
• https://galileo.ai/blog/test-driven-development-ai-systems
• https://www.builder.io/blog/test-driven-development-ai

Human-in-the-Loop (HITL) AI Development

Authority Rank: 2 | Perception: Positive - Responsible

Sources:

1. Google Cloud Documentation — 2026
2. Encord Research — December 2024
3. Atlassian Engineering — HULA framework, September 2025

Description: Humans actively involved in AI system lifecycle. Continuous feedback and validation loops. Hybrid approach: human judgment + AI execution.

HULA (Human-in-the-Loop AI) — Atlassian framework for software development agents.

Links:

• https://cloud.google.com/discover/human-in-the-loop
• https://encord.com/blog/human-in-the-loop-ai/
• https://www.atlassian.com/blog/atlassian-engineering/hula-blog-autodev-paper-human-in-the-loop-software-development-agents

Quality-First AI Coding

Authority Rank: 3 | Perception: Positive - Professional

Sources:

1. Qodo.ai (formerly CodiumAI) — December 2025

Description: Code integrity at core. Qodo.ai — platform with agentic AI code generation and comprehensive testing.

Production-ready focus: automatic test generation for every code change. Direct contrast to "vibe coding" — quality non-negotiable.

Link: https://www.qodo.ai/ai-code-review-platform/

Deterministic AI Development

Authority Rank: 4 | Perception: Positive - Enterprise/Compliance

Source: Augment Code Research — August 2025

Description: Identical outputs for identical inputs. Rule-based architectures for predictability. Best for: security scanning, compliance checks, refactoring tasks.

Hybrid approach: probabilistic reasoning + deterministic execution.

Link: https://www.augmentcode.com/guides/deterministic-ai-for-predictable-coding

---

Domain 4: Collaborative Development Patterns

AI Pair Programming

Authority Rank: 1 | Perception: Positive - Collaborative

Sources:

1. GitHub Copilot (Microsoft) — January 2026
2. Qodo.ai Documentation — March 2025
3. GeeksforGeeks — July 2025

Description: AI as "pair programmer" or coding partner. Based on traditional pair programming: driver (human/AI) and navigator (human/AI) roles.

Real-time collaboration and feedback. Tools: GitHub Copilot, Cursor, Windsurf.

Links:

• https://code.visualstudio.com/docs/copilot/overview
• https://www.qodo.ai/glossary/pair-programming/
• https://www.geeksforgeeks.org/artificial-intelligence/what-is-ai-pair-programming/
• https://graphite.com/guides/ai-pair-programming-best-practices

Mobbing with AI / Mob Programming with AI

Authority Rank: 2 | Perception: Positive - Team-Focused

Sources:

1. Atlassian Engineering Blog — December 2025
2. Aaron Griffith — January 2025 (YouTube)

Description: Entire team works together, AI as driver. AI generates code/tests in front of team. Team navigates, reviews, refines in real-time.

Best for: complex problems, knowledge transfer, quality assurance.

Links:

• https://www.atlassian.com/blog/atlassian-engineering/mobbing-with-ai
• https://www.youtube.com/watch?v=BsFPbYX4WXQ

Agentic Coding / Agentic Programming

Authority Rank: 3 | Perception: Positive - Advanced

Sources:

1. arXiv Research Paper — August 2025
2. AI Accelerator Institute — February 2025
3. Apiiro Security Platform — September 2025

Description: LLM-based agents autonomously plan, execute, improve development tasks. Beyond code completion: generates programs, diagnoses bugs, writes tests, refactors.

Key properties: autonomy, interactive, iterative refinement, goal-oriented.

Agent behaviors: planning, memory management, tool integration, execution monitoring.

Links:

• https://arxiv.org/html/2508.11126v1
• https://www.aiacceleratorinstitute.com/agentic-code-generation-the-future-of-software-development/
• https://apiiro.com/glossary/agentic-coding/

---

Domain 5: Workflow & Process Integration

Prompt-Driven Development (PDD)

Authority Rank: 1 | Perception: Neutral to Positive

Sources:

1. Capgemini Software Engineering — May 2025
2. Hexaware Technologies — August 2025

Description: Developer breaks requirements into series of prompts. LLM generates code for each prompt. Critical: developer MUST review LLM-generated code.

Critical distinction from vibe coding: code review mandatory.

Links:

• https://capgemini.github.io/ai/prompt-driven-development/
• https://hexaware.com/blogs/prompt-driven-development-coding-in-conversation/

AI-Augmented Development

Authority Rank: 2 | Perception: Positive - Practical

Sources:

1. GitLab Official Documentation — December 2023
2. Virtusa — January 2024

Description: AI tools accelerate SDLC across all phases. Focus: code generation, bug detection, automated testing, smart documentation.

Key principle: humans handle strategy, AI handles execution.

Links:

• https://about.gitlab.com/topics/agentic-ai/ai-augmented-software-development/
• https://www.virtusa.com/digital-themes/ai-augmented-development

Copilot-Driven Development

Authority Rank: 3 | Perception: Positive - Practical

Sources:

1. GitHub/Microsoft Official — January 2026
2. Emergn — September 2025

Description: Specifically using GitHub Copilot or similar tools as development partner (not just assistant).

Context-aware, learns coding style. Enables conceptual focus instead of mechanical typing.

Links:

• https://code.visualstudio.com/docs/copilot/overview
• https://www.emergn.com/insights/how-ai-tools-impact-the-way-we-develop-software-our-github-copilot-journey/

Conversational Coding

Authority Rank: 4 | Perception: Neutral to Positive

Sources:

1. Google Cloud Platform — January 2026
2. arXiv Research — March 2025

Description: Natural language interaction with AI for development. Iterative, dialogue-based approach. Context retention across sessions.

Links:

• https://cloud.google.com/conversational-ai
• https://arxiv.org/abs/2503.16508

---

Domain 6: Code Review & Maintenance

AI Code Review

Authority Rank: 1 | Perception: Neutral to Positive

Sources:

1. LinearB — March 2024
2. Swimm.io — November 2025
3. CodeAnt.ai — May 2025

Description: Automated code examination using ML/LLM. Static and dynamic analysis. Identifies bugs, security issues, performance problems, code smells.

Tools: Qodo, CodeRabbit, SonarQube AI features.

Links:

• https://linearb.io/blog/ai-code-review
• https://swimm.io/learn/ai-tools-for-developers/ai-code-review-how-it-works-and-3-tools-you-should-know
• https://www.codeant.ai/blogs/ai-vs-traditional-code-review

---

Domain 7: Specialized & Emerging Approaches

Ensemble Programming/Prompting with AI

Authority Rank: 1 | Perception: Positive - Advanced

Sources:

1. Kinde.com — November 2024
2. Ultralytics ML Research — December 2025
3. arXiv — June 2025

Description: Multiple AI models/prompts combined for better results. Aggregation methods: voting, averaging, weighted scoring.

Links:

• https://kinde.com/learn/ai-for-software-engineering/prompting/ensemble-prompting-that-actually-moves-the-needle/
• https://www.ultralytics.com/blog/exploring-ensemble-learning-and-its-role-in-ai-and-ml

Prompt Engineering for Development

Authority Rank: 2 | Perception: Neutral to Positive

Sources:

1. Google Cloud — January 2026
2. OpenAI — April 2025
3. GitHub — May 2024

Description: Crafting effective prompts for AI models. Critical skill for AI-assisted development.

Techniques: few-shot learning, chain-of-thought, role prompting.

Links:

• https://cloud.google.com/discover/what-is-prompt-engineering
• https://platform.openai.com/docs/guides/prompt-engineering
• https://github.blog/ai-and-ml/generative-ai/prompt-engineering-guide-generative-ai-llms/

Intentional AI Development

Authority Rank: 3 | Perception: Positive - Thoughtful

Sources:

1. Tech.eu — January 2026
2. ghuntley.com — August 2025

Description: Purpose-driven AI design. Clear roles and boundaries for AI. Deliberate practice and learning approach.

Links:

• https://tech.eu/2026/01/05/adopting-an-intentional-ai-strategy-in-2026/
• https://ghuntley.com/play/

---

Domain 8: General & Cross-Cutting Terms

AI-Assisted Coding / AI-Assisted Development

Authority Rank: 1 | Perception: Neutral to Positive

Sources:

1. Wikipedia — July 2025
2. GitLab — 2025

Description: Broad umbrella term for AI enhancing software development tasks. Includes code completion, documentation generation, testing, debugging assistance.

Developer remains in control, reviews all suggestions. Most common adoption pattern globally.

Links:

• https://en.wikipedia.org/wiki/AI-assisted_software_development
• https://about.gitlab.com/topics/devops/ai-code-generation-guide/

---

Key Takeaways

Domain 1 (Experimental): Only Vibe Coding — only term with explicitly negative connotation, backed by high-authority sources (OpenAI founder, Collins Dictionary).

Domain 2 (Enterprise): Most authoritative domain with AWS, GitHub Engineering, ThoughtWorks as sources. Focus on production-grade, governance, architecture.

Domain 3 (Quality): Research-heavy domain (Galileo AI, Google Cloud, Atlassian) with emphasis on responsible development.

Domain 4 (Collaborative): Practical patterns, backed by major platforms (Microsoft/GitHub, Atlassian) and research (arXiv).

Domains 5-7: Workflow integration, code review, specialized techniques — more narrow but important practices.

Domain 8: General term serving as baseline for all other approaches.