424 lines
15 KiB
Markdown
424 lines
15 KiB
Markdown
# 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.
|