Skip to content

OpenLAIR/awesome-vibe-researching

BranchesTags

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

5 Commits
.github
.github
 
 
assets
assets
 
 
CODE_OF_CONDUCT.md
CODE_OF_CONDUCT.md
 
 
CONTRIBUTING.md
CONTRIBUTING.md
 
 
LICENSE
LICENSE
 
 
README.md
README.md
 
 

Repository files navigation

Awesome Vibe Researching Awesome

GitHub stars GitHub forks

License: CC0-1.0 PRs Welcome Resources Papers Tools Last Updated

From Assisted to Autonomous: L1 Assist → L2 Partial → L3 Conditional → L4 High → L5 Full

A curated list of resources for Vibe Researching -- AI-driven research automation across the scientific discovery pipeline, from literature review to publication.

Vibe Researching extends the concept of "vibe coding" (conversational, AI-assisted programming) to the entire research workflow. This collection covers tools, papers, and frameworks spanning six research pipeline stages (Survey -> Ideation -> Experiment -> Analysis -> Writing -> Promotion) and six autonomy levels (L0 Manual -> L5 Fully Autonomous).

Legend: NEW = published/released 2025-2026 | OSS = open-source | FREE = free to use

Contents

What is Vibe Researching?

Vibe Researching is the application of AI agents with skills to the full research pipeline, enabling researchers to:

  • Automate literature reviews -- 80-90% time savings with tools like Elicit, Rayyan
  • Generate code from academic papers -- DeepCode: 75.9% success, beats human experts at 72.4%
  • Run autonomous experiments -- A-Lab: 10x data collection speedup, 71% success rate
  • Produce complete research papers -- AI Scientist-v2: first peer-reviewed AI-generated paper, 2025

Key Paper: Vibe Researching as Wolf Coming (arXiv:2602.22401, Feb 2026) -- Introduces the vibe researching paradigm with cognitive task framework

Community: VibeX 2026 Workshop -- 1st International Workshop on Vibe Coding and Vibe Researching @ EASE 2026

Leveraged Cognition: Traditional Manual Research vs Vibe Researching vs Fully Automated AI Scientist
Manual work is too slow. Fully automated AI is too generic. Vibe Researching is the new frontier.
Dr. Claw turns your Research Taste into outsized outcomes with Agentic Execution -- so you can move faster, think bigger, and still hold the line on scientific rigor.

Navigation Guide

By Research Stage

This list is organized by the research pipeline (Survey -> Ideation -> Experiment -> Analysis -> Writing -> Promotion). Each stage has tools and papers at different autonomy levels:

Level Name Human Role AI Role Example
L1 Assist Drives all decisions Suggests Code completion, grammar checking
L2 Partial Defines task, validates Executes specific subtasks Automated screening, figure generation
L3 Conditional Sets goals, approves at checkpoints Handles full workflows End-to-end literature review
L4 High Monitors for failures Handles extended workflows Self-driving labs
L5 Full None (reviewer/consumer) Operates independently AI Scientist: idea -> paper

Level x Stage Matrix

                 | Survey | Ideation | Experiment | Analysis | Writing | Promotion |
-----------------+--------+----------+------------+----------+---------+-----------+
Level 1 (Assist) |   8    |    2     |    12      |    3     |    6    |     1     |
Level 2 (Partial)|   6    |    3     |    10      |    7     |    2    |     0     |
Level 3 (Cond.)  |   2    |    2     |     8      |    4     |    1    |     0     |
Level 4 (High)   |   1    |    1     |    12      |    3     |    0    |     0     |
Level 5 (Full)   |   0    |    1     |     6      |    1     |    1    |     0     |
-----------------+--------+----------+------------+----------+---------+-----------+
Total            |  17    |    9     |    48      |   18     |   10    |     1     |

Key Statistics

Metric Value Source/Year
Total resources 107 (45 papers, 62 tools) This survey, Mar 2026
Developer AI adoption 92% of US developers use AI coding tools daily 2025 industry data
Code generation share 41% of all code globally AI-generated (256B lines) Google, 2024
Lab automation market $7.84B -> $14.78B (CAGR 6.55%) 2024-2034 forecast
Multimodal AI market $391B -> ~$2T (CAGR 35.9%) 2025-2030 forecast
Key milestone AI Scientist-v2: first peer-reviewed AI paper ICLR 2025

Survey & Literature Review

Tools and papers for literature search, systematic review, knowledge extraction, and gap analysis. (17 items)

Level 1 -- Assisted Search

Papers:

  • Vibe Coding in Practice: Motivations, Challenges, and a Future Outlook NEW -- Grey literature review (arXiv:2510.00328, Sep 2025). Analyzed 101 practitioner sources with 518 firsthand behavioral accounts. Revealed speed-quality trade-off paradox: 62% motivated by speed, but code often 'fast but flawed'. Tags: empirical study, grey literature, developer experience.

Tools:

  • Google Scholar FREE -- Classic academic search engine with citation tracking, h-index metrics, and researcher profiles. Free access to academic literature across disciplines. De facto standard for citation counts. L1 Assist.
  • Semantic Scholar FREE -- AI-powered academic search from Allen Institute for AI. Features citation contexts, paper recommendations, TLDR summaries, and influential citations analysis. Covers 200M+ papers. L1 Assist.

Level 2 -- Automated Screening & Extraction

Tools:

  • Elicit -- AI-powered systematic review platform. 125M+ papers, 545K clinical trials. 80% time savings for systematic reviews, high precision for preliminary searches. Automated data extraction and evidence synthesis. L2 Partial. Commercial.
  • Rayyan -- AI study screening for systematic reviews. 90% reduction in screening time, ML-based study suggestion, automatic deduplication. Widely used in biomedical systematic reviews. L2 Partial. Freemium.
  • Consensus -- AI academic search engine. 200M+ peer-reviewed papers, evidence-based answers with citation support, automated data synthesis. Focuses on extracting research findings as direct answers. L2 Partial. Freemium.
  • Scite -- Smart citation analysis evaluating how publications are cited: supporting, contrasting, or mentioning. Reveals citation context beyond simple counts. L2 Partial. Commercial.

Level 3 -- End-to-End Literature Review

Tools:

  • Paperguide NEW -- Fully automated systematic review: research question -> search -> screening -> report generation. Improved by Deep Research integration (June 2025). Competitive features at lower price than SciSpace. L3 Conditional. Commercial.
  • SciSpace -- 10M+ researchers (2026). Upload PDF, chat with paper, semantic search, automated summarization, reference management, AI copilot. Comprehensive research assistant platform. L3 Conditional. Freemium.

Stage Summary

Finding Detail
Time savings 80-90% demonstrated by Elicit and Rayyan in systematic review workflows
Adoption pattern Commercial-first: tools emerged 2021-2024, academic evaluation papers followed 2024-2026
Gap No Level 4-5 tools for autonomous literature synthesis exist yet

Ideation & Hypothesis Generation

Tools and papers for research question formulation, hypothesis generation, and novelty checking. (9 items -- under-represented stage)

Level 1-2 -- Brainstorming & Templates

Tools:

  • Brainstorming Assistants -- GPT-powered conversational agents for research question ideation and problem formulation. General-purpose LLMs (ChatGPT, Claude) serve as interactive brainstorming partners. L1 Assist.
  • Hypothesis Generators -- Template-based tools for structured research question development using PICO/FINER frameworks. L1 Assist.
  • Gap Analysis Tools -- AI-powered literature analysis for identifying research gaps and understudied areas by analyzing citation patterns and topic coverage. L2 Partial.

Level 5 -- Autonomous Ideation

Papers:

Tools:

  • Sakana AI - AI Scientist OSS -- (GitHub | 5k+ stars) End-to-end autonomous research: idea generation -> code implementation -> experiment execution -> LaTeX paper. Open-source reference implementation. L5 Full.

Critical Evaluation

Papers:

  • Evaluating Sakana's AI Scientist for Autonomous Research NEW (arXiv:2502.14297, 2025, ACM SIGIR Forum) -- Critical evaluation revealing 42% experiment failure rate due to coding errors, poor novelty assessments (misclassifying established concepts as novel). Important counterpoint to optimistic claims. Evaluation paper.

Stage Summary

Finding Detail
Critical gap Ideation stage under-represented (9 items total, mostly Level 1-2)
Need Standalone autonomous hypothesis generation tools (Level 4-5), novelty checking systems
AI Scientist limitation 42% failure rate requires human review of all outputs

Experiment & Implementation

Tools and papers for protocol design, code implementation, laboratory execution, and data collection. Largest category (48 items).

Vibe Coding (Level 1 -- Code Completion)

Papers:

Tools:

Tool Type Key Metric Pricing
GitHub Copilot Commercial 20M users, 90% of Fortune 100 $10-39/mo
Cursor Commercial 4.9/5 ratings, $60M Series A Free-$20/mo
Claude Code Commercial 46% "most loved" (2026) Subscription
Tabnine Commercial Air-gapped support, privacy-first Freemium
Continue OSS Open Source GitHub 15k+ stars Free
Aider OSS Open Source GitHub 20k+ stars, ~75% success Free

Tool Details:

  • GitHub Copilot -- Pioneering inline autocomplete (launched 2021). 20M users (mid-2025), powers 90% of Fortune 100. Supports Claude 3 Sonnet, Gemini 2.5 Pro. Agent Mode introduced 2025. L1 Assist. Commercial: $10/mo individual, $19/mo Business, $39/mo Enterprise.
  • Cursor -- Developed by Anysphere (2023), $60M Series A (Aug 2024). Averages 4.9/5 user ratings across 2025-2026 roundups. Top-ranked vibe-coding tool in independent benchmarks. L1 Assist. Commercial: Free tier, $20/mo Pro.
  • Claude Code NEW -- 46% "most loved" rating (early 2026) vs Cursor 19%, GitHub Copilot 9%. Terminal-native agentic coding from Anthropic. L1 Assist. Included in Claude subscription.
  • Tabnine -- Privacy-focused pioneer. Local/private cloud/VPC deployment, air-gapped support. Enterprise Context Engine learns org architecture. Best for regulated industries. L1 Assist. Freemium + Enterprise.
  • Continue OSS -- (GitHub | 15k+ stars) Open-source, developer-first coding assistant. Flexible model selection, extensible architecture. L1 Assist. Free.
  • Aider OSS -- (GitHub | 20k+ stars) Repository-level agent handling 50k+ LOC codebases with ~75% success rate. Multi-file refactors, debugging loops, scoped task execution. L2 Partial. Free.

Code Generation from Papers (Level 2)

Papers:

  • Paper2Code: Automating Code Generation from Scientific Papers in Machine Learning (arXiv:2504.17192, 2024) -- Multi-agent LLM framework: planning -> analysis -> generation. 77% user preference, 83% practical utility on PaperBench, performance on par with author-released repos. Benchmark paper.
  • DeepCode: Open Agentic Coding NEW (arXiv:2512.07921, 2025) -- 73.5% success vs PaperCoder 51.1% (+22.4%). Beats human experts (75.9% vs 72.4%) on ICML 2024 paper reproduction. Agentic framework for Paper2Code, Text2Web, Text2Backend. SOTA on PaperBench.

Tools:

  • PaperCoder / Paper2Code OSS -- (2k+ stars) Multi-agent framework: ML paper -> operational code. 77% user preference, performance on par with author-released repos (no statistical difference vs Oracle). L2 Partial.
  • DeepCode OSS NEW -- (1k+ stars) Open agentic coding for Paper2Code, Text2Web, Text2Backend. 73.5% success on PaperBench, beats human experts (75.9% vs 72.4%). L2 Partial.
  • Devin -- Built by competitive programming champions (Cognition), backed by Founders Fund. $4B valuation. Autonomous agent with integrated dev environment, terminal, tests, browser. L3 Conditional. Commercial.

Agentic Frameworks (Level 3)

Papers:

  • Automatic Prompt Engineer (APE) (Zhou et al., 2022) -- (GitHub) Instruction generation as natural language synthesis, black-box optimization. Discovered better zero-shot CoT prompt than 'Let's think step by step' (+3% improvement). Foundational paper for prompt optimization.
  • Is It Time To Treat Prompts As Code? NEW (arXiv:2507.03620, 2025) -- Multi-use case study for DSPy prompt optimization. Prompt evaluation criterion task: 46.2% -> 64.0% accuracy improvement (+17.8%). Empirical study.

Tools:

Tool Stars Key Feature Type
LangChain 90k+ Most complete production stack OSS
LangGraph 30k+ Graph-based stateful agents OSS
AutoGen 30k+ Conversational multi-agent OSS
CrewAI 25k+ Role-based, 40% Fortune 500 OSS
DSPy 15k+ Programmatic prompt optimization OSS

Tool Details:

  • LangChain OSS -- (GitHub | 90k+ stars) Most complete stack for production agent workflows. Evolved from chaining library to full orchestration platform. Best-in-class observability with LangSmith. L3 Conditional.
  • LangGraph OSS -- (GitHub | 30k+ stars) Modern production interface for LangChain. Graph abstraction for stateful multi-agent apps with branching, state machines, error handling, checkpointing. L3 Conditional.
  • LangSmith -- Best-in-class observability for agents. Traces every LLM call, tool invocation, chain step (latency, tokens, errors). Standard production runtime for debugging and monitoring. L3 Conditional. Commercial.
  • CrewAI OSS -- (GitHub | 25k+ stars) Role-based multi-agent workflows. 40% of Fortune 500 use CrewAI agents. Fastest prototyping (minutes). Structured memory with RAG. L3 Conditional.
  • AutoGen OSS -- (GitHub | 30k+ stars) Microsoft's conversational multi-agent model. Agents communicate via natural language. Async task execution, human-in-the-loop oversight. L3 Conditional.
  • DSPy OSS -- (GitHub | 15k+ stars) Stanford framework (2023). Abstracts prompts into modular Python (Signatures, Modules). Programmatically creates and refines prompts. Up to 17.8% accuracy improvement. L3 Conditional.

Self-Driving Laboratories (Level 4-5)

Papers:

  • Self-Driving Laboratories for Chemistry and Materials Science (Chemical Reviews, 2024) -- Comprehensive review: SDLs automate experimental workflows with autonomous planning, accelerating chemistry and materials discovery. Demonstrates 10x data collection speedup. Review paper. Venue: Chemical Reviews (IF 54.3).
  • Autonomous 'self-driving' laboratories: a review of technology NEW (Royal Society Open Science, 2025) -- Most capable SDLs automate entire scientific method: hypothesis generation, experimental design, execution, data analysis, conclusion drawing, hypothesis updating. Review paper.
  • A Survey of AI Scientists NEW (arXiv:2510.23045, 2025) -- Comprehensive survey of autonomous research agents and hierarchical multi-agent architectures with meta-orchestrators spawning domain specialists. Survey paper.

Tools:

Tool Autonomy Domain Key Result
A-Lab L5 Materials 71% success, 10x speedup
ChemCrow L4 Chemistry 18 expert tools
Coscientist L4 Chemistry <4 min protocol design
FutureHouse L4-5 Biology 4 specialized agents
Edison Scientific L5 Multi $70M seed, 79.4% accuracy
Agent Laboratory L4 ML 84% cost reduction

Tool Details:

  • A-Lab (Berkeley Lab) -- Fully autonomous solid-state synthesis. 41/58 DFT-predicted materials synthesized in 17 days, 71% success, minimal human intervention. 10x data collection speedup. Published in Nature. L5 Full.
  • ChemCrow -- LLM for chemical tasks with 18 expert-designed tools. Finds molecules, plans synthetic routes, executes synthesis on cloud robotic platforms. L4 High.
  • Coscientist -- LLM-driven system for autonomous chemical experiment design, planning, robotic control. Successfully optimized Nobel Prize-winning palladium-catalyzed cross-couplings without human intervention (<4 min protocol design). Published in Nature. L4 High.
  • AlabOS NEW -- Autonomous Laboratory Operating System. Reconfigurable workflow management for autonomous materials labs. Orchestrates instruments, robots, and AI planners. L4 High.
  • FutureHouse -- Nonprofit AI-for-science lab (SF, launched Sep 2023). Building AI Scientist with world models. Specialized agents: Crow, Falcon (literature search), Phoenix, Owl (experimental design). L4-5.
  • Edison Scientific NEW (FutureHouse spinout) -- $70M seed at $250M valuation (co-led Spark Capital, Triatomic Capital). Kosmos platform: reads 1500 papers, runs 42k lines of analysis. Beta users: 6 months work -> 1 day, 79.4% conclusion accuracy. L5 Full. Commercial.
  • Agent Laboratory NEW -- o1-preview driven, 84% reduction in research expenses. Supports literature review -> experiment -> report writing as end-to-end pipeline. L4 High.

Code Quality & Review

Tools:

  • CodeRabbit -- AI code review assistant. Reports AI-generated code has 1.7x more issues, 2.74x more security vulnerabilities. Achieves 4x faster PR merge times. Essential for quality assurance in vibe coding workflows. L2 Partial. Commercial.

Stage Summary

Finding Detail
Developer adoption 92% of US developers use AI coding tools daily (2025)
Code generation 41% of all code globally AI-generated, 256B lines in 2024
Quality concerns 1.7x more issues, 2.74x more security vulnerabilities (CodeRabbit 2025)
Productivity paradox Developers felt 20% faster but took 19% longer after debugging (Stack Overflow 2025)
Materials discovery Years -> weeks compression with self-driving laboratories
Market Lab automation $7.84B (2024) -> $14.78B (2034), CAGR 6.55%

Analysis & Interpretation

Tools and papers for statistical analysis, visualization, results interpretation, and reproducibility. (18 items)

Reproducibility Frameworks (Level 2)

Papers:

  • SciRep: Framework for Reproducibility NEW (arXiv:2503.07080, 2025) -- Configuration, execution, packaging of computational experiments. 89% success vs 61% for Docker/Singularity baselines. Addresses the reproducibility crisis in ML research. Benchmark paper.
  • ReproSchema NEW (JMIR, 2025) -- Schema-centric survey design for reproducible data collection. Reusable assessments, validation/conversion tools. Published in peer-reviewed medical informatics journal. Venue: JMIR (IF 7.4).

Tools:

Tool Stars Focus Type
Docker 100k+ Standard containerization OSS
Nextflow 25k+ Bioinformatics pipelines OSS
Snakemake 20k+ Declarative workflows OSS
MLflow 18k+ MLOps platform OSS
DVC 13k+ Data version control OSS
Apptainer 10k+ HPC containerization OSS
Weights & Biases 8k+ ML experiment tracking Freemium

Tool Details:

  • SciRep NEW -- 89% success rate vs 61% for other tools. Automated configuration, execution, and packaging of computational experiments into reproducible artifacts. L2 Partial.
  • Docker OSS -- (GitHub | 100k+ stars) Industry-standard container platform for reproducible scientific computing environments. Foundation for most reproducibility workflows. L2 Partial. Free.
  • Singularity / Apptainer OSS -- (GitHub | 10k+ stars) HPC-friendly containerization for scientific workflows. Runs without root access on shared computing clusters. L2 Partial. Free.
  • Snakemake OSS -- (GitHub | 20k+ stars) Declarative workflow management for bioinformatics and data science. Python-based rule definitions, automatic parallelization. L2 Partial. Free.
  • Nextflow OSS -- (GitHub | 25k+ stars) Scalable workflow orchestration for bioinformatics pipelines. Supports Docker, Singularity, cloud execution. Active nf-core community. L2 Partial. Free.
  • DVC (Data Version Control) OSS -- (GitHub | 13k+ stars) Git-like version control for datasets and ML models. Tracks data pipelines, enables experiment reproducibility. L2 Partial. Free.
  • MLflow OSS -- (GitHub | 18k+ stars) Open-source MLOps platform: experiment tracking, project packaging, model registry, deployment. L2 Partial. Free.
  • Weights & Biases -- (GitHub | 8k+ stars) ML experiment tracking, dataset versioning, model registry, hyperparameter sweeps. Industry-leading visualization. L2 Partial. Freemium.

Stage Summary

Finding Detail
Reproducibility crisis Only 19.5% of top ML conference papers (2024) provide code
SciRep advantage 89% success vs 61% for Docker/Singularity
2026 vision Automated reproducibility as byproduct of AI-assisted research

Writing & Publishing

Tools and papers for manuscript drafting, citation management, figure/table creation, and formatting. (10 items)

Level 1 -- Grammar & Style

Tools:

  • Grammarly -- Grammar, style, and tone suggestions. Widely used academic writing assistant with clarity improvements and plagiarism detection. L1 Assist. Freemium.
  • QuillBot -- Paraphrasing and grammar correction. Flow workspace for end-to-end writing process. Reliable for rephrasing, checking grammar, summarizing. L1 Assist. Freemium.
  • LanguageTool OSS -- Open-source grammar checker supporting 30+ languages. Free alternative to commercial tools. Self-hostable for privacy. L1 Assist. Free.

Level 2 -- Research Writing Assistants

Tools:

  • Jenni AI -- Research and academic writing assistant. Strong for citations, outlines, academic formatting. Specialized for research-driven writing tasks with inline citation support. L2 Partial. Commercial.
  • Yomu AI -- Excels at paragraph development and PDF interaction. Quick paper gist understanding for staying current with literature. L2 Partial. Commercial.
  • SciSpace (writing mode) -- 10M+ researchers (2026). Upload PDF, chat with paper. Semantic search, automated summarization, reference management, AI copilot for writing. L2 Partial. Freemium.

Level 5 -- Fully Automated Paper Generation

Papers:

  • The AI Scientist-v2 NEW (arXiv:2504.08066, 2025, Sakana AI) -- First AI-generated paper accepted through peer review (ICLR 2025 workshop). Removes human templates, uses progressive agentic tree search. Milestone paper. Venue: ICLR Workshop.

Tools:

  • Sakana AI - AI Scientist OSS -- (GitHub | 5k+ stars) End-to-end paper generation: hypothesis -> code -> experiment -> LaTeX paper. First system to produce peer-review-accepted output. L5 Full.

Stage Summary

Finding Detail
Gap Level 3-4 tools missing (between Jenni AI L2 and AI Scientist-v2 L5)
Need Automated draft generation with human refinement (RAG for review papers)
Milestone AI Scientist-v2: first AI paper to exceed human acceptance threshold (2025)

Promotion & Dissemination

Tools and papers for presentation slides, video/audio narration, social media, and homepage creation. (1 item -- critical gap)

Level 1 -- Slide Templates

Tools:

  • Slide Templates -- AI-suggested layouts and content generation for academic presentations. Basic automation of slide creation from paper content. L1 Assist.

Stage Summary

Finding Detail
CRITICAL GAP Promotion stage severely under-researched: 1 tool, 0 papers (96% gap vs Experiment 48 items)
Missing L2-3 Automated slide generation, poster creation from papers
Missing L4-5 Video narration, social media automation, conference talk generation
Potential additions Beautiful.ai, Tome, Gamma (slides); Descript, Synthesia (video); Buffer, Hootsuite (social media)

Cross-Cutting Tools

General-purpose platforms, multimodal AI, and infrastructure supporting multiple research stages.

Research Platforms

Papers:

  • AI for Science 2025 NEW (Nature, 2025) -- Interdisciplinary knowledge graphs, RL-driven closed-loop systems, interactive AI interfaces for scientific theory refinement. Venue: Nature.

Tools:

  • NVIDIA BioNeMo -- LLM for biology (announced fall 2022). Major expansion to full open development platform. Lab-in-a-loop workflows for biology and drug discovery. RNA structure, molecular synthesis, toxicity prediction. Commercial.
  • Eli Lilly x NVIDIA Partnership -- $1B 5-year strategic partnership. AI co-innovation lab (SF Bay Area). Generative AI for drug discovery. Partnership.
  • NSF RAISE -- Research in AI for Science and Engineering. Democratizing AI access for researchers through infrastructure and funding. Government initiative.
  • National AI Research Resources -- NSF infrastructure for democratizing AI access to computational resources for the research community. Government initiative.

Multimodal AI for Science

Tools:

  • InternVL3-78B OSS -- (5k+ stars) 72.2 MMMU (open-source record). Vision-language model for multimodal scientific tasks including figure understanding and data extraction. Open-source.
  • Qwen2.5-VL-32B-Instruct OSS -- (10k+ stars) Vision-language instruction tuning (Alibaba). Strong multimodal understanding and generation capabilities. Open-source.
  • GLM-4.5V -- MoE architecture, 3D-RoPE, 72.2 MMMU benchmark (tied with InternVL3 for top score). Advanced vision-language capabilities. Commercial.
  • MaCBench NEW -- Benchmark for chemistry/materials tasks: data extraction, experimental execution, results interpretation. Published in Nature Computational Science. Venue: Nature Computational Science.

Stage Summary

Finding Detail
Market size Multimodal AI: $391B (2025) -> ~$2T (2030), CAGR 35.9%
Strategic investments $1B+ partnerships emerging (NVIDIA-Eli Lilly)
Timeline compression Discovery cycles compressed from years to weeks with AI + autonomous experimentation

Related Resources

Meta-Research

  • Vibe Researching as Wolf Coming NEW (arXiv:2602.22401, Feb 2026) -- Introduces vibe researching concept using scholar-skill (26-skill plugin for Claude Code, full pipeline idea -> submission). Cognitive task framework: codifiability x tacit knowledge. AI agents excel at speed, coverage, scaffolding but struggle with theoretical originality.
  • VibeX 2026 Workshop -- 1st International Workshop on Vibe Coding and Vibe Researching @ EASE 2026. Mixed audience: junior/senior researchers, practitioners. Focus on autonomous AI agents in SE research.

Awesome Lists

Academic Conferences

Conference Relevance
ICLR AI Scientist-v2 first peer-reviewed AI paper (2025 workshop)
EASE 2026 Hosts VibeX 2026 workshop on vibe coding and vibe researching
NeurIPS PaperBench benchmark papers, ML reproducibility initiatives
ICML DeepCode benchmark (75.9% success on ICML 2024 reproduction)

Contributing

Contributions welcome! Please read the contribution guidelines first.

How to Contribute

  1. Add a resource: Submit a pull request with the new tool/paper. Include: title, URL, description (1-2 sentences), and classification (Stage, Level).
  2. Update existing entries: Corrections, updated metrics (GitHub stars, user counts), or additional information.
  3. Report gaps: Identify missing categories, tools, or papers via Issues.

Quality Criteria

  • Papers: Must be peer-reviewed, arXiv preprints, or published in reputable venues. Include arXiv ID or DOI.
  • Tools: Must be actively maintained (not archived), have clear documentation, and be relevant to research automation.
  • Descriptions: Concise (1-2 sentences), factual (no marketing language), include quantitative metrics when available.

Classification Guide

Level Definition Human Role Example
L1 (Assist) AI suggests, human drives Decides all logic Code completion, grammar
L2 (Partial) AI executes specific subtasks Defines task, validates Automated screening
L3 (Conditional) AI handles workflows with checkpoints Sets goals, approves End-to-end literature review
L4 (High) AI handles extended workflows Monitors for failures Self-driving labs
L5 (Full) AI operates independently None AI Scientist: idea -> paper

Stages: Survey, Ideation, Experiment, Analysis, Writing, Promotion

See CONTRIBUTING.md for the full classification decision tree and formatting standards.

License

CC0

To the extent possible under law, the contributors have waived all copyright and related rights to this work. See LICENSE for details.

Acknowledgments

This list was compiled through systematic web research (March 2026) covering academic databases (arXiv, Google Scholar, Semantic Scholar), GitHub repositories, and commercial platforms. The survey identified 107 resources (45 papers, 62 tools) across 10 research categories.

Maintained by: OpenLAIR Version: 2.0 (2026-03-16) Last Updated: March 2026

Citation: If you use this resource in your research, please cite:

@misc{awesome-vibe-researching,
  title={Awesome Vibe Researching: A Curated List of AI-Driven Research Automation Resources},
  author={Dingjie Song and Lichao Sun},
  year={2026},
  howpublished={\url{https://github.com/OpenLAIR/awesome-vibe-researching}},
  note={107 resources across 6 research stages and 5 autonomy levels}
}

About

No description, website, or topics provided.

Resources

Readme

License

CC0-1.0 license

Code of conduct

Code of conduct

Contributing

Contributing
Activity
Custom properties

Stars

10 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

 
 
 

Contributors