AI use cases
2977 ways people actually use AI, compiled from the videos in this
library. Each names the job and why it is worth doing, and links to the
moment in the source video.
-
Research software architectures
— Make a coding agent consult arXiv papers and converge on a recommendation with documented prior failure modes.
-
Execute large implementation plans
— Coordinate persistent worker loops for implementation, review, repair, and independent rechecking across context resets.
-
Evolve coding agents
— Preserve, mutate, or replace agents' prompts, code, tools, and dependencies based on benchmark evidence.
-
Translate speech locally
— Perform two-way voice translation on a Raspberry Pi 5 without internet access after setup.
-
Generate video and audio
— Create up to 15 seconds of video with native stereo audio from mixed text, image, video, and audio references.
-
Run persistent coding workflows
— Preserve sessions, memory, schedules, checkpoints, and reusable lessons while delegating work to isolated sub-agents.
-
Modify sandboxed applications
— Allow agents to work on private per-user copies while queuing simulated permission outcomes for bulk approval.
-
Compress coding-agent context
— Reduce upstream tool-output and history context while retaining expandable references.
-
PDF retrieval
— Token-Saver uses local BM25 and embedding search to send only relevant passages with page citations to Claude Desktop.
-
Video-generation acceleration
— ComfyUI-Spectrum-MiniMax-H3 forecasts intermediate hidden features to avoid running the transformer on every step.
-
AI-agent workflow evaluation
— RealReplicaBench tests whether agents can complete 107 long business workflows in reproducible service replicas.
-
Video generation
— MAGI-2-preview generates 10-second videos with synchronized audio from prompts or prompts plus still images.
-
Document visual understanding
— Doc7 renders document pages and sends them to a vision model to preserve charts, diagrams, relationships, and visible UI state in Markdown.
-
Video captioning and transcription
— Captivo generates captions on-device through whisper.cpp, while Open Edit transcribes video and designs stylized subtitles.
-
Large-model inference on Apple devices
— Swiftlet runs 35B and 80B Qwen mixture-of-experts models by streaming routed experts from SSD.
-
Local LLM fine-tuning
— Soup fine-tunes LLMs locally with QLoRA and includes reward-hack mitigation for GRPO runs.
-
Model-driven office editing
— Gen Office embeds model-driven editing into document, spreadsheet, presentation, and PDF workspaces.
-
Video-to-video generation
— Finger Frame AI regenerates a video while preserving the subject's movement and facial alignment.
-
3D scene reconstruction
— QuerySplat predicts 3D Gaussian splats, cameras, depth point clouds, and appearance from a handful of photos.
-
Personal assistant memory retrieval
— Waku-agent demonstrates an agent harness, loop, memory, and evaluation structure while deciding whether each turn needs memory.
-
Agent file-system execution
— Cloudflare Computer gives an agent a durable SQLite-backed file system projected into containers, shells, or ECMAScript workers.
-
Business-context retrieval
— AWS's context ontology accelerator converts databases and files into validated ontology and knowledge-graph context for agents.
-
Coding assistance
— Soul advisor, DSCode, Scopey, and related projects coordinate coding agents, isolate work, checkpoint changes, or detect scope drift.
-
Research and CRM maintenance
— CRM uses a durable research agent to process queued database rows while preventing unsupported personal inferences.
-
Large-model inference
— WASTE and Kimi K3 in C stream model components from storage and use caches to run models without conventional GPU requirements.
-
Code-change review
— Ratchet checks agent-produced edits for unnecessary dependencies, duplicate helpers, thin wrappers, and other patterns.
-
Writing analysis
— Humanizer CLI identifies mechanical signs of machine-written prose and provides editing guidance.
-
Procedure recording
— Skill recorder reconstructs the intent and ordered steps of a real work session for reuse as an agent skill or scheduled automation.
-
Workplace coordination and campaign production
— QM and the marketing team Eve template isolate agents, route work among specialists, and preserve shared context.
-
Generating interactive video worlds
— Lingbot World V2 generates long interactive video with causal frame generation, KV caching, and planning agents.
-
Building a personal model from application-use context
— Persome exposes receipts-backed memory over MCP so trusted clients can access work context outside the chat window.
-
Reviewing and improving user interfaces
— The skills collection gives coding agents structured checks for visual design, accessibility, layout, details, and product writing.
-
Rebuilding 3D objects from reference images using procedural code
— IMG3JS and the Three.js Object Sculptor Codex Plugin create animation-ready models with pivots, sockets, and colliders.
-
-
-
Evaluate multimodal perception
— PerceptionBench tests visual skills such as counting, depth, localization, OCR, and hallucination resistance before reasoning.
-
Generate multilingual speech and clone voices
— Audio8_TTS provides multilingual speech generation and zero-shot voice cloning in a 0.6 billion parameter model.
-
Optimize generative-engine visibility
— GeoLook samples brand visibility across AI engines, audits citation gaps, and produces implementation tickets and deployable assets.
-
Control robots from vision and instructions
— TurboVLA encodes vision and instructions separately and predicts continuous action chunks with a compact decoder.
-
Coordinate software delivery
— CodeJury divides delivery into six terminal-driven stages and uses multiple configurable review roles before execution.
-
Monitor and control coding agents
— Numbat applies local CEL rules to agent events for live alerts, forensic timelines, and supported pre-action blocking.
-
Reduce coding-agent tool context
— Ratel searches tool schemas and skill metadata and injects only the capabilities relevant to the current turn.
-
Route requests among models
— World model optimizer scores registered models on held-out tasks and serves an endpoint that chooses a model per request.
-
Large-model local inference
— DeltaFin streams routed experts from a CDN into a growing cache so a 2.8 trillion-parameter mixture-of-experts model can run on a 64 GB MacBook.
-
Robot task planning and control
— Tau Zero VLA splits long-horizon robot work into subtasks and uses a world model to search alternative actions when plans become uncertain.
-
Agent framework execution
— labs-OO-Agents hands methods with ellipsis bodies to a model while retaining deterministic Python implementations for methods with real bodies.
-
Agent-session accounting
— agentacct joins Claude Code and Codex logs with tasks and test evidence to estimate tokens and cost and distinguish verified from agent-reported work.
-
Writing and marketing-copy generation
— AI copywriter interviews the user, writes marketing copy, and removes listed AI-writing patterns without inventing product facts.
-
Agentic reinforcement-learning environments
— AgentENV provides sandboxed, forkable environments for Kimi K3's agentic RL training.
-
Coding-agent game generation
— Claude-of-Duty uses a fleet of coding agents to generate a browser FPS and evaluates the result with screenshot tests and profiling.
-
Scientific research assistance
— Open Science can read files, run Python or R, search the web, query permissioned life-science connectors, and preview generated scientific artifacts.
-
Home automation and personal information assistance
— Renfield's agent works across Home Assistant, calendars, email, media, and documents using local context from a RAG knowledge base and long-term memory.
-
News summarization
— News dashboard offers optional generated briefings to compress a backlog of technical articles.
-
Health-data pattern analysis
— Getbased creates deterministic biology scores and optionally performs model analysis without presenting the scores as diagnoses.
-
Coding-agent control and session management
— Fact adds prompt-cache stability, session compaction, crash recovery, and explicit policies around tool calls without replacing the existing coding agent or model.
-
Financial statement import and transaction analysis
— Expense Budget Tracker exposes a SQL API that an agent can use to import statements or analyze transactions.
-
Coding-agent operations
— Fractera provides browser-based coding tools, shared RAG memory, and an MCP-based design for coordinating coding agents.
-
Natural-language business intelligence
— Helical Insight provides natural-language analysis using the user's own language model.
-
Meeting transcription, summarization, speaker labeling, and search
— Steno processes microphone and system audio with models on the user's machine and makes meetings searchable.
-
Multi-model arbitration
— Collective Intelligence sends requests to several models, arbitrates between them, and returns reasoning metadata, model selections, costs, and disagreements.