AI use cases
3056 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.
-
Learning from large datasets
— AI systems are trained to learn and produce intelligent behavior without humans writing every rule line by line.
-
Automating compliance workflows
— AI can handle document processing, compliance checklists, and initial reviews while an expert retains final signoff and risk responsibility.
-
Supporting fractional executive work
— AI can help fractional CFOs close books and run scenarios faster, increasing the number of clients one person can handle.
-
Accelerating software documentation and onboarding
— AI enables one software implementation expert to take on more clients.
-
Processing business documents
— AI can read documents, sort them, extract required data, and flag errors in bookkeeping, billing, and payroll workflows.
-
Handling basic administrative work
— AI agents are expected to handle inbox management, scheduling, data entry, and basic bookkeeping for a couple of dollars an hour.
-
Answering user questions directly
— ChatGPT can answer questions without users clicking through to affiliate sites.
-
Generating and publishing internet content
— AI has flooded the internet with content, contributing to the decline of affiliate SEO.
-
Radiation testing and mitigation
— Measure GPU failure behavior under heavy ions and high-velocity protons and inform shielding and software strategies for bit flips.
-
High-power inference on satellite imagery
— Process satellite imagery in orbit and downlink insights such as vessel coordinates instead of transmitting all raw data.
-
Training and fine-tuning models in orbit
— Demonstrate that AI model workloads can run on satellite-based GPUs.
-
Operate a personal marketing system
— Bring together calendars, social-media schedules, media files, client work, community activity, and SEO workflows in one interface.
-
Analyze LinkedIn posts
— Extract post URLs, engagement scores, formats, and hooks to inform social-media content strategy.
-
Transform YouTube videos into SEO-oriented blog posts
— Repurpose video content into searchable website content using keywords, links, sitemaps, and other SEO instructions.
-
Convert an existing Claude project into a skill markdown file
— Preserve the project's instructions, memory, and recent chat context in a reusable format.
-
Neural-network design
— Apply insights from modeled biological neural connectivity to AI systems.
-
Improving frontier models and their efficiency
— Use AI systems to contribute to the development of subsequent models.
-
Cybersecurity testing by an AI agent
— Evaluate an unreleased model's ability to identify and exploit vulnerabilities.
-
Training frontier AI models on books
— Acquire additional training data and improve model capabilities.
-
Ask for the latest news and sentiment about XRP, Luxo and TRON.
— Get a direct, consolidated overview instead of manually searching social-media posts.
-
-
Payment tracking
— Double-check incoming payments and log them in an Excel sheet.
-
Research and prospecting
— Conduct research and prospecting while recording results in a human-readable spreadsheet or table.
-
Short-form video editing
— Retrieve videos from Google Drive, cut them up, and add captions, producing approximately 80% of the finished result.
-
File organization
— Organize accounting, vendor, and HR files across a Dropbox account.
-
Multi-step document and file processing
— Search an inbox, locate an invoice, create a job folder, move a PDF, extract information into an Excel sheet, save it to Dropbox, and send a link.
-
Administrative and central-office workflows
— Set up jobs, process field reports and logs, work across QuickBooks, PDFs, contracts, folders, and related business records.
-
Writing, reviewing, and shipping code
— Create a software factory in which agents handle the full coding workflow.
-
Model routing
— Use the appropriate model for each task to reduce token costs while retaining stronger models for demanding work.
-
Conversation grading and self-improvement
— Evaluate chats against a rubric and trigger fixes for results below the required score.
-
Production activity summarization
— Summarize events for paid customers and surface important activity and bugs for daily review.
-
Browser-based end-to-end testing
— Test signup, case creation, client onboarding, and discovery flows, then identify and help fix production bugs.
-
Running software-development agents in cloud virtual machines
— Work on multiple coding tasks in parallel while keeping development environments separate.
-
Generating proofs that are later formalized in Lean
— Convert AI-generated mathematical arguments into formally verified presentations.
-
Generating a counterexample for non-Sothic groups
— Show that some infinite mathematical objects cannot be approximated by any finite set.
-
Combining ideas across mathematical disciplines
— Create solutions by connecting areas of knowledge that individual human specialists may not understand deeply enough to combine.
-
Producing mathematical proofs and new mathematical results
— Address longstanding open problems that had resisted human mathematicians and supercomputers for decades.
-
Predict when an emergency may occur
— Help a volunteer emergency-response network position and prepare its volunteer force for potential incidents.
-
Inference serving
— Process text, audio, images and video using local model engines.
-
Generating embeddings
— Support RAG and related language-model applications.
-
Building AI agents and tool-using systems
— Use local models for agent workloads and tool usage.
-
Retrieval augmented generation
— Use local language models to support RAG workflows.
-
Running GPT-style models as AI assistants and code editors
— Provide local assistance and coding support from a personal computer or laptop.
-
Performance and bias monitoring
— Deployed AI systems are monitored for performance, bias drift, and unintended consequences.
-
AI system development and deployment
— AI systems are developed and deployed with controls within a governed life cycle.
-
Loan decision-making
— An AI system denies a bank loan, illustrating the need for explanation, appeal, validation, monitoring, and accountability.
-
System evaluation
— Critically assess agent performance, identify how it should improve, and guide development and deployment metrics.
-
World simulation
— Generate realistic behavioral and sensor scenarios for closed-loop training and evaluation, including rare situations not previously encountered in the real world.
-
Semantic scene understanding
— Interpret complex situations, such as a vehicle on fire, and choose an appropriate route or action.
-
Behavior prediction and planning
— Understand how other actors behave and generate the vehicle's driving decisions.
-
Multimodal perception
— Fuse camera, LiDAR, and radar inputs into a precise view of the surrounding world.
-
Autonomous driving
— Operate a vehicle safely and smoothly without a human driver in the physical world.
-
Identify personal use cases through an interview
— Help a person discover ways AI could assist with their work, family responsibilities and recurring problems.
-
Assist with research and data analysis
— Shorten the cycle between asking a research question, retrieving data, checking results and refining the analysis.
-
Answer customer questions in call centers
— Automate an increasing share of customer interactions that were previously handled directly by human agents.
-
Execute defined projects and generate software
— Allow people to focus on defining the problem and evaluating the result while AI handles more execution.
-
Rewrite restaurant-template listing text for a pickleball directory
— Generate information about South Austin Rec Center and why it is a good place to play pickleball in Austin, Texas.
-
Refactor codebases and synthesize regulatory documents with Claude
— Claude was presented as an enterprise assistant for complex, sustained tasks.
-
Run AI search experiments
— Google used an A/B test to opt people into AI search.
-
Train AI systems on YouTube's video corpus
— YouTube's large video archive provides data for the AI era.