AI use cases
2978 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.
-
Multi-model arbitration
— Collective Intelligence sends requests to several models, arbitrates between them, and returns reasoning metadata, model selections, costs, and disagreements.
-
Drafting email replies
— Agentic Inbox uses a Workers AI agent to draft replies for confirmation before sending.
-
Interacting with virtual machines through generated code
— Models can use code, conditionals, loops, installed binaries, files, and data sources to complete complex tasks.
-
Coding and long-horizon software work
— Poolside uses coding tasks as a route toward general reasoning and AGI.
-
Modifying data and model-training pipelines
— Agents increasingly improve the systems used to train future models.
-
Evaluating model-run results
— Agents assess experiment outputs and help determine subsequent changes.
-
Launching training jobs
— Agents automate the execution of model experiments.
-
Writing training and research code
— Agents increasingly implement researchers' ideas inside Poolside's Model Factory.
-
Operating companies and industrial systems
— Potentially improve businesses, chip fabs, factories, robots, and other real-world systems.
-
AI safety research and alignment
— Evaluate model behavior, train against reward hacking, and help align future AI systems.
-
Bug detection and debugging
— Identify subtle bugs in training recipes, codebases, and distributed infrastructure.
-
Understanding unfamiliar codebases
— Rapidly gather context and implement complex changes in large software systems.
-
Building RL environments
— Create environments that provide training and verification signals for AI capabilities.
-
Training models
— Optimize models on tasks including image classification, video generation, image generation, game playing, and online learning.
-
AI research and development
— Improve algorithms, model architectures, training methods, and future AI systems through iterative feedback loops.
-
Automate operations work
— Reduce miscellaneous founder tasks and increase time spent on higher-leverage work.
-
Evaluate prompts and model changes
— Measure output quality and prevent regressions in notes and action-item extraction.
-
Review pull requests
— Check code quality, architecture, product decisions, and copy for consistency.
-
Write software end-to-end
— Have agents implement work while humans revisit and review the results.
-
Support customer and sales operations
— Investigate customer emails, organize inbound sales work, and make activity visible to the company.
-
Prepare for meetings
— Combine emails, calendars, and past meetings to generate interview preparation.
-
Capture and organize company conversations
— Record and transcribe meetings, produce notes, assign action items, and preserve company context.
-
AI- and robotics-enabled manufacturing
— AI is described as converting human factory labor into energy consumption by coordinating robots, potentially making energy the dominant cost of manufactured goods.
-
Directly powering an AI chip with nuclear electricity
— Valar connected an NVIDIA Blackwell system directly to its reactor and used it to host a website.
-
Power demand for AI compute
— AI compute is presented as a major driver of increasing U.S. electricity demand.
-
Identify an advertising outlier with unusually high impressions relative to its age.
— Prioritize advertisements that may be receiving comparatively high spending or early traction.
-
-
-
Work augmentation
— AI is presented as a way for people to perform five or ten times as much work and create new business opportunities, though adoption is a condition.
-
Personalized agents and assistants
— Zuckerberg's proposal for a personal AI agent for everyone is discussed alongside Grokbot and other personalized AI systems.
-
Coding and software development
— Anthropic's coding focus and OpenAI's subsequent pivot toward coding are discussed as major drivers of AI adoption and revenue.
-
Provide autonomy for specialized industrial machines
— Enable robots and vehicles to move and act in industrial environments at scale.
-
Automate mining operations
— Increase productivity, reduce operating expenses, improve mine safety, and process materials into metals or minerals.
-
Automate food production and logistics
— Produce and deliver meals efficiently enough to approach the cost of going to the grocery store.
-
Automate operations across industrial sectors
— Combine software, sensors, robotics, and AI to operate physical industries.
-
Dental back-office administration
— Daydream Dental uses AI, computers, and specialized staff to handle insurance calls, billing codes, claims, and patient insurance checks for dentists.
-
Control a humanoid robot from camera input
— Convert visual information directly into motor and joint commands for autonomous physical work.
-
Manage personal and professional tasks
— Track engineering projects, recruiting, email, Slack messages and to-do lists for Brett.
-
Operate a general-purpose computer through a virtual computer environment
— Complete tasks such as building financial models, booking flights and ordering DoorDash without relying on consumer APIs.
-
Candidate evaluation
— Use independent structure prediction, confidence estimates and generation diversity to assess generated designs before laboratory testing.
-
Selectivity and cross-reactivity design
— Design molecules to bind desired targets or variants while avoiding similar proteins and unwanted interactions.
-
Sequence-structure co-design
— Iteratively generate a protein sequence and three-dimensional structure that are mutually self-consistent.
-
Protein binder design
— Generate candidate proteins or antibodies designed to bind a specified target.
-
Protein structure prediction
— Predict the three-dimensional structure produced by a protein sequence.
-
Training, inference, backtesting and reinforcement learning
— AI infrastructure can shift capacity among these workloads to improve utilization and economics.
-
Cybersecurity attack and defense
— Advanced AI systems may be able to compromise systems, while defenders can use comparable models to improve security.
-
Coding, research and tool calling
— Muse Glimmer is designed to support agentic workloads on a laptop.
-
Generate voice narration for market updates.
— Add a voice-reading feature to the Bullmania AI product.
-
-
Research current crypto activity and compare Pump.fun with the FOMO app.
— Provide financial data, trend comparisons and summaries of current on-chain developments.
-
Review recent commentary and sentiment about Alibaba.
— The AI is used to examine the narrative surrounding Alibaba's cloud, AI, valuation, and market position.
-
Determine whether ETH is still deflationary.
— The AI is used to check whether ETH supply is declining under current Layer 1 demand and Layer 2 usage conditions.
-
Summarize recent developments and sentiment around CoreWeave.
— The AI is used to interpret the company's earnings report, revenue growth, backlog, and narrative shift.
-
Investigate what is driving the rise in U.S. earnings.
— The AI is queried to distinguish AI-related growth from broader earnings and economic drivers.
-
Explain why a market asset has turned bullish and summarize relevant market commentary.
— The AI feature is used alongside chart and trend information to produce an explanation and identify analysts discussing the asset.
-
Select and run models locally
— Allow local model use and automatically select an appropriate model for an MLX architecture.
-
Search and retrieve video content
— Use an agentic RAG pipeline to identify relevant videos and return exact moments from their timestamped transcripts.
-
Transcribe audio and video
— Convert video speech into text using Whisper Turbo within the Docling pipeline.
-
Podcast analytics, A/B testing, and recommendations
— Improve titles, topics, guests, and content decisions for How I AI.
-