Where Intelligence
Enters the World
AI is not one thing. It becomes different tools in different domains: protein models in biology, vision transformers in medicine, time-series models in finance, retrieval systems in law, diffusion policies in robotics, neural operators in science, agents in software. This atlas maps what is used, where, why, and what still breaks — with sources, confidence flags, and a public verification queue so every claim can be challenged.
Every entry is tagged sourced / inferred / market context / forward-looking / needs verification. High-risk claims (clinical, finance, defence, robotics) are cited where possible and softened where not. The source library at the bottom of the page lists every reference and the open verification queue.
Trust & confidence
This atlas separates sourced facts, inferred relationships, market context and forward-looking claims. Papers, models, companies and use cases are labelled by confidence and maturity.
- sourced direct citation
- inferred multi-source consensus
- market context industry-known dynamics
- forward-looking roadmap signal
- needs verification placeholder
- production commercial / clinical scale
- early production narrow real deployments
- research frontier published methods
- experimental pilot or demo
- speculative claimed only
This atlas covers the major domains where AI is currently used, researched, commercialised or strategically deployed. It is designed to be expandable, source-backed and updated over time. Where a claim is not verified, we mark it — not invent it.
How to use this atlas
Featured questions
Curated entry points across audiences. The full question explorer is below.
Domain map
52 domains across 7 categories. Click any tile to open the profile.
Domain explorer
Search and filter the full map. Each profile shows architectures, papers, bottlenecks, opportunities and hype-vs-real.
Architecture explorer
The toolbox. Each entry shows what the architecture is, where it shines, where it fails and which domains use it.
Workflow explorer
End-to-end pipelines: drug discovery, radiology, quant research, robotics, AI coding, weather, manufacturing, support and more.
Papers & models
Landmark papers across foundation, vision, biology, robotics and scientific AI. Sources linked where possible.
Bottleneck map
16 categories of bottleneck across applied AI. The binding gate is rarely model quality.
Hype vs real
For every major domain: what is real today, and what is overhyped right now.
Founder opportunity map
Underserved workflows by domain. Each row shows buyer, pain, data and regulatory risk, and competition.
Where the money might be hiding
Not every AI idea is a startup. These dossiers separate validated demand, second-mover opportunities, boring cashflow businesses, venture-scale wedges and overbuilt zones — with the brutal capitalist read on each.
What actually blocks adoption
Detailed bottleneck intel: what it is, why it exists, who is solving it, and what would break it open.
How the biggest companies are playing the AI game
Strategy profiles of frontier labs, infrastructure giants, robotics players, healthcare and finance leaders. Architecture claims are best-effort and labelled by confidence.
The questions that define the AI era
Curated interview banks for Jensen Huang, frontier-lab CEOs, researchers, founders, investors, civilizational thinkers and domain experts.
Industry workflows — how AI actually enters the work
Step-by-step workflows: input, actor, AI role, data, architecture, output, human review and failure mode per step.
Where AI turns into money
Workflow → buyer → pain → AI stack → revenue logic. The bridge between capability and product.
Architectures that let you understand the game
You do not need to know every model. You need the architectures that keep reappearing across industries.
What to learn next
Practical study paths for builders, engineers, researchers, investors, robotics, bio, finance, infrastructure and interviewers. Each module is concrete.
Research frontier
Where new architectures and methods are still needed. Domains with research-frontier or experimental maturity.
Question explorer
Search the full Q&A by audience, category and difficulty. Every answer aims to teach a mental model.
Sources & assumptions
Methodology, source library and the audit log of claims tightened or flagged for verification.
Methodology
- Each domain entry is anchored on what AI is currently used for, not on what is claimed.
- Architectures, papers and datasets are matched per domain; specific authorship and venues are only asserted when verifiable.
- Companies are listed as market context. Adoption claims are softened or marked ‘needs verification’ where data is partial.
- Regulated domains (medicine, law, finance, defence, government) use cautious wording — AI assists, augments, triages, drafts, ranks, recommends.
- Forward-looking labels apply to any claim about future capability or commercial scale.
Source library
This is a living atlas. High-risk claims are tracked in the verification queue
(NEEDS_VERIFICATION_QUEUE in the data file) and updated as stronger
primary sources become available. Load ?devcheck=1 in the URL to
run an integrity check that confirms every sourced entry resolves to a
real SOURCE_LIBRARY or DOMAIN_PAPERS id.
Audit log
Three atlases, one map
The Accelerated Computing Atlas explains how AI is physically manufactured: power, chips, TSMC, HBM, NVIDIA, cloud and data centers. The Applied AI Atlas explains where that intelligence is deployed across domains. The AI Map sits underneath both with the taxonomy of techniques.