What AI Actually Is (and Isn't)
A practical route into AI—explained from zero, sourced where the facts move, and organized so you can read, reference, and return without scrolling through one enormous document.
Choose a way in
Start with your question.
The guide still works in order. These shorter paths are for readers arriving with a specific job to do.
If you read the news
Make sense of the headlines
If you want to build
Do the work yourself
If you're picking tools
Compare agents by harness, not hype
The quick tour
A short orientation
If you're non-technical
Skip the engineering deep dives
If you're investing / exec
Money and the board
The publication
Five durable surfaces.
Read long-form explanations in the Guide. Use the other sections when you need a definition, coding-app vocabulary, a product comparison, or a dated market snapshot.
Guide
8 readings that build a durable mental model, one page at a time.
02Reference
331 plain-English terms with permanent, shareable pages.
03Workshop
216 coding-app terms—Git, shell, SQL, packages, CI, and harness jargon.
04Workbench
10 agent products compared by what they can see, do, and verify.
05Board
A dated map of the model, compute, and silicon layers. Snapshot 2026-07-13.
AtlasScope · Now
The guide is becoming a publication.
Current focus: Ship multipage Working Knowledge (guide/reference/workshop/workbench/board). See what Atlas is working on now →
After the reading
Where to actually begin.
You don't need to install anything or write a line of code. Six moves turn the guide into something you can use this week.
- Pick one assistant and live in it.
Open ChatGPT, Claude, or Gemini and use it daily for a week on real tasks — email, summaries, planning. Fluency comes from reps, not reading.
- Give it the context this task needs.
Some products can store preferences or retrieve past information; the model still answers from the context assembled for this turn. State the goal, show one example of “good,” and name the format you want back.
- Verify anything that matters.
It writes what sounds right, which isn't the same as right. Check every fact, figure, quote, and citation before you lean on it.
- Feed it your own material.
Paste in the document, the thread, the spreadsheet. The moment it answers from your stuff instead of the open internet, it gets genuinely useful.
- Start free; pay when it buys back time.
The free tiers are enough to learn on. Upgrade only once a tool is clearly saving you hours of your week.
- Be careful the moment it can act.
Answering is low-stakes; doing — sending, buying, deleting — is not. Hand an agent real access slowly, and watch what it does.
How to read the whole field.
Researchers live in the architecture, training, and prompting layers. Builders and operators care about inference, tasks, and evaluation. Safety and policy people orbit alignment. Investors track the hardware — with scaling laws as the thesis tying it together.
And the one thing to never forget: these systems are built to sound right, which isn't the same as being right. Anything that matters gets an independent check.
On the numbers.
Capital spending, product behavior, regulation, and infrastructure relationships move quickly. Each time-sensitive section carries a snapshot date, and its sources show when they were checked.
Reported facts, forecasts, compiled estimates, and editorial interpretation are labeled differently. Treat any specific number as a snapshot, not a fixed point.
Decoded & designed · Working Knowledge · 2026 Edition
Time-sensitive claims carry checked dates and source links.