Antigravity for PMs
A free course · twelve lessons · nothing to sign up for

Most PMs use AI like a vending machine.

Paste in a fragment, get back a paragraph, re-explain the company tomorrow. Antigravity is built differently, and the difference only shows up if you work differently too.

The chat-window habit

Context you retype every morning

  • 01You explain the product, the personas and the metric before every task.
  • 02Sources live in your head, so nothing is checkable afterwards.
  • 03Output arrives as a wall of text you paste somewhere and reformat.
  • 04Quotes come back tidied. You find out in the review.
  • 05Every session starts from zero, so nothing compounds.
Working in a folder

Context that stays put and compounds

  • 01The brief, the personas and the metric definitions live in files you point at.
  • 02Every claim traces to a file in research/, or it is marked unevidenced.
  • 03Output is written back into the folder, next to what it came from.
  • 04You spot-check five quotes before anything reaches a deck.
  • 05Week four is faster than week one, because the folder got better.

What you are actually learning

Point at a file.
Read the plan.
Then let it work.

Format
12 lessons, 3 modules
Time
About 7 hours
Cost
Free, no account
Coding
None at any point
Practice company
TaskFlow (fictional)
Agent runrun-8f24417planning mode
>
plan4 steps · awaiting your approval
1read research/interviews-march.md — 38 quotes, 6 sessions
2group by theme; keep single-mention themes separate
3check every claim against context/company.md
4draft ≤1 page; cite the file behind each claim
okapproved → writing drafts/activation-problem.md
task list plan draft sources

An illustration of the loop the course teaches — a plan you can reject, then artifacts you can check. Not a live connection to anything.

View one

Editor

Your files, your document, and an agent panel beside it. For work that is a conversation: drafting, cutting, arguing, deciding.

View two

Manager

Several agents working in parallel while you are elsewhere. For work that is long, mechanical and needs no decisions from you.

What you review

Artifacts

Task lists, plans, screenshots, browser recordings. Read these instead of the transcript; they are where a wrong assumption shows up first.

Artifact

Task list

What it intends to do, before it does any of it.

Artifact

Implementation plan

Ordered steps and the files each one reads. Reject it here or not at all.

Artifact

Screenshots

What the agent actually saw when it went and looked.

Artifact

Browser recording

Your evidence for the number you quoted six weeks ago.

Artifact

File diff

Exactly what changed in your document, line by line.

Artifact

Run log

Which files were opened. If your notes are not in it, it guessed.

How the course works

Four habits, repeated until they are boring

Every lesson is a short piece of reasoning, an exercise you do in your own workspace, and a note on what usually goes wrong.

Habit one

Attach the file, do not describe it

An @ mention puts the actual text in front of the agent. Describing the same file from memory gets you an answer built on what documents like that usually say. The gap between those two outputs is the whole course.

  • @context/company.md
  • @research/interviews-march.md
  • @shipped/last-prd.md — for the house style
  • דwe’re a Series B PM tool” — not a source
Habit two

Read the plan before the work

Ask for the steps, the files each step will read, and the two places it is most likely to go wrong. A plan with no file names in it is a warning: the output will be generic, and you will not notice until someone senior does.

  • 1read research/support-export.csv
  • 2count tickets in first 7 days
  • 3check categories against context/metrics.md
  • !risk: 'other' may be 60% of rows
Habits three and four

Spot-check, then give it permission to find nothing

Check five quotes against the source, every time, for the first month — a tightened quote is what ends up in a deck. And end hard questions with “if the evidence does not support this, say so”, or you get manufactured rigour, which is worse than agreement.

  • quote 1 — verbatim, line 214
  • quote 2 — verbatim, line 88
  • ×quote 3 — tightened; two words removed
  • quote 4 — verbatim, line 301
Straight from the lessons

Six things the course actually tells you to say

Before you write anything, give me a plan: the steps in order, which files you will read for each, and the two places you are most likely to get this wrong.
Lesson 1.4Planning mode
Ask me the eight questions a sceptical director would ask before approving this, one at a time. Do not answer them yourself.
Lesson 2.1Write a PRD
Propose how to handle each problem you found. For each: what you would do, and what it would hide.
Lesson 2.2Analyse an export
What is the one question about this strategy that I have not asked and would least like to answer? Ask it. Do not soften it and do not answer it.
Lesson 2.3Pressure-test a strategy
Every claim about users cites a file in research/. If there is no source, write ‘unevidenced’ rather than softening the claim.
Lesson 1.5Project rules
If you are editing the workflow every time you run it, the task is not stable and the workflow is costing you more than it saves.
Lesson 1.6Re-runnable workflows

These are extracts from the lessons, not testimonials. This site publishes no reviews, no ratings and no customer quotations.

Before you start

The five things people ask first

Do I need to know how to code?+

No. Nothing in the course asks you to write, read or run code. Antigravity is a development platform, but the parts a PM uses are file management and conversation. If you can find a folder on your computer, you can do all twelve lessons.

Is this an official Google course?+

No. This site is independent and has no connection to Google LLC. Antigravity is Google’s product; we are readers of it, writing about how to use it for product work. The software is downloaded from Google and only from Google — we host nothing and distribute nothing.

What does it cost, and what do you want from me?+

Nothing, and nothing. There is no sign-up, no email capture, no account and no upsell. The site sets no cookies and runs no analytics or advertising scripts. Antigravity itself was free during its public preview when we last checked, on 7 September 2026; confirm the current position with Google.

Will this still be accurate in six months?+

Partly. The habits — grounding in files, reading the plan, spot-checking quotes — will outlast several interface revisions. Specific menus and button labels will not, and the preview moves quickly. Where a lesson leans on the exact position of a control, it says so and tells you what to search for instead.

I already use a chatbot for this. What actually changes?+

Three things: context lives in a folder instead of your head, output lands as files next to the material it came from, and long tasks produce a plan you can reject before any work happens. If none of those are your bottleneck, the honest answer is that the gain is small.

All questions, including the ones about data and privacy

Start with lesson 0.1.

Twelve lessons, about seven hours, nothing to sign up for. Begin by understanding what the thing is — including what it is bad at.

Begin the course Antigravity itself comes from Google, at antigravity.google

Independent and unofficial. This site is published by readers of Antigravity, not by Google. It is not affiliated with, endorsed by or connected to Google LLC. Nothing is hosted here — no software is mirrored, repackaged or distributed, and no installer, licence key, sign-in or payment is ever requested. Antigravity itself is downloaded only from Google, at antigravity.google. Never enter a Google password, a payment card or an API key on this site — nothing here needs one.