Guide
Autonomous GTM, with you approving every move
Autonomous GTM is go-to-market work run by AI agents: they read your business, plan the channels, draft the work and measure it. In InstinctGTM, you approve what goes out, or a rule you approved does, and every channel runs on a clock with kill rules locked before it starts.
The steps in order
Why fully autonomous is the wrong goal
An agent that sends without a person looking spends your name. A wrong message lands on your domain and your reputation, and no retry undoes it. The cost of checking a draft is a few seconds; the cost of an off-brand message to the wrong person is the relationship.
Unchecked agents also keep weak channels alive. Without a rule written in advance, every week of thin results reads as early days, and the budget drifts to whatever the agent finds easiest to produce.
The fix is two constraints that hold in the database rather than on a screen. Nothing goes out without an approval on record, from you or from a rule you approved. And every channel is judged on a date, against numbers chosen before the work started.
The InstinctGTM Loop, step by step
Nine steps, each one running in the product today.
Read your website and what you tell it
- What happens
- The agents read your site: what you sell, who it is for, the words you use. Then they ask once about what they could not infer. What you confirm is kept as fact and every later step reads it; what they only inferred is labelled as a guess and never becomes a claim in your name.
- What you do
- Paste your address and correct anything that is wrong.
Get a GTM plan of channels, each tied to one goal
- What happens
- The GTM plan picks the demand generation channels that fit your price, your hours and your budget, and ties each to the one number it is meant to move. Every change makes a new version, so you can see what the GTM plan said last month and why it changed.
- What you do
- Approve the GTM plan, or remove a channel you will not run.
Set the clock and the rules before work starts
- What happens
- Each channel runs as a test on a clock of 14, 21, 30 or 90 days, depending on how fast that channel shows a signal. Its success, kill and scale rules are written before the clock starts and locked by the database, so nobody moves the goalposts when the numbers come in.
- What you do
- Read the rules once. They are the ones the verdict will use.
Draft in your voice
- What happens
- The agents write the posts, emails and pictures each channel needs. Every draft is scored against your voice and passes a quality check for invented claims and stock phrasing before it reaches you.
- What you do
- Edit what is close, reject what is wrong.
Approve, or let a rule you approved say yes
- What happens
- Nothing is sent or published without an approval on record, and the database refuses a send that lacks one. Once a channel is running you can approve rules for it, such as a daily cap, the topics and the audience. Work that meets every rule is approved by the rule; anything else waits for you, with the reason.
- What you do
- Check the daily sample of what your rules approved. Mark two bad in a week and that channel goes back to asking you every time.
Run on the channel’s schedule
- What happens
- Approved work goes out on the schedule the channel set, through your own accounts and your own mailbox. A pause button holds the whole team from the next request: no new drafts, no rule approvals, and nothing the team scheduled goes out.
- What you do
- Nothing, unless you want to pause.
Measure each channel, on to the deal
- What happens
- Each channel is measured against its own rules: replies, clicks, visits, booked calls. Where you connect your site and your deals, a visit is followed to the booked call and the payment, so a channel is judged by buyers and not by likes.
- What you do
- Connect your site once, and your deals if you track them here.
Read the verdict on the clock’s last day
- What happens
- On the last day the verdict is read from the numbers by code, never by a model. Too little data extends the clock once. Success keeps the channel going. A miss changes one thing and tests again, and after two misses in a row the team proposes a new angle instead. A kill is always a proposal to you, never automatic; left unanswered for a week, it pauses the channel rather than killing it.
- What you do
- Answer the proposals in your Monday report.
Learn, and start the next GTM plan smarter
- What happens
- What the verdicts show goes back into the next GTM plan. Your edits become proposed preferences you confirm. A learning about the work itself is used only after it beats a holdout group in a real test and a person approves it.
- What you do
- Confirm or dismiss what it learned from your edits.
How the loop improves itself
Your messages
Every edit keeps the original draft beside what you kept. When you remove the same habit in most of your edits, it becomes a proposed preference, labelled as learned from your edits, and counts only after you say yes.
Your channels
Each channel is a test with its rules set first. The verdict on the last day decides what gets more time and what you are asked to stop.
Your business
Booked calls, deals and payments show which channels bring buyers. A learning about the work only shapes plans after it beats a holdout group in a real test.
Who does what
| Agents do this alone | Waits for your yes | Never automated |
|---|---|---|
| Reading your site | Anything sent or published, unless a rule you approved covers it | LinkedIn messages: drafted for you, sent by your hand |
| Drafting posts, emails and pictures | Any increase in spend | Anything that needs your login to another site |
| Measuring each channel | Pivoting or killing a channel | Your card details |
| The Monday report |
An example channel
| Channel | Cold email |
|---|---|
| Goal | Booked calls |
| Clock | 14 days |
| Variable under test | The first line names a dated change at the company |
| Success rule | Reply rate at or above the line set before the start |
| Kill rule | Reply rate below the kill line once the minimum sample is sent |
| Verdict on the last day | Read by code from the numbers; a kill is proposed, never applied |
| What carries forward | The winning first line becomes the next test’s starting point |
How to start
No workflows to wire. Tell InstinctGTM the goal and approve the work.
- Paste your website and get a free GTM plan.
- Approve your first channel and its rules.
- Read your first Monday report.
Questions people ask
What is autonomous GTM?
Go-to-market work where AI agents do the planning, drafting, running and measuring, instead of a person doing each step by hand. The useful version keeps a person on what goes out and on when a channel stops.
Can AI run my go-to-market on its own?
It can do most of the work: reading your market, drafting, scheduling and measuring. It should not decide alone what goes out under your name or when to give up on a channel, because both are cheap for you to check and costly to get wrong.
Is it safe to let AI agents send messages?
Only with a record of approval the software cannot skip. In InstinctGTM the database refuses a send without one, and a rule you approved can only approve work inside the caps, topics and audience you set.
How does a GTM engine learn?
From verdicts and from your edits. Each channel is judged against rules set before it started, and what wins becomes the next starting point. Your edits become preferences you confirm before they count.
What should never be automated in go-to-market?
Anything sent from your personal accounts on networks that forbid automation, such as LinkedIn messages, anything that needs your login, and any decision to spend more. Drafting those is fine; sending them is yours.
How long before results improve?
As long as the channel’s clock: 14, 21, 30 or 90 days, depending on the channel. A verdict needs enough volume to mean something, so the first clock tells you whether the channel works, and the next ones improve it.
Do I still need a marketer?
Not to produce the work week to week. You still need someone to decide positioning and approve what goes out, and in a young company that is usually the founder.