Zway.ai

How it works

AI plans it. A person ships it.

Zway runs growth marketing in two layers. An AI layer researches your company, writes the strategy, and drafts every asset against five source documents. A trained human operations team then publishes that work daily on your own channels, in your voice, and handles anything that needs judgement, standing, or a real conversation.

Last reviewed 27 August 2026

The source of truth

Everything reads from five documents.

Before anything ships, we write five documents about your company. Every agent and every operator reads them before doing any work. This is the whole quality control system: when output drifts, it is almost always because a document was wrong, and fixing the document fixes every asset downstream at once.

  • 01-product.md

    Product

    What you have built, who it is for, how it actually works, and the three things it does better than the alternatives. Written from your demo and a founder interview, then approved by you.

  • 02-market.md

    Market

    The buyer, the trigger that makes them look, the language they use for the problem, and where they already gather. This is the document that decides which channels are worth running.

  • 03-competitors.md

    Competitors

    Who you lose to and why, in plain terms. Includes what each competitor is genuinely better at, because an operator who cannot concede a point is not credible in a real conversation.

  • 04-voice.md

    Voice

    Sentence length, vocabulary, what you would never say, how you handle disagreement, and how much you swear. Built from your existing writing and calibrated against every draft you correct.

  • 05-content-strategy.md

    Content strategy

    The themes you are trying to own, the claims you are allowed to make, and the topics that are off limits. Everything downstream is generated against this.

You own these documents. They are written in the first month, corrected by you, and refreshed as the company changes. An agency that starts over every month is really just rebuilding this context badly, every month, from scratch.

The machine layer

What the AI actually produces.

The AI layer does the work that scales: reading, planning, drafting, watching, and reporting. It produces artefacts, not decisions. Every output below lands in a queue that a human works from.

Each agent in the Zway machine layer and what it produces
AgentWhat it produces
ResearchCompany, market, and competitor documents, refreshed monthly
StrategyThe 30-day plan: channels, volumes, owners, targets
BriefPer-asset briefs with angle, audience, sources, and constraints
DraftPosts, replies, articles, outreach, all in the approved voice
MonitorThreads, mentions, and questions worth a human answering today
ReportWeekly results against target, and the plan diff for next week

The human layer

Who actually posts, and how they are trained.

Operators are the product. They are hired for judgement rather than output speed, trained on your five documents before they touch an account, and assigned by channel fit rather than by whoever has capacity.

  • Growth operator

    Owns the account. Runs the weekly plan, posts, replies, and decides what to escalate.

  • Community operator

    Holds the accounts with standing. Participates in the subreddits, Discords, and forums where a new account gets ignored.

  • SEO operator

    Ships technical fixes as pull requests, publishes briefs, and owns the AI-search prompt coverage.

  • Creative operator

    Runs creator and UGC relationships past the first message, and owns the long-form calendar.

Training

Two weeks before touching a live account: platform norms, your five documents, and a supervised queue where every draft is reviewed by a senior operator.

Review gates

Anything naming a competitor, quoting a metric, making a claim about the product, or responding to criticism passes a second operator before it ships.

Escalation

Anything where a wrong move damages the brand permanently comes to you first. Operators are told explicitly that escalating is never penalised.

The operator roles are real jobs with real descriptions. If you want to see how we hire for them, they are listed on the careers page.

Your control

What you approve, and what runs without you.

You pick the approval mode per channel, and you can change it whenever you want. Nothing here is locked.

  • Full review

    Weeks one and two

    Nothing ships without your explicit yes. Slower, and the right setting for regulated categories or a brand-new account.

  • Spot check

    From week three

    You see everything in the queue and approve a sample. Operators ship the rest. Most companies settle here.

  • Autopilot

    By request

    Named channels run without approval, with an escalation rule for anything sensitive. You still see every asset after it ships.

What we will not do

The limits are the product.

  • We do not create fake personas, fake reviews, or sockpuppet accounts.
  • We do not astroturf communities or manufacture organic-looking demand.
  • We do not post from accounts you have not authorised.
  • We do not claim results you did not get, on your site or on ours.
  • We do not run a channel where the only way to win is to break its rules.

These are not disclaimers. They are the reason the work compounds. A community account that gets caught astroturfing is worth less than nothing, and it takes years to rebuild. The constraint is what makes the asset durable.

Read the full trust and ethics policy

Onboarding

The first thirty days.

  1. Day 1

    Founder interview and product demo. The 30-day plan is written and sent the same day.

  2. Day 3

    The five documents are drafted and sent for your corrections. First assets drafted against them.

  3. Day 7

    First assets ship. Accounts, access, and approval mode are set. Voice is already being corrected against your edits.

  4. Day 30

    First full reporting cycle. The plan is rewritten against what actually worked, not against what we assumed.

How it works, in detail

Who writes the first draft, the AI or a person?
The AI layer writes the first draft of almost everything, against the five documents and the approved voice. A human operator edits it, decides whether it should ship at all, and takes responsibility for it once it does. Nothing goes out without a person reading it first.
How many companies does one operator carry?
A small number, deliberately. Standing in a community cannot be split across dozens of accounts, and an operator who is spread thin starts posting generically, which is the exact failure mode we exist to avoid. Capacity limits are why we cap intake each month.
What accounts do you post from?
Accounts you own and authorise, or your own account with your sign-off. We do not create fake personas or sockpuppet accounts. Where an operator uses their own established community account, they participate as themselves and disclose the relationship when the platform or the context calls for it.
What happens if an operator gets something wrong?
It gets corrected publicly and fast, and the correction goes into the voice document so the same mistake does not repeat. You are told the same day, before you find it yourself. Anything that could damage the brand permanently is escalated to you before it ships, not after.
Can I approve everything before it goes out?
Yes. Full review is the default for the first two weeks and you can stay there permanently. Most founders move to spot checks by week three because the approval queue stops surfacing surprises, but that is your call and you can move back at any time.
How do you learn my voice?
From your existing writing, a founder interview, and every correction you make. Each edit you make to a draft is fed back into the voice document. In practice the correction rate drops sharply after the first fortnight, which is the point at which most founders relax the approval mode.
What do you do when there is no news to post about?
Most of what works is not news. Answering a real question in a thread, publishing the thing you learned building the product, and participating in a discussion you have a genuine view on all work without a launch. Announcement-shaped content is a small fraction of the plan.
What happens if we stop working together?
The accounts are yours, because they always were. The five documents are yours. The content is yours. There is no platform to be locked into and nothing stops working when the engagement ends, which is deliberate: the compounding asset is your channels, not our software.

See the plan for your company.

Tell us what you are building. We will tell you exactly what we would run in the first 30 days, whether or not you hire us.

Book a demo

30 minutes. You leave with the plan either way.