A small business owner does not need a $2,000-a-month retainer to start posting consistently. The bottleneck is not ideas. It is the production loop: plan, write, design, approve, publish, measure, repeat.
I run Postmately, an AI social media autopilot. The format is borrowed from the case studies that work best: one concrete workflow, real numbers, and the parts that did not work. This is our version, based on the workflow we run at Postmately and the failure modes we see in small businesses.
The short version: one website link becomes a 30-day plan, three carousels and one short video in the first 24 hours. The owner approves everything in Telegram. The agent publishes through official APIs and reads the numbers back. The autopilot costs $59 a month.
Here is the full workflow, including the mistakes.
The old way vs the new way
| Freelancer or agency | Postmately autopilot | |
|---|---|---|
| Starting price | $500 to $2,000 a month | $59 a month |
| First batch | 3 to 7 days | 24 hours |
| Content | Depends on the person | Plan, carousels, short video |
| Approval | Email, docs, calls | One tap in Telegram |
| Publishing | Manual or scheduler | Official APIs |
| Analytics | Monthly report | Per post, next to the content |
This is not a claim that an agency is obsolete. A good agency brings strategy, filming, community management and taste. The autopilot replaces the production loop for a business that already knows what it sells and does not have an in-house SMM person.
What one link actually gives the agent
A website is not just a source of text. It contains:
- the product and the price;
- the audience and the cases;
- the photos and the tone;
- the questions customers ask before they buy.
The agent reads the site, builds a month plan, and turns the plan into formats:
- three carousels for the main offer;
- one short video for the visual proof;
- text posts for the routine weeks;
- a CTA for each post.
For a detailing studio, the plan looks like this: before and after, the process, the materials, the price, the booking. For a furniture studio: the kitchen, the wardrobe, the measurement, the delivery, the review. The formats are different. The loop is the same.
The six-step workflow
1. Paste the link
The owner starts with one link and a short brief: what to sell, which channels, what to avoid.
2. Build the month plan
The agent returns a table: week, topic, format, hook, CTA. The owner edits the plan before any content is produced. This is the cheapest place to change direction.
A prompt that works:
You are the content lead for [business].
Read https://example.com.
Build a 30-day plan for VK and Telegram.
30 posts in total: 8 carousels, 4 short videos, 18 text posts.
Show the hook, the format and the CTA for each post.
Do not publish. Send the drafts to Telegram for approval.
3. Make the first batch
The first batch is deliberately small: three carousels and one short video. The owner sees the visual style, the text length and the tone before committing to a month.
The carousels are rendered as 1080x1920 slides. The video is vertical, with a hook in the first two seconds and captions. The agent does not generate fake photos. It uses the owner's library or a licensed stock pack.
A carousel prompt that works:
Build a carousel for [offer] from https://example.com.
Six slides, 18 words per slide maximum.
Slide 1 is the hook, slide 6 is the CTA.
Use the photos from the site, not stock.
Do not invent prices or guarantees.
The first version of this prompt asked for ten slides and set no word limit. The result was a wall of text that nobody would read on a phone. The limit is the feature.
4. Approve in Telegram
Every draft goes to Telegram or email a day before its date. The owner can approve, edit or skip. Nothing goes out without a tap.
This is the most important part of the workflow. AI is good at production. The owner is still responsible for the claim, the price and the brand.
5. Publish through official APIs
The agent publishes to the channels that are live in the workspace. In our case: VK, Telegram, YouTube, TikTok, Pinterest, Bluesky and Mastodon. Instagram, Facebook, LinkedIn and Threads are waiting for platform review.
No shared logins. No browser automation that pretends to be a person. The owner can disconnect a channel at any time.
6. Read the numbers
After a week, the agent reads views, comments and clicks for each post. The next plan starts from what worked, not from a content calendar built in a vacuum.
The weekly review prompt:
Read the last 10 posts from the analytics.
Group them by format and by topic.
Tell me which two posts got saves and replies, not just views.
Propose three topics for next week that repeat the pattern.
Do not schedule anything yet.
The first version asked for "best performing posts". It returned the ones with the most views, which were also the ones with the most reach from a repost. Asking for saves and replies instead changed the answer completely.
The first two weeks, day by day
The work is not the content. It is the waiting for approvals and the corrections.
- Day 1. Paste the link, write the brief, get the plan. Twenty minutes of owner time.
- Day 2. Edit the plan. Kill two topics that sound like a brochure, add one that customers actually ask about.
- Day 3. The first batch is rendered: three carousels, one video. The owner watches all four once.
- Day 4. Approve, edit or skip. This is the only manual step that repeats every week.
- Day 5. The first posts go out on two channels.
- Day 12. Read the numbers from the first week and adjust the next batch.
The credits are easy to count: three carousels at 20 credits each plus one video at 60 is 120 credits. That is exactly the free tier, which is why the first example can be free.
Reach in week one is not the goal. A realistic first week for a new account is a few hundred views per post and a handful of saves. If the first carousel gets 400 views and the video 1,100, that is normal. What matters is whether the saves and replies come from the right people. A post with 300 views and 12 saves from potential customers is worth more than 3,000 views from people who will never buy.
The failure modes we see most often
Most of them are not technical.
Generic AI captions
The first drafts from a generic model are usually grammatically fine and completely forgettable. The fix is not a better model. It is more context: the owner's words, the customer questions, the real objections, the price.
Publishing without approval
Direct publishing looks faster, but it removes the quality gate. In practice, the owner catches a wrong price, a missing disclaimer or a tone that sounds like a different company. Approval is not bureaucracy. It is the quality gate.
Too many channels at once
In our own workspace, we connected six channels on the first afternoon because it took minutes and felt like progress. The next batch went to all six. The Telegram post was too long, the Pinterest pin had no vertical crop, and the TikTok caption read like a press release. Nothing was broken. All of it was mediocre.
We went back to two channels for the next two weeks. The formats got better because there were only two sets of rules to satisfy. The rest came back one at a time, each with its own batch and its own review.
Chasing trends without a plan
A trend can bring one good post and a month of inconsistency. The plan keeps the account recognizable. Use trends as an exception, not as the strategy.
Expecting viral reach in week one
The first month is for consistency and learning. A new account does not have the trust or the data to go viral on demand. The goal is a repeatable loop.
The guardrails
- Official APIs only.
- The agent never sees the social network password.
- The owner approves before publishing.
- The live channel list is visible in the workspace.
- The owner can disconnect any channel.
- Paid partnerships are labelled as paid.
These are not features to hide in a settings page. They are the reason a business can hand production to an agent without handing over the account.
The only rule
If the owner would not say it to a customer in person, it does not go out.
The economics
The autopilot is $59 a month. It includes 6,000 credits, about 300 carousels or 100 videos, and up to 50 channels. There is also a free start with 120 credits and 5 channels, which covers the first example: a plan, three carousels and one short video in 24 hours.
A part-time social media manager costs more than that before taxes, tools and management. A small agency retainer usually starts somewhere between $500 and $2,000 a month, depending on scope. The autopilot is not cheaper because it is worse. It is cheaper because the production loop is automated and the owner keeps the approval.
What it does not replace:
- strategy for a new market;
- filming a founder or a product;
- community management and direct messages;
- a creative agency for a launch campaign.
If the business needs those, the autopilot can still produce the routine content while the agency handles the campaign.
How to try it
Start with the free example. Send one link, get a month plan, three carousels and one short video in 24 hours. If the style fits, connect two channels and run the autopilot for a month.
The MCP setup for agents is here:
https://postmately.com/en/skills/
The product overview is here:
https://postmately.com/en/
The skill file for Claude Code, Codex and Cursor is here:
https://postmately.com/skills/postmately/SKILL.md
The takeaway
The winning workflow is not "let the agent post everywhere". It is:
- One clear source of truth: the website.
- One plan the owner can edit.
- One small first batch.
- One approval tap.
- Official APIs.
- A measurement loop.
That is how one link becomes 30 posts without an SMM agency, and without losing control of the account.
I am Leonid, founder of Postmately. We build an AI social media autopilot for small businesses and AI agents. If you are testing a similar workflow, the MCP page above has the current setup.
Top comments (0)