How we run our agency with AI agents
A group of AI agents runs our website and six YouTube channels: finds news, selects what matters, writes, fact-checks, voices, edits, publishes and reports. The same work done by hand would cost $5,500-$6,300 a month.
The task
We need articles, market news, videos and reference tools all the time. We did not want to hire an editor, authors and a video editor just for that.
What we did
- Built a pipeline of seven AI agents. Each owns one step, like an employee in an editorial office, and passes the result on.
- Put checks between the steps: if a number cannot be confirmed against the source, the material is not published and goes to a person.
- Added an idea bank fed by competitors, audience questions and our best videos, so topics never run out.
- Moved the website to our own server so that the agents can publish news, articles and reference sections without a human editor (see the separate case study on the move).
- Added a watchdog: 19 system checks every three hours, daily backups and an alarm in Telegram if something breaks.
Before and after
| Before | After | |
|---|---|---|
| Articles | 1-2 a month, each written by hand over several evenings | Prepared from our videos in the author's style and published automatically |
| Video | Irregular, whenever we had energy to edit | 110 videos a week on 6 channels, on schedule, with voice-over and editing |
| Market news | We read and retold it when we found time | The system reads official channels, selects what matters and publishes an analysis within an hour of the post |
| Topic ideas | In our heads and notes | Automatic idea bank |
| Control | We noticed a problem when something stopped appearing | 19 checks every 3 hours, alerts in Telegram |
Results
| Before | After | |
|---|---|---|
| Cost of the work | $5,500-$6,300 a month by hand | About $170 a month on AI services and tools |
| Output | A few articles and irregular videos | 110 videos a week and a constantly updated news and reference section |
| Source posts processed | Whatever we could read ourselves | 400+ a week, read and rated automatically |
Takeaways
- Start with one process where the effect is visible fastest, then expand.
- Build quality control into the system. Fact-checking is the reason the output can be trusted.
- Keep people for decisions and approvals, not for moving files.
- The AI is the smallest part: the value is in the system around it (sources, rules, checks, publishing, monitoring).
Amounts are shown in US dollars, converted at a fixed rate.
More case studies
Moving a 700-page website from Tilda to our own server
We moved more than 700 pages and 650 blog articles from a site builder to a private server, kept every address, replaced forms and payment with our own, and made automatic content publishing possible.
Read the case study →
How SEO became our cheapest source of customers
We built a steady flow of customers from search engines for our own agency. SEO now pays back all the investment and brings customers 1.5-2 times cheaper than paid and social ads, with 1-2 hours a month of management time.
Read the case study →
Ozon: +$15,100 of monthly sales in one month
After a 35-parameter audit we scaled advertising, reworked product card content and changed business processes for the online store PROGamer, which sells gaming desks on Ozon.
Read the case study →Want results like this?
Request a free consultation. Tell us about your project and we will say honestly what we can do.
- Free, no obligation
- Reply within one business day
- Straight answer, even if the answer is "you do not need us"