Routine campaign setup can exhaust even an experienced specialist. Fitting headlines into strict character limits, cropping a dozen banners for different screens, entering ad labeling, setting up UTM tags, and double-checking targeting—all of this eats up half a day of monotonous work.
Expert Alexander Gorbachev (channel «Я — Авитолог») tested a practical hypothesis: can this entire cycle be entrusted to an autonomous agent powered by Claude? The goal of the test was an end-to-end launch of a live campaign in Avito Reklama with minimal human involvement. The promoted product was a website with a free guide on integrating AI into classifieds workflows, funneling subscribers into automated chatbots. The initial brief was short: "We are advertising a website with a free guide to attract subscribers. Creatives are attached. Generate the headlines and texts yourself based on the info on the site and send them to me for approval."
Character Limits and Banner Resizing
The first barrier was the platform's strict requirements for native ads. The dashboard enforces rigid boundaries: the headline can be no longer than 30 characters, and the main text up to 45 characters, including spaces and punctuation. At the same time, all-CAPS and pointer arrows are prohibited.
The neural network analyzed the landing page structure and prepared six pairs, strictly staying within the limits:
- "AI makes ads on Avito" (29 of 30 characters) | "Connect Claude or ChatGPT in 30-40 minutes" (44 of 45 characters).
- "AI prepares ads on Avito" (27 of 30 characters) | "Case: 30 min of AI work, 2 min of owner time" (44 of 45 characters).
- "A neural network runs your Avito" (25 of 30 characters) | "Texts, images, and publishing. No code" (39 of 45 characters).
The second difficulty came up with the creatives. The platform requires 4:5 and 4:3 formats. When uploading a universal banner, Avito's cropping algorithm trims the edges, cutting off the offer text. The agent noticed this cropping in the preview, independently generated versions in 4:5 and 4:3 proportions with safe margins, and re-uploaded the corrected files to the dashboard. On the banners, text took up less than 20% of the area with no visible phone numbers.
Division of Labor: Browser Agent, API, and Human Oversight
The official Avito Reklama API cannot create ad campaigns from scratch yet. You can manage bids and download statistics via the API, but building the initial structure requires working in the web interface.
That is why Gorbachev used a hybrid approach: the agent controlled the browser (More Akkauntov), logging into the dashboard as a specialist. Claude selected the "Target Actions" goal with cost-per-click (CPC) bidding, set a weekly budget of 20,000 rubles, configured geo-targeting for all of Russia, and filled in the legal ad labeling (advertiser TIN, contract number, and service code for transmission to ERIR via ORD). We covered how to connect knowledge bases and classifieds system regulations in our breakdown of automating Avito promotion with AI.
Setup from the brief to the "Ready to serve" status took 1 hour and 10 minutes. The specialist sent just three short confirmations in chat:
- "Texts look good, budget 20,000 ₽ per week, all of Russia."
- "Yes, all correct, proceed."
- "Launch it."
At the same time, two security boundaries remain:
- OS system barrier: the file selection dialog on macOS is a system process that the script does not have direct access to. It required a one-time screen access confirmation, after which the agent entered system file paths.
- Financial control: the neural network has no access to bank cards. At the final step, the dashboard requested a balance top-up, and the avitologist completed that payment manually.
The takeaway is simple: neural networks are great at taking over mechanical routine like ad formatting and labeling. But strategy, offer approval, and payment remain with the human.


