Running an AI agency in Russia since 2014. Notes on AI sales agents, marketing agents and what works for real business.
Novosibirsk, Russia
Joined November 2024
- Tweets463
- Following5
- Followers2
- Likes1
Pinned Tweet
I run La Chatte, an AI agency in Novosibirsk. We build AI that answers leads in seconds, writes marketing and tracks competitor prices. Here I post what works, what does not, and what it costs. No hype.
Every new model is "the smartest ever" in its launch post. Run it on your own task and the ranking often changes. Before you pick one, test it on 20 real examples from your business. It takes an afternoon.
Model prices keep falling. For most business tasks a cheaper model is enough. What decides quality is what you feed it: your prices, your rules, your past answers. Choose the model on your own data, not on a leaderboard.
IT job posts keep asking for every framework at once and paying for one. If you hire, write what the job really is. A wish list does not attract good people. It makes them close the tab.
With an AI assistant, one developer can now do some work that used to need a small team. Writing code got faster. Deciding what to build, and checking that it works, still takes a person.
Owners update the CMS. Nobody updates the contact form plugin. "It works."
But the form is the one place anyone can send something to your server. An old bug there is an open door.
30-minute check:
- last form plugin update
- any unused plugins
- who owns updates, by name
AI agents are the loudest phrase in tech right now. Plain version: a program that takes a task, does the steps and checks its own work. The useful ones do one boring job well, like answering a lead or updating a price list. No magic.
We have built software in Russia since 2014. The biggest lesson: clients do not buy AI. They buy a problem going away. Start the conversation with the problem.
If an AI project has no number it should move, do not start it. Replies faster? More leads that reach a call? Fewer manager hours on routine? Pick one, measure before, measure after.
Price tracking sounds easy until you do it. Same product, different name in every shop, different city, different price. Most of the work is matching products, not scraping pages.
Small retail question we hear a lot: "Can a new shop assistant know every rule on day one?" With an assistant in the messenger that answers from the company's own documents, yes. Without it, they ask the manager 40 times.
Honest take: a chatbot that only answers FAQ is not an AI salesperson. A salesperson asks questions, finds out what the buyer needs and moves them to the next step. That is the whole job.
A lead writes at 11pm. A human answers at 9am. By then the lead has talked to three competitors.
The fix is not a smarter bot. It is a bot that knows your prices and answers in seconds.
Most "AI sales bots" fail for a boring reason: nobody told them what the company actually sells. The model is fine. The product list, prices and rules are missing.
экосистему — от инфраструктуры до конечных решений. OpenAI это понимает и пытается контролировать весь стек технологий, но в условиях ценовой войны это уже не выглядит гарантией успеха.
Что думаете, есть ли будущее у таких стратегий, или рынок ИИ-сервисов так и останется
убыточным?
Рогачев Дмитрий: ИТ бизнес без рокетсайн 🚀 t.me/itrubiz
Компании по-прежнему готовы платить тем, кто умеет решать реальные задачи, а не просто красиво рассказывать на собесе. Главное — не вестись на откровенно тухлые условия и выбирать, где работать, а не просто «лишь бы взяли».
Как у вас с поиском работы или наймом? Чувствуете, что
стало сложнее?
Рогачев Дмитрий: ИТ бизнес без рокетсайн 🚀 t.me/itrubiz