AI agents for business: sales in Instagram and Telegram, calls and support

An AI agent for business, also called an AI employee, is software that gets a goal rather than a script: take the customer from the first message to payment, a booking or a resolved question. It understands free speech, including Russian and Uzbek mixed in one message and voice notes, picks the next step itself and acts inside your systems: it replies in Instagram and Telegram, calls with a live voice, sends a Payme or Click payment link and fills in the CRM card. A scripted bot only reacts to pre-written keywords and buttons, and a human operator works only during a shift. An agent works around the clock and hands complex cases to a person with the full context. There are four main types: sales, support, voice and call analytics agents. Small and mid-sized businesses in Uzbekistan launch one themselves from 2 hours; an enterprise with integrations takes up to 30 days.

Demand for these agents in Uzbekistan comes from simple arithmetic. According to DataReportal's «Digital 2026: Uzbekistan» report, 33.1 million people in the country use the internet, or 89% of the population, and a customer's first contact with a business increasingly happens in a DM rather than on the phone. They write at 11 pm Tashkent time and expect an answer right away, not in the morning.

Below: how an AI agent differs from a chat bot and from an AI employee, which types exist, how a sales agent works in Instagram and Telegram step by step, a comparison with a scripted bot and a human operator, a buying checklist, and an honest answer on when you don't need an agent.

What is an AI agent for business, in plain words

An AI agent for business is a worker, not a chat window. OpenAI's guide to building agents draws the line clearly: systems that use a language model but don't let it control the workflow, such as simple chatbots and classifiers, are not agents. An agent chooses the next step itself and calls the right tool: the catalogue, the calendar, a payment link, the CRM.

Take a clothing shop in Tashkent that sells through Instagram. Dozens of messages arrive in an evening: «narxi qancha?», «do you have size M?», a voice note in Uzbek, a question about delivery to Samarkand. The agent answers each one by meaning, checks size and stock, sends a payment link and opens a deal in the CRM. In the morning the manager sees not a hundred unread chats but a list of paid orders and a few conversations the agent handed over with a note on why.

That gives you a simple test for any tool labelled «AI agent»: what appears in your systems after it has done its work? If it's only a chat thread, it's a bot. If it's a payment, a booking, a call and a filled-in customer card, it's an agent.

How is an AI agent different from a chat bot and an AI employee

The difference from a chat bot is who runs the process. In «Building effective agents», Anthropic splits systems into two classes: workflows, where the model follows predefined steps, and agents, where the model decides which step and which tool it needs. A scripted bot from a builder sits lower still: it matches the customer's words against keywords and doesn't understand meaning.

In practice it looks like this. A customer writes «qancha turadi», «сколько выйдет» or sends a voice note, the bot's rule doesn't fire, and the conversation ends with «please contact a manager». An agent understands the question in any wording, remembers the customer across Instagram and Telegram, and keeps working outside the chat: it calls, sends a payment link, creates a task. The keyword-versus-meaning difference is broken down in AI replies vs chatbot builders.

As for the AI employee, there is less confusion than it seems: it's the same thing. «AI agent» is the engineering term, and «AI employee (AI agent)» is the same agent named in business language: not a technology but an area of responsibility on your team. In Russian-language sources it appears as «ИИ-агент», with no change in meaning. Terms and levels of agent autonomy are covered in AI agent and AI employee: what it is.

What types of AI agents do businesses use

Four types of AI agents show up in customer-facing work. They differ not by the model inside but by channel and outcome: where the agent talks to the customer and what's left after the conversation. The first three talk to customers directly; the fourth works with your own team's calls.

• Sales agent in Instagram and Telegram. Replies in Direct and Telegram private messages, moves the author of a comment under a post or ad into Direct, qualifies with a few questions, recommends a product, sends a payment link and records the deal in the CRM. More: Sales AI and comment auto-replies.
• Support agent. Answers routine questions from the knowledge base: delivery, order status, opening hours, returns. Opens a prioritised ticket and hands a complex case to an operator together with the thread. More: Service AI.
• Voice agent. Holds a live phone conversation in Uzbek, Russian and English: takes inbound calls, calls back on a lead, reminds about a visit or a payment, runs surveys. This is not an IVR with «press 1». More: voice AI agent.
• Call analytics agent. Doesn't talk to customers; it reviews your managers' calls: transcribes them, scores every call, finds lost deals and shows what to fix. More: call analytics.

In practice these types work together. The sales agent gets a phone number in Direct, the voice agent calls back, and the call analytics agent checks how the manager's conversation went. If the agents live in different services, the customer's context is lost at every hand-off. So when choosing, look less at the number of agents on the price list and more at whether they all work on one CRM card.

How does an AI sales agent work in Instagram and Telegram, step by step

Speed decides more than eloquence here. HBR's study of 2,241 companies found that firms replying to a lead within an hour were nearly 7 times more likely to qualify it, while the average response time was 42 hours. A sales agent closes exactly that gap.

1. The customer writes or comments. At 10:40 pm Tashkent time, a comment saying «narxi?» appears under an Instagram ad. The agent replies publicly by meaning and messages the author in Direct.
2. It understands the request. The customer answers with a voice note, half in Uzbek and half in Russian. The agent transcribes it and pulls out the essentials: product, size, delivery city.
3. It qualifies. It asks a few clarifying questions under your rules: budget, timing, pick-up or delivery. The answers go straight into the customer card.
4. It recommends. It checks the catalogue and prices in soum, then suggests a matching option and a complementary product in your brand's tone.
5. It takes payment. It sends a payment link via Payme, Click, Uzcard or Humo and confirms the order as soon as the payment goes through.
6. It runs the CRM. It opens a deal, sets the pipeline stage and saves the thread. The manager doesn't copy anything by hand.
7. It follows up and hands off. If the customer goes quiet, the agent follows up on your schedule. A non-standard request goes to a manager with the full history.

Instagram also has a technical limit: under Meta's Messaging API rules, a business replies to a customer within 24 hours of their last message. An agent that answers in seconds always fits that window; a manager who reads Direct in the morning doesn't always. The rules are covered in are auto-replies allowed on Instagram.

AI agent, scripted bot or human operator: which should you choose

Each option is strong at its own job. In Salesforce's CRMArena-Pro benchmark, even the best language-model agents succeeded on about 58% of single-turn CRM tasks and about 35% in multi-turn conversations. That's why an agent needs a defined process and a human for hard cases, not unlimited freedom.

The takeaway isn't «an agent instead of everyone». In our implementation experience, the combination works best: the agent takes the volume and the night hours, the bot stays on rigid steps such as code confirmation, and the operator gets only the conversations that need judgement. The operator's job changes too: fewer repeats, more decisions.

How do you choose an AI agent: a checklist for businesses in Uzbekistan

The choice starts with language, not with the model. The state is betting on AI: the strategy approved by Resolution PQ-358 targets 1.5 billion US dollars in AI-based products and services by 2030. But for your Direct inbox another question matters more: will the agent understand a customer who writes in Latin or Cyrillic, switches to Russian mid-sentence and sends a voice note in Uzbek? Check it against the list below using your own real messages.

There are cases where you don't need an agent. If you get five enquiries a week and answer them yourself in a minute, or if sales happen through personal meetings and tenders, response speed decides nothing. There an agent adds cost, not results.

Zukko.AI: a platform of AI agents for business

Zukko.AI is a platform of AI agents for business: AI employees sell in Instagram and Telegram, answer support requests, call customers in a live voice and run the CRM themselves. The platform understands Uzbek, Russian and English, including voice messages, takes payments via Payme, Click, Uzcard and Humo right in the chat, and keeps data of Uzbek customers in Uzbekistan. Developer: SUMMERTECH, Tashkent.

• Sales AI: a sales agent in Instagram Direct and Telegram, from the first message to payment and a deal in the CRM.
• Inbox AI: every channel in one window: Telegram, Instagram, WhatsApp, web chat; Uzbek voice notes and document photos.
• Voice AI: a voice agent in Uzbek, Russian and English on an in-house STT/TTS stack, connected to your telephony over SIP. AI calls are available only on the Enterprise plan: from 15,000,000 soum a month, with 2,200 minutes and 1 AI line included.
• AI automation: a no-code builder for your own AI agents and «trigger, condition, action» scenarios.
• Call analytics: transcription and scoring of calls, up to 300 a month on the Start plan, without that limit from the Business plan.

The product runs in production: AI answers customers in aviation, healthcare, services and sales, and clients include Uzbekistan Airways, where AI answers passengers' text enquiries. Small and mid-sized businesses launch the platform themselves from 2 hours: connect channels, upload the catalogue and rules, switch the agent on. For large companies with their own systems and security requirements, implementation takes up to 30 days.

AI agent and AI employee: what it is, how to launch →