What Is a Deployed AI Agent? (And How Is It Different From a Chatbot?)
A deployed AI agent is an AI model running continuously on real infrastructure — with permanent memory, tool access, and permission to act — instead of waiting for you to open a browser tab. A chatbot answers questions when you show up. A deployed agent holds a job whether you show up or not.
The word "agent" gets slapped on everything right now, from prompt templates to customer-service widgets. Most of what's sold as an agent is a chatbot with better marketing. This guide draws the line precisely, so when you evaluate anything — including MAKO — you know exactly what you're buying.
What does "deployed" actually mean?
Deployed means the agent exists as a persistent process on cloud infrastructure, not a session in your browser. Three things follow from that:
- It's always on. The agent doesn't stop existing when you close a tab. It's reachable at 3am from your phone the same way it's reachable at 3pm from your desk.
- It has its own inbox. A deployed agent receives work through a real channel — in MAKO's case, your Telegram account. You message it like a contact: text, voice notes, photos, PDFs.
- It keeps state. Because the process persists, memory can persist. MAKO stores every conversation, file, and decision in a PostgreSQL brain with semantic search, so context accumulates across weeks and months instead of evaporating per session.
A chatbot has none of those properties. It's request-and-response: you type, it replies, the session dies, and the next session starts from zero.
How is a deployed agent different from a chatbot?
| Chatbot | Deployed agent | |
|---|---|---|
| Exists | While your tab is open | 24/7, as a running process |
| Memory | Per session, mostly forgotten | Permanent, searchable, cumulative |
| Initiative | Answers when asked | Runs scheduled tasks and queued work on its own |
| Tools | Usually none | File access, code execution, integrations, APIs |
| Interface | A website you visit | A channel you already live in (e.g. Telegram) |
| Output | Text answers | Completed work |
The last row is the one that matters. A chatbot's product is an answer. An agent's product is a finished task — the report written, the file analyzed, the research done and waiting in your messages when you wake up.
What can a deployed agent do that a chat window can't?
Concretely, using MAKO's feature set as the reference implementation:
- Scheduled tasks. "Send me a weekly report every Monday at 9am." Said once, runs forever.
- Task queues. Drop five tasks before bed; the agent grinds through them autonomously and reports back with results.
- Files as inputs. PDFs, contracts, spreadsheets, images, code — 30+ file types dropped straight into the chat.
- Voice in and out. Send a voice note; it's transcribed and acted on. Toggle voice mode and the agent replies with voice messages.
- Self-extension. Tell it to connect to GitHub, Notion, or your CRM, hand it an API key, and it wires the integration itself.
None of that is possible in a session-based chat window, no matter how smart the underlying model is, because all of it requires a process that outlives the conversation.
What infrastructure does a deployed agent need?
Four components, and the quality of each determines whether the agent is an employee or a toy:
- Compute that never sleeps. Managed cloud infrastructure running the agent process continuously.
- A memory store. Not a context window — an actual database. MAKO uses PostgreSQL with semantic search.
- An interface you already use. Adoption dies when the agent lives in yet another app. Telegram means the agent is in your pocket by default.
- Security gating. A deployed agent holds your context, so access control is non-negotiable. MAKO is owner-only — gated to your Telegram ID — with credentials scrubbed before the AI ever sees them and output auto-filtered for accidental leaks.
Why does the model behind the agent matter?
An agent is only as good as its brain. Plenty of "agents" run cheap, small models because inference costs less — and consumer chat products quietly delegate work to lighter models behind the scenes. That's fine for casual questions. It's not fine for an agent you hand real business work to, because a weaker model makes subtly wrong decisions autonomously — the most expensive kind of wrong.
MAKO runs Claude Opus, Anthropic's most capable model, exclusively — the same Claude Code toolchain professional engineers use, with full tool use and autonomous execution. It never downgrades to a smaller model. And it uses your existing Claude Pro or Max subscription via OAuth, so there's no API key and no metered pay-as-you-go billing.
If you want to see what a full agent framework looks like under the hood, the engine that powers MAKO is documented at FAST Framework.
FAQ
Is a deployed AI agent the same thing as a Telegram bot?
No. A Telegram bot is just an inbox and outbox. A deployed agent is the intelligence behind it — a frontier model running on cloud infrastructure with memory, tools, and autonomy. In MAKO's case, Claude Opus runs on managed cloud infrastructure and uses the Telegram Bot API purely as its interface.
Do I need to know how to code to use a deployed AI agent?
Not with a done-for-you deployment. MAKO's setup is two commands pasted into your Terminal plus three tokens — about 10 minutes. If you can copy and paste and follow instructions, you can deploy an agent.
How is a deployed agent different from just using ChatGPT or Claude in a browser?
A browser session ends when you close the tab, and the model forgets you between sessions. A deployed agent runs continuously, keeps permanent memory, executes scheduled tasks, works through a task queue while you sleep, and connects itself to your other tools.
What does it cost to run a deployed AI agent?
MAKO's managed deployment is $297/month (or $2,997/year), plus your existing Claude subscription of $20–$200/month paid directly to Anthropic. No API key and no pay-as-you-go metering.