I built a Telegram AI tool for myself. Users immediately wanted different things
Today I received valuable feedback from the first users testing CoTel.
When I started building this service, I was mostly solving my own problems and thinking from my own experience of using Telegram. I wanted a tool that could help me analyze chats, answer questions about message history, and find important information faster.
Of course, I thought this could also be useful to other people. But until you actually put a product in front of real users, you still see it mostly through your own context.
And today I felt that very clearly.
One of the first things I heard was:
“It would be really useful to search not only channel posts, but also comments under posts.”
And I realized I had never even thought about that use case myself. So I never implemented that functionality.
But the moment someone said it out loud, it immediately made sense to me why that could be extremely valuable for many users.
I also received very interesting feedback from a journalist who works heavily with Telegram.
He said he really liked the idea of the product and would genuinely use a tool like this in his workflow. But almost immediately, he pointed out things I also hadn’t thought about.
For example:
— he needs direct links to the exact Telegram messages found through AI;
— manually searching for those messages afterward is inconvenient;
— it would be much more useful for him to work with entire groups of chats instead of a single chat.
That part was especially interesting to me.
Because for him, Telegram is no longer just a messenger — it’s a research environment and an information source. He has dozens of topic-based chats organized into folders, and his natural workflow is asking questions across an entire group of sources at once.
Meanwhile, I hadn’t even implemented Telegram folders in the interface because I personally barely use them.
Now I’m already thinking about how to build that functionality.
I think this is one of the most interesting moments in building a product:
at some point, you stop building a tool only “for yourself” and start seeing how differently other people approach their own workflows and problems.
That’s why early feedback feels so important.
The most encouraging part is realizing that the service may actually be genuinely useful to people. Some users already told me they would actively use it — and even pay for it — if the features most important to them appear in the product.
Right now I’m continuing to work on the landing page, improving UX, and collecting more feedback.
Some ideas will go into the second iteration of development, because at some point it’s important to stop adding features and actually release the MVP instead of disappearing into endless development forever.
The 'users immediately wanted different things' moment is the most useful data point in early product life — way more signal than feature requests in a survey. The trick is figuring out which divergent ask is a real new segment vs. just one loud user. Did you see any clustering in the asks, or was it all over the map?
This is a strong signal that CoTel is not just a Telegram search tool. The journalist use case points to something bigger: Telegram as a research layer, where people need to search across channels, comments, folders, and source groups with exact message-level citations.
That feels much sharper than “AI chat history search.” The real pain is workflow continuity. If someone uses Telegram as their knowledge base, they need answers that link back to the original message, not just summaries that create more manual search work.
One thing I’d watch is the CoTel name. It is short, but it may not immediately carry the research/workflow intelligence angle. If this becomes a broader AI layer for Telegram-heavy researchers, journalists, analysts, and operators, Xevoa .com would feel more expandable and platform-grade.
This is one of the hardest transitions in product building:
realizing users are not trying to reproduce your exact workflow, they’re solving their own.
The journalist example is especially interesting because it changes the positioning from:
“AI assistant for Telegram chats”
to:
“AI research layer for Telegram knowledge networks.”
The requests also make total sense:
Feels like the product becomes much more powerful once the workflow expands beyond personal chat search into structured research workflows.
Also agree with your MVP point — there’s always tension between:
“one more feature”
vs
“ship before it becomes endless internal development.”
This is the part of building that humbled me too 😅
You think people will use the product like you do, then one user shows a completely different workflow you never imagined.
This is the core tension of building for yourself vs. building for a market. We ran into the exact same wall built an AI research tool that solved our own pain (manual competitor analysis), then realized users were arriving with completely different expectations. How different were their requests from what you'd built?
This is such a real product-building moment.
One of the hardest transitions is realizing:
users are not buying your exact workflow — they’re trying to solve their own.
And often the most valuable use cases come from people using the product in environments you never originally imagined.
The journalist example is especially interesting because it changes the positioning completely:
from “AI assistant for Telegram chats”
to potentially:
“AI research layer for Telegram knowledge networks.”
That’s a much bigger and more powerful category.
The requests also make total sense:
• direct deep-links to messages → critical for trust & verification
• comments indexing → where a lot of real signal actually lives
• folder/group-wide querying → transforms it into a research workflow instead of chat search
The strongest insight in your post is probably this:
“unsupported assumptions from the founder become obvious only after real usage.”
That’s why releasing early matters so much.
Also completely agree with your MVP point. There’s always tension between:
“just one more feature”
and
“ship before the product becomes an endless internal project.”
From an AI product perspective, explainability + source traceability will probably become huge advantages here, especially for researchers, journalists, analysts, and investigators.
Emmanuel here 👋
I work heavily with AI workflows, developer tooling, automation, and product systems, and I’d genuinely be interested in testing CoTel and providing detailed UX / workflow feedback from a real usage perspective.
Very promising direction.
https://teams.live.com/l/invite/FAAk3iOSJkDyS11JQE?v=g1