Production-ready AI agents for regulated enterprises.
Edge
Joined December 2025
- Tweets84
- Following1
- Followers21
- Likes206
A real thing that happens when you build voice AI:
You test it in normal conditions.
Then you test it with background noise.
Then someone switches language.
Then the live demo environment behaves nothing like your desk setup.
Voice AI is not built only in clean rooms.
It is built through messy tests, edge cases, accents, interruptions, noisy environments, and multilingual conversations.
Every one of those moments makes the product better.
A small feature with a big operational impact:
In-call forms.
During a conversation, Elba can collect structured information like:
license plates
reference numbers
IDs
appointment details
case information
Without breaking the flow of the call.
Because the goal is not just to “have a conversation.”
The goal is to turn the conversation into usable work.
Self-serve and enterprise are not opposite motions.
Enterprise buyers still want to explore without booking a call.
Growth teams still want control without waiting for support.
Admins still want setup to be obvious.
Operators still want guardrails.
The best enterprise products feel self-serve until the moment you need enterprise control.
The model is not the product.
The recovery system around the model is the product.
What happens when input is wrong?
What happens when the user gets stuck?
What happens when a required field is missing?
What happens when escalation is needed?
That is where production AI actually lives.
Global Assistance, part of Vienna Insurance Group, runs 24/7 emergency assistance across the Czech Republic with a fleet of about 700 rescue vehicles.
Elba now supports inbound emergency calls, dispatch workflows, paperwork, and follow-ups across languages.
Over 70% of these tasks are handled end-to-end without a human touching them.
That is what we mean by AI workforce.
Not a chatbot.
Not a phone assistant that creates more work.
A system that helps complete the job.
One thing teams ask after deploying AI agents:
“How do we know if the conversations are actually good?”
That is why Elba includes AI-powered call evaluation.
It helps teams review quality, accuracy, and outcomes across interactions, not just a random sample.
Because automation without visibility is not enough.
You need to know what happened.
Conference chaos is real.
Too many booths.
Too many sessions.
Too many “we should catch up later” moments.
So for DMEA, we built a version of Elba that people could actually call for help navigating the event.
A small example of what we believe:
AI should make human connection easier, not replace it.
AI agents won’t become useful just because they are smarter.
They become useful when customers can reach them where they already are.
That’s why we’re working on WhatsApp connectivity for Kolsetu agents.
Create an agent -> connect WhatsApp -> let users chat/call without changing behavior.
Distribution is UX.
During our DMEA launch prep, one piece of feedback changed the page:
“The phone number is the whole point.”
Simple, but correct.
If the experience is “Call Elba,” the page should not make people search for how to call.
So we moved the number above the fold, made it tappable on mobile, and built the page around the action.
Good product marketing often comes down to this:
Make the next step obvious.
2/ The voice agent was handling roadside assistance calls in production.
The hard part wasn’t making it “sound human.”
The hard part was all the boring, high-stakes messiness.