Armin Daryabegi

founder of Future of AI customer service & marketing for E-Commerce

[Armin Daryabegi @saasjesus]

We grew from €3M ARR to €6M ARR last year.

While killing competitors who have €100M funding.

Currently just a UX Design Engineer 🐐 is needed

If you are interested: https://t.co/9w6UNoJzN9

Let's see if anyone is as cracked as we are

[Armin Daryabegi @saasjesus]

@dwhitedesign @typesafeai Bro Daniel, open your DMs 😅

[Armin Daryabegi @saasjesus]

Feature request came in last night. Live the next morning.

We ship almost daily. No cutting corners on quality or stability.

One of our customer's support agents was on vacation for two weeks.

First day back, she messaged us. Said our AI sounds way more natural now.

Good startup dev teams need 26 to 78 hours just from first commit to live.

That doesn't even include the wait between a customer request and someone actually starting on it.

For us, it's straight to build, test, ship. New features in a few days. Smaller changes often within hours.

While others are still scheduling and reviewing the first change, ours is already live, and the next one's already in progress.

And our speed compounds.

2 extra improvements a week add up to 100+ a year.

So our features aren't just live sooner. We're building noticeably more product in the same amount of time.

We deploy almost every day now, without stability taking a hit. 100+ days, zero outages.

Same pace as before. If not faster.

Which means more, in the same time: better performance, stronger agents, simpler workflows for our customers' support teams.

[Armin Daryabegi @saasjesus]

We've paid @vercel close to €1,000,000.00. Why put that much into a single vendor?

Vercel is the infrastructure Chatarmin runs on. Every deployment, all logs, AI Agents and more.

Their developer experience makes us faster and our whole architecture AI ready.

One vendor. All data, all logs, the entire infrastructure in one place.

Massive leverage for us and our coding agents.

Elon Musk said something similar about Tesla once: the real innovation wasn't the car, it was the machine that builds the car. The factory.

@vercel makes the factory we build chatarmin with even better.

A little while back we shipped our biggest product shift yet, AI Agents for every customer. And we recently became Vercel's first Enterprise Partner in Vienna. The partnership came with roughly 1M in credits.

With armincx we're applying the same principle to customer communication:

support, marketing, every channel, every interaction, one tool.

All in one. Full cycle.

So brands always know exactly what's happening between them and their customers.

Working with Florian Binswanger at @vercel, we're also bringing their hackathons and events to Vienna. They already run those in plenty of cities. Just not here yet.

We're building that together now. More top talent to Vienna, soon.

Still day zero.

#vercel #chatarmin #startup #aiagents #armincx

[Armin Daryabegi @saasjesus]

Our dev Aviral analyzed 1,000,000 support tickets from 100+ e-com brands. By hand. Pen and paper.

He went ticket by ticket. 36 hours straight.

Bedding. Drinks. Auto parts. Food.

Completely different products. Same problems underneath.

We tried clustering first.

AGNES, DBSCAN, the usual suspects. At this volume, the language is too messy.

"Product damaged" can mean a warehouse error, carrier damage, or just bad quality that shows up after use. An algorithm throws all of that into one bucket.

So Avi did it by hand. 100k+ topics filtered, printed, read one by one, brand by brand.

Cure a disease, become a doctor.

Build a car, become a mechanic.

Categorize every ticket, see every topic yourself first.

The result: a game changer for our armincx customers.

- See what's actually behind their tickets, without opening each one
- Spot a sudden spike, missing deliveries, leaking packaging, immediately
- See exactly what AI already resolves for them, and what should get automated next

Same on our end. We see which topics are becoming relevant across our customers, and build or sharpen for them.

Voice & another dashboard update are next. 🤫

Coming soon.

#armincx #cx #ecom

[Armin Daryabegi @saasjesus]

Two weeks ago, Stephan Wassink and I set ourselves a goal: Increase Zelesta’s automation rate by more than 50% in just 4 days.

A few days later: From 48% to 74%.

Mission accomplished.

Stephan joined us in Vienna for those 4 days.

Together, we reviewed the cases where our AI agents were still handing over to a human.

Stephan knows exactly how his team handles complex requests in practice.

We saw where our agents were still missing context or the right logic.

Then we repeated the same loop:

Real customer case. Feedback. Adjustment. Test.

We improved the agents, covered new scenarios, and tested every change directly on real tickets.

The result: Almost 3 out of 4 actions in Zelesta’s customer service are now automated by AI.

This means less manual work for Zelesta’s support team. At the same time, the improvements we built benefit our other customers as well.

The days with Stephan have now become the blueprint for our CX AI Agents Bootcamp in September.

Customers will join us in Vienna to build their CX agents together with our team or take their existing setup to the next level.

Because we learn the most from our customers.

They show us what to solve next and help decide how our software evolves.

Every day, one step closer to a 100% automation rate.

#armincx #aiagents #customerservice #ecommerce

[Armin Daryabegi @saasjesus]

Customer goes live with armincx on Thursday.

Automation rate near 50% by Friday.

No weeks of onboarding.

No integration phase where nothing happens.

How a setup actually runs:

1. Migrate or start fresh

Full ticket history imported. Inbox rebuilt with the same views, so nothing changes for the team day to day. Then everything gets sorted by topic to find the biggest lever.

2. Deep dive

Go through tickets one by one. Not just categorize, understand. Where do duplicate follow-ups come from. Where does the wrong escalation happen. Every escalation has a reason behind it, missing info, unclear case, whatever it is. That's what the roadmap gets built on.

3. Tech stack

Shop, shipping, payment provider, everything an answer needs to be based on fact instead of a guess. Whatever's missing gets connected or built before anything else happens.

4. Build the agents

Usually the 5 biggest topics. Order status, social, warranty, returns, cancellations. One agent per topic, clean logic, so it stays maintainable once more get added.

5. Onboard the support team

Before anything goes live, the client's team gets briefed. What the agent owns, what stays human, how to read a ticket at a glance.

6. Approval mode

2-3 days with a human checking every reply before it sends. Once that's clean, full autonomy.

We don't just sell software and walk away.

Work with us, and you get the right people.

100+ setups deep, even the ugliest tech stacks aren't new to us anymore.

Our team understands the customer's process and build the agents so they run from day one.

Automating tickets from minute one.

Taking load off the support team.

Reducing costs.

Hard to find that combination of speed and depth anywhere else.

Kudos to our success team. You're doing a fantastic job.

#aiagents #cx

[Armin Daryabegi @saasjesus]

A mid-size e-com brand gets 1,000+ support tickets a month.

Automating them is one thing. Bringing that number down is another.

At that volume, most teams lose sight of where the actual problems are. And scale keeps adding more tickets.

The default reaction: automate harder. Doesn't answer why the tickets exist in the first place.

Example: if 50% of the last 2,000 deliveries to a region arrive damaged, faster automated refunds aren't the fix.

Switching carriers or fixing packaging is.

So we built Dashboard 2.0.

Every ticket gets categorized by root cause. Brands see what's driving next month's ticket volume, not just this month's queue.

Two sides of the same system:

- AI agents resolve the tickets coming in
- Dashboard shows what to fix so fewer come in next time

Inbox, dashboard, agent: all fully customizable.

Pull in your own data, prioritize your own insights, configure the agent around your own workflow.

Building armincx with no limitations. The product adapts to the business, not the other way around.

#armincx #dashboard #productupdate #chatarmin #cx

[Armin Daryabegi @saasjesus]

100% uptime for 3+ months. And we've gained speed on top of it.

A year ago, that pace would've meant way more bugs.

Here's how speed stopped costing us stability:

When we started @chatarmin and armincx, we never really had time to stress test anything.

We just went for it. Sell, then build.

Huge customers from day one. Before we even knew if the system would hold.

It showed. Stuff broke, more than once.

So we learned. Brought in outside help.

Daniel, for example. Spent 5 months with us, just on the database. We learned a lot from him.

Together with him, Jerry, and the whole team, we changed quite a bit:

→ End-to-end tests for every critical function
→ Playwright frontend tests for performance, even on the oldest devices
→ 300+ tests for our AI
→ 10+ CI checks on every PR
→ Automated code reviews for our engineering standards
→ Database performance checks on every commit
→ No deploys during the day

Result: 100% uptime for 90+ days.

Without losing speed. If anything, we're faster now.

Because one thing hasn't changed: we can never get slow.

That's the moment others kill you.

#cto #chatarmin #softwaredevelopment #armincx #saas

[Armin Daryabegi @saasjesus]

We launched our AI agents in July. The idea existed 1.5 years ago.

Why now, not then?

Feb 2025, we started bringing AI into customer service.

And we had to answer one question:

Build agents? Or ship static workflows?

We went with workflows.

Models couldn't reliably run 20 steps in a row back then. An agent would've gotten stuck mid-flow. Building it properly would've taken months and a ton of engineering we didn't have. We also thought customers wanted to see and check what's happening, not just trust a black box.

Workflows were what actually helped our customers at the time.

That's flipped now. Models can do what they couldn't a year ago.

So we're building agents.

Some customers are already automating 50-70% of tickets. Real resources, real money saved.

Sure, you could say I took the wrong exit a year ago. Missed the shortcut.

I don't regret it. I only have today's learnings because I can look back at the decision now.

More interesting question: how many right turns have I already taken in the last 4 years, without noticing?

#armincx #agents #chatarmin #cx #supporttool #aiagents

[Armin Daryabegi @saasjesus]

Some of our customers have more say in our product roadmap than my own co-founder & me as CTO.

Stephan Wassink is one of them.

He moved Zelesta from Trengo to us over a year ago.

✓ 500.000+ tickets fully automated
✓ response time: days → hours
✓ automation: 50% → 80%, this week

He knows ecom inside out. And he's an AI power user himself.

He tells us exactly what's broken. We watch him work, live.

I blocked my whole week to shadow him and spar on the product.

He pushes us in the right direction. We ship him the features to match.

Same goal, different seats: 100% automation with a 5/5 CSAT score.

Happy to welcome you to Vienna, Stephan!

We love building with you and shaping the future of customer service together.

Still #dayzero

#zelesta #armincx #chatarmin #clientstory #customersservice

[Armin Daryabegi @saasjesus]

Saturday. Support's offline.

A customer wants to cancel 2 orders.

The AI agent verifies her identity and triggers the refund. No human touches it.

Minutes later, she asks for a return label on those same orders.

The agent knows there isn't one. They're already cancelled.

Then it catches what she actually meant: a different order, one she never gave a number for.

Finds it in the conversation history. Loops in the return partner. Closes it out.

Team's still offline.

One of our customers now runs 45k+ tickets/month through this.

One of our customers now runs 45k+ tickets a month through this.

Automation rate went from 27% to over 45% in two weeks.

Over 1,000 requests a day, and weekend response time is still 2 minutes. Some customers even add a random 5-15 minute delay so it feels more human.

Only tickets that actually need a human reach the team.

Live on every armincx account as of this week.

And it's just the beginning!

#armincx #customersupport #aiagents

[Armin Daryabegi @saasjesus]

@dwhitedesign @dwhitedesign want to open DMs? think got something valuable for you

[Armin Daryabegi @saasjesus]

@jacobmparis doesn't look too bad at all

[Armin Daryabegi @saasjesus]

@DanielWhit21874 yeah letsgo, want to work together? open DMs
https://t.co/9w6UNoJzN9

[Armin Daryabegi @saasjesus]

@markproduct https://t.co/MIUmnNxU6M

[Armin Daryabegi @saasjesus]

@iamncdai @shadcn absolutely amazing. Chanh I was wondering if you could open your DMs for me 😅

[Armin Daryabegi @saasjesus]

We are looking for an UX Designer for @chatarmin and armincx!

Are you eager to take end-to-end ownership of the entire UI/UX for both of our products?

At Chatarmin we compete against tools that have raised $100M+. 

Bigger design teams, more budget, more everything.

What we put against that is speed and standards. 

We are looking for someone who notices the small things, can explain why they matter, and turns that judgment into production React.

Someone who cares about taste, speed, clarity, motion, spacing, and the small details users never consciously notice but always feel.

What we are looking for is 3+ years across frontend development, UX, and product interface design, with strong React, TypeScript, CSS, and Tailwind experience. 

And you have a portfolio with real product work, not just clean Dribbble shots.

We are a young team, no ego, fast pace. 

Good ideas should ship, not die in review loops. 

You get direct ownership of the product experience and real impact on a company that is early enough for your work to matter and ambitious enough for it to scale.

If that sounds like you, apply here:

https://t.co/HQMnluxism

#uxdesigner #chatarmin #startup #hiring #uxdedesign

[Armin Daryabegi @saasjesus]

@uixmat @tacosta @benjitaylor @iamdavidhill @ryolu_ @DanHollick is there some design engineer community haha? We are looking desperate for cracked ones haha

in case someone sees this https://t.co/9w6UNoJzN9

[Armin Daryabegi @saasjesus]

@DanielWhit21874 Nice one! Wdyt @DanielWhit21874 down to join us crackheads?

https://t.co/9w6UNoJzN9