Typewise AI Customer Service

Automate customer support across systems with AI agents

Productivity

Customer Communication

Artificial Intelligence

Visit WebsiteSee on Product HuntTwitter ⧉

Proud of the team for getting this out. The simulator does both: historical replay against new instruction versions for regression coverage, and synthetic generation for the long tail.
We argued on the engineering side about which to build first, and eventually decided neither one on its own makes the evaluation loop useful rather than theatrical. Both landing cleanly is the part I'm most invested in.

Response-quality grading on its own never catches the interesting failures. Action-sequence validation against an expected workflow, invariants on which tools get called for a given intent, custom policies beyond simple output checks; that's where the real agent bugs live.

Getting that into the harness as a proper API rather than a checkbox was the thing we kept pushing for on the QA side.

Turning existing docs into live automations through natural language is the future of complex tools. Flowchart builders are too cumbersome; users don’t want to wrestle with messy diagrams, they want to write in plain language and let the system handle the rest. I've spent a lot of time designing Typewise to remove this friction, and the team and I have obsessed over every detail along the way. I’m proud to see it finally out in the wild, and I’m really looking forward to hearing what you all think!

A persistent conversation object with channel-agnostic events turned out to be the right abstraction for cross-channel context, rather than a thread-per-channel stitched at read time. Took a few iterations to land on, and most of that work isn't visible from the outside, which is exactly the bar we set on the engineering side. A customer moving from WhatsApp to email shouldn't have to notice the platform at all.

I’m really proud to have been working on this as an AI engineer. I’ve been dogfooding it at my side company and it's become the thing I'd miss most if it disappeared. It’s saved me so much time responding to customer emails. Tickets get solved automatically with relevant context and tools, but also there’s a human in the loop element for tickets the AI can’t fully solve.

What about initiating refunds, e.g. we need to make them via Postman or in the Google Play Order management. Can it handle? Operate on cross-platforms?

Could be very useful! Honestly, I am so frustrated with some chatbots at many companies that provide no resolution at all. One question though if users don’t define flows then where does control actually live like in prompts, policies or learned behavior??

One thing I've learned working on agents in production: what kills you isn't bugs, it's drift. Three weeks in, CSAT dips a few points and nobody on the team can actually tell you what changed.

That's why on the AI side we pushed hard to have simulation and evaluation as first-class primitives, not bolted-on features. It's the piece of this release I'm most confident about!

Going MCP-native from day one saved us from the usual “we’ll add that integration next quarter” loop. And with 6000+ connectors most customer requests are just a quick config change instead of something we have to slot into the roadmap.
From the engineering side, that’s what made this launch feel steady instead of a sprint we’d regret later.

Hey Product Hunt 👋
I'm Janis, CTO at Typewise, and it honestly feels a bit surreal to finally put this out into the world.

One of the things I'm most proud of is the UX. We didn't want another tool that requires a week of onboarding and a 40-page manual. Setting up Typewise should feel closer to onboarding a new team member than configuring enterprise software, you tell it what it should and shouldn't do, point it at the systems it needs, and it gets to work.

The hard part was making that feel effortless without cutting corners on safety. We spent a huge amount of time on the layers underneath: what the AI is allowed to know, what it's allowed to do, where a human needs to stay in the loop. That interplay between AI and human was the real design challenge for us, not the model itself.

I genuinely believe customer service is one of the clearest cases where AI can take the grind out of the job without replacing the people who care about it. And seeing it actually land with real teams is the part I still can't quite get used to.

After such a long build, I'm really curious to hear what you think of it. Honest first impressions, what stands out, what feels off, all of it very welcome.

Hey PH! I'm David, co-founder of Typewise. Here's a hot take: most "AI customer service" tools are just glorified chatbots with better marketing. They still need you to build flows, write rules, and babysit them.

We built Typewise to be fundamentally different. It's an AI agent system that builds your AI customer service for you. Describe your goals in natural language, and our platform creates specialist agents, connects to your systems, and starts resolving tickets autonomously.

But it also knows its limits. The AI supervisor detects when a human should step in and hands off seamlessly with full context. Your team picks up in a clean, beautiful UI that makes customer service actually enjoyable.

No flowcharts. No code. No manual tuning. The AI manages itself.

We're already live with enterprises and YC startups across Europe & US, and the early results speak for themselves. Excited to share this with the PH community. Would love to hear what you think!