The tension: An AI agent is only as good as its setup, and setup is where most agents fail. There's a lot to configure: how it talks, what it knows, which channels it covers, how it routes, the details of the business. Ask a non-technical owner to do all of that and you get a half-built agent. That's exactly the useless bot their customers will try to escape.

The obvious move: The usual approach is a behaviour builder, where you draw a flowchart of what the agent should do. That's fine for a technical team. For the small business owner Ventaz is built for, it's a wall. It feels like programming, and most people give up before they get a working agent. I didn't want to make the flowchart easier. I wanted to remove it altogether.

The decision: Don't ask people to configure an agent. Build it for them and show it to them. You type in your website address. Ventaz reads your site, builds an agent from it, and fills in most of the settings. Then it shows you that agent, already in your branding, answering questions about your business, live in a preview, before you've signed up for anything.

Why it works: The preview comes before the signup, not after. That order is the whole point. People commit to things they've already seen working, not things they've been promised. Seeing your own branded agent answer a real question about your business changes the feeling from "now go configure this" to "this already exists, I just have to keep it".

Setup stops being a flowchart and becomes a moment of "oh, it already works."

The tension: People reach for voice over typing. It's faster and more natural, especially on a phone or when they're busy. But voice is also where trust is hardest. Only about a third of people are comfortable with an AI agent on a phone call, against nearly nine in ten on web chat. And the usual way a voice bot fails is the worst one: it mishears you, you repeat yourself, and you give up. An agent that only does text pushes voice-first people away. An agent that does voice badly pushes them away harder.

The obvious move: Stay text-only because it's simpler to build, or add voice as a separate mode you have to switch into on purpose. Both treat voice as second-class, and both ignore how people actually want to talk. There was a third bad option: ship voice as a closed loop, where a misheard word means starting over.

The decision: Add voice as a first-class channel, with a realtime transcript and smooth switching between text and voice in the middle of a conversation. Someone can start by typing and finish by talking, or the reverse, without losing the thread.

Why it works: The live transcript means nothing is lost when someone speaks, and you can see straight away if the agent heard you right. If it didn't, you switch to text and keep going, mid-conversation, with no reset. That's the safety valve. Voice gets to be the natural, fast option because failing on voice never costs you the conversation. The benefit: people talk to it the way that comes naturally, and when voice gets it wrong they don't have to repeat themselves or hang up.

Evidence: The speech-to-text is a general model today. Local accents are where it's weakest, which is one more reason the text switch has to be there.

The agent meets people the way they'd rather communicate, without making them choose up front.

The tension: People don't reach a business in one place. They message on WhatsApp, Facebook, Instagram and Messenger, they use the website widget, they call. Owners and their customers kept telling us the same thing: questions come from everywhere, and an agent that only lives in the website widget misses most of them. Answering on every channel was the easy part. The hard part is memory. When the same person shows up on two channels, should the agent treat them as someone it already knows?

The obvious move: The tempting version is to merge everything that looks like the same person, so the agent always seems to remember you. It looks great in a demo. But merge two people by mistake and you've shown one customer's history to another. And a loose match is exactly the kind of gap someone could use to pull data that isn't theirs. For a product that's all about trust, that's the one mistake we couldn't afford.

The decision: Answer on every channel, and manage it all from one place. But only merge a person's history when the agent can confidently verify it's the same person: the exact same name with the same phone number, or the exact same name with the same email, backed by similarities in the conversations themselves. When it can't be sure, it doesn't merge. Today that combined history is shown to operators. We're extending it carefully, so the agent itself can use that context, and so no past data is ever pulled up without proper verification, like an OTP. That part is a work in progress on purpose. A careless merge is the fastest way to lose the trust the whole product is built on, so we'd rather get it right than rush it.

Why it works: Once identity is verified, an operator can open a Conversation History that brings that person's conversations together across channels. A lead panel next to it holds what the agent has gathered: company, contact details, and what they actually need. Whoever picks it up gets the full picture instead of starting cold. The benefit: one continuous relationship, without betting a customer's privacy on a guess.

Evidence: We tested multi-channel answering on ourselves and with business owners, and it holds up across channels. The combined memory we're still perfecting, on purpose. A careless merge is a trust problem, not just a bug. And the caution matches what customers already expect. About half of people are uncomfortable sharing personal or financial details with an AI agent, two in three are uneasy about an agent holding their full history, and nearly three in four expect a verification step like an OTP before an agent acts on their behalf. Slowing the merge down isn't us being timid. It's the product meeting people where their trust actually is.

One agent, every channel, one place to manage it.

Ventaz is a more focused product than Craftit, so I gave it a more focused system. Rather than build a full bespoke design system, I kept it on a customised Tailwind base with the custom components it actually needed.

That was a deliberate call. Part of doing this well is knowing when a product doesn't need more system than it already has.

Building it was a two-person effort with my co-founder, and I was hands-on in the front-end code for a good part of it, directing Claude Code and Cursor against the design system rather than treating them as a black box. The parts that depend most on implementation, like the live branded preview and the real-time conversation view, I built myself, so their timing and live states never got lost in a handoff. On a team of two, that's how design and a fast-moving codebase stayed in step.

Ventaz is still you, growing through it’s early stages with continues discovery. But we already see positive signals.

300+ booked leads and appointmentsdriven for customers, across1.8K+ conversations.Every prospect gets a working, branded agent to try before they commit.