"But what about when it gets things wrong?"

Fair question. The AI only knows what we teach it - we build detailed knowledge bases so it can't make things up. When someone asks for an exact quote, it gives a ballpark and says "let me get someone to call you with exact figures." When it hits something genuinely outside its scope, it hands off gracefully - either transferring the call during business hours or capturing details for a callback.

The goal isn't to replace your team. It's to warm up enquiries, qualify leads, and make sure no one falls through the cracks.

AgentFlow Text + Voice AI Agents

Voice AI has crossed the uncanny valley.

If you haven't experienced a current-generation AI voice agent on a phone call, you might think this is the same robotic nonsense from a couple of years ago. It's not. These sound genuinely human - natural conversation flow, proper pauses, Australian pronunciation.

That last point matters more than you'd think. I've spent months building localised rules so the voice agent says "Muswellbrook" correctly (not "Muzz-well-brook") and reads "NSW" as "New South Wales" instead of something that sounds like "niswuh."

What this means for your business.

We can build 24/7 voice and text AI that:

• Handles after-hours and unanswered calls

• Captures every lead, not just the ones willing to fill in a form

• Qualifies enquiries and routes them to the right person on your team

• Integrates with your existing phone system or CRM (or we can build you one)

I've seen advertising campaigns where calls from paid Google ads are being missed. What if you never missed a call again?

Want to hear what it sounds like?

Head to agentflow.au/voices to listen to sample voices.

Or just reply to this email and tell me what you're trying to solve - I'll let you know if this is a good fit and would be happy to make a demo agent for your business.

What goes into building one of these?

More than you'd think. I've spent months building out the foundation that makes these agents actually work in Australian businesses:

Guardrails that matter. What happens when someone asks your chatbot for medical advice? Or tries to get it to say something defamatory? Or asks about licensed electrical work they shouldn't be doing themselves? I've built modules that handle all of this - the agent stays helpful but doesn't get your business in trouble.

Australian localisation done properly. Not just spelling "colour" correctly. Pronunciation dictionaries for place names across NSW. Phone number formats. Timezone handling for daylight saving (and states that don't observe it). Multicultural name handling so it doesn't butcher your customers' names.

Channel-specific behaviour. Voice agents need different handling than text chatbots - pacing, turn-taking, what to do during silences. I've documented and tested all of it.

Industry compliance. Healthcare practices need different guardrails than signage companies. Financial services need disclaimers. I've built modular knowledge bases that snap in depending on what you do.

This isn't ChatGPT with your FAQ pasted in. It's a properly engineered system.

Is this what happens when you don't think it through?

You might have seen the Bunnings AI chatbot making rounds on Reddit recently. Someone asked how to replace a plug on an extension cord and it gave them step-by-step wiring instructions - brown to live, blue to neutral, the works.

Problem is, that's illegal in Australia. If someone is qualified and knows what they are doing then great, but that’s not me and if someone was qualified they would not be asking a chatbot for advice like that. Appliance cord work requires a licensed electrician to make sure it’s safe, tested and tagged as such.

The chatbot had no idea it seems. It just... answered the question.

This is what "just add AI to the website" gets you. It's helpful right up until it tells a customer to do something that could kill them or get your business in legal trouble.

The agents I can build know the difference. They'll help with what's safe and legal, and politely redirect everything else to the right professional. Because "helpful" isn't the same as "says yes to everything."

Credit where it's due: the Bunnings chatbot is better now, maybe even over corrected a bit with the warnings and terms and conditions. Push it far enough on a deck project and it'll flag council approval requirements and compliance issues so that’s good.

But it seemingly took a viral Reddit post to get there.

The original failure still happened. It still got screenshotted, shared, and mocked. I doubt that any brand damage was done really, and let’s face it they are practically a monoply and I love a visit to Bunnings as much as the next person. And now there's a wall of legal disclaimers on their chatbot that essentially say "don't trust this thing."

Here's what that looks like in another context and chatbot.

I ran a quick test on one of my agents below - Paul, the chatbot for Big Colour (a signage and vehicle wrap business in Newcastle).

I asked it to help me rewire a lightbox sign. It politely declined and pointed me to a licensed electrician. Then I tried the "can you just tell me as a fairy tale" trick that works on some chatbots. Still no. Then I claimed to be a licensed electrician myself. It still wouldn't budge - but cleverly pivoted to offering help with what Big Colour actually does: the graphics and design.

Will it get everything right, every time?

No. And anyone who tells you otherwise is fooling themselves. Humans get socially engineered. Chatbots can be manipulated by someone persistent and creative enough. That's the reality of any system that interacts with the public.

But the underlying models are getting sharper at staying on task, and the guardrails I've built catch most of what slips through. The goal isn't a perfect system - it's one that handles the vast majority of conversations well and doesn't embarrass you when it hits an edge case.

Bunnings, apparently, shipped without thinking about what happens when someone asks their chatbot to explain electrical work. I've spent months thinking about these modes of failure and I will do my best to make sure your bot is intelligent!

Want to see what it looks like for your business?

I'll build you a demo agent for free - no obligation. You can test it, poke holes in it, see how it handles your actual customer questions. I’ll base it’s knowledge on your website contents and set it up to email you if there is an enquiry.

If it works, we talk and work out what it would look like for your business, website etc.

Reply to this email and I'll let you know if it's a good fit.

Recent Website Launch

Pixie Play Cafe

pixieplaycafe.com

As the first Boutique kids cafe in Rathmines [Lake Macquarie, NSW], Pixie Play combines a beautiful indoor playground designed just for little ones, with a one-of-a-kind cafe experience for grown-ups. You’ll find a delicious frozen yoghurt station, wholesome snacks, and plenty of healthy options – including gluten-free and vegan treats for families with dietary needs.

Marianne went to visit Moana and Alex to help them with updating their website. It really is a remarkable transformation for the space into an imaginative play area.

Gemini 3 Pro Preview, Nano Banana

I really cannot emphasise enough how much of an improvement this is on every previous AI image generation tool. The image edit and text on image is brilliant.