A long-form deep dive into 8x8 AI Studio for Australian small and mid-sized businesses. Covers the AI Studio Builder and building an agent by describing it in plain English, the difference between conversational and agentic AI, real-time voice straight into the language model, agent actions through integrations, webhooks and APIs into systems like Salesforce and Jira, deployment across voice, digital, SMS and WhatsApp, escalation and handoff to humans with full context, governance and access controls, the unified data layer, consumption-based pricing after a free build-and-test tier, common pitfalls, and how StartCloud scopes and deploys a first agent as an 8x8 Partner.
8x8 AI Studio: AI agents on the phone system you already run
Describe the agent you want in plain English. It gets built, tested and put live on voice, SMS, WhatsApp and web chat, using the same numbers and routing you have today. Here is what it can genuinely do, what it costs, and where you still want a human on the line.
What is 8x8 AI Studio and can it really answer our phones?
8x8 AI Studio is a builder inside the 8x8 platform where you describe in plain English what you want an AI agent to do, and it creates a voice or chat agent that can answer calls, look things up, take actions in systems like Salesforce or Jira, and hand the conversation to a person when it should, deployed across voice, SMS, WhatsApp and web chat without a separate integration project.
StartCloud's verdict: The fastest way to put a genuinely useful AI agent on your phone line, provided you start with one narrow job and keep a human escape hatch
The short version
8x8 AI Studio is an agent builder inside the 8x8 platform. You tell the AI Studio Builder what you want an agent to do, in normal language, and it puts together the prompt, the tools and the connections. You set the limits, test it, and publish it to voice, digital, SMS and WhatsApp.
Agents do two kinds of work. Conversational means answering, understanding and routing. Agentic means taking action in your other systems mid-conversation: raising a ticket, checking a booking, updating a CRM record. Most businesses start with the first and add the second once they trust it.
The thing that makes it different from bolting a chatbot onto a phone line is that it lives in the platform. Voice goes straight into the language model, so it does not have that laggy robotic feel. The channels are already connected. Governance, access control and reporting sit where your phone system already sits.
It is in early availability. Free to build and test, no extra licence for existing 8x8 customers during that period, then consumption-based fees once an agent handles real traffic. We are an 8x8 Partner, so we can scope it, build it and tell you honestly if your use case is not ready.
8x8 AI Studio launched in April 2026 in early availability. Features, limits and regional availability are still moving, so confirm what is enabled for an Australian tenant before you plan around it. Details in this guide are current as of September 2026.
Not a chatbot bolted onto your phone line
Most businesses have met AI on a phone line by now, and most of those experiences were bad. A menu that mishears you. A bot that repeats itself. A long pause, then a clipped robotic sentence. That happens because the usual setup transcribes your voice, sends the text off to a language model somewhere else, waits, then reads the reply back. Every hop costs half a second, and half a second is what makes a conversation feel wrong.
8x8 AI Studio takes a different route. It is built into the 8x8 platform, so it sits on the same infrastructure your calls already travel over. Voice goes into the language model directly, with the live call and the full interaction context available to it, rather than being handed through a translation layer first. You find it in the 8x8 App Store inside your Admin Console, not as a separate product with its own login and its own bill.
The other half of the story is how you build one. There is no flowchart to draw and no script to write. You describe the job to the AI Studio Builder, and it generates the prompt, creates the tools, writes the webhooks and walks you through the configuration. Someone who knows the business can build a useful first agent. What they should not do alone is decide what it is allowed to touch, which is the part we spend our time on.
Conversational: the agent talks
It answers the phone or the chat, works out what the person actually wants, answers from what it knows about your business, and passes them to the right team. Useful on its own, and this is where nearly everyone starts. Think of it as an auto-attendant that listens properly instead of reading out a menu.
Agentic: the agent does things
It reaches into your other systems mid-conversation. Checks a booking, raises a ticket in Jira, looks up an account in Salesforce, confirms an appointment, updates a record, then carries on talking. Multi-step jobs that used to need a person to click through three screens.
Why the calls sound different
The usual bolt-on setup
Every hop adds delay. That is the pause and the clipped delivery callers hate.
8x8 AI Studio
Fewer hops, so the conversation keeps pace with the person having it.
The features that matter
8x8 groups AI Studio into four ideas: create any agent, deploy it instantly, control it with confidence, and scale it. Here is what sits underneath those, without the marketing gloss.
AI Studio Builder
Describe the agent you want in plain English and the Builder designs it: the prompt, the tools it needs, the webhooks, the configuration. No decision trees to draw and no scripted paths that fall apart the moment a caller says something unexpected.
Real-time voice
Voice goes straight into the language model rather than being transcribed first and passed along. That removes the lag and the robotic pauses that make most phone bots painful, so the conversation actually feels like a conversation.
Agent actions
Connect the agent to the systems you already run using built-in integrations, webhooks or APIs. Salesforce, ticketing platforms, booking systems, your own line-of-business app. This is the difference between an agent that talks and one that finishes the job.
Channel deployment
Voice, digital, SMS and WhatsApp are part of the platform already, so an agent can go live on the channels you use without a separate integration project first. Same numbers, same routing, same admin console.
Escalation and handoff
You define exactly when the agent stops and a person takes over. The full context of the conversation goes across with the transfer, so the customer is not asked to explain themselves twice. This is the feature that decides whether customers tolerate the agent or resent it.
Governance and controls
One governance model for the whole platform rather than a separate set of rules for the AI. Access controls, permissions and audit sit where your phone system already sits, which matters a great deal when the agent can take actions on customer records.
Unified data
Every interaction feeds one data layer, so the agents get better at your specific business over time and you can actually see what people are ringing about. Call analysis is one of the quieter wins here: you find out what customers keep asking for.
Agentic automation
Once the basics work, agents can run multi-step workflows end to end. Take the call, verify who it is, do the thing in the back-end system, confirm it, log it, and only involve a human at the point where a human adds something.
If you are weighing this up against building something yourself on top of a general AI platform, the trade is worth naming plainly. Rolling your own gives you total control and a longer project. AI Studio gives you the channels, the routing, the governance and the phone numbers already in place, which is most of the work you would otherwise be doing before your first useful call.
From a sentence to a working agent
This is the bit that surprises people. There is no flowchart. You write down what you want in the way you would explain it to a new starter, and the Builder does the assembly. The work that remains is deciding the boundaries and testing properly.
Steps three and four are where the value is. Anyone can get an agent talking. Getting one that knows its limits and hands over cleanly is the actual job.
A worked example
Say you want the main line answered when reception is busy. You would describe it roughly like this:
"Answer calls to the main number. Find out whether the caller is an existing client or a new enquiry. Existing clients go to their account manager, and if that person is unavailable, take a message and raise a ticket. New enquiries get asked what they need and roughly how many staff they have, then get booked into a call with sales. Anything urgent, or anyone who asks for a person, goes straight through to reception."
The Builder turns that into a working agent, including the ticket-raising tool and the calendar booking. Then you tighten it: what counts as urgent, what it must never promise, how it words the transfer. That refinement is a conversation too, not a config screen.
Build your first agent with a partner who will tell you if it is a bad idea
We are an 8x8 Partner, which means we can scope AI Studio properly, build and test agents with you, and support the platform and the network underneath it afterwards. It also means we have seen which use cases work and which ones sound clever in a meeting and annoy customers in practice.
Being a partner does not make us a salesperson for it. If your call volume does not justify an agent yet, or your source information is not in good enough shape, we will say so and tell you what to fix first. That advice is free and it is usually the most useful part of the conversation.
First-agent scoping
We pick the use case with the best odds, not the flashiest one.
Safe integrations
CRM and ticketing wired up with permissions scoped tightly.
Testing that counts
Handovers, exceptions and the awkward calls, before customers hit them.
Perth-based support after go-live
A local team who answer, and who manage the network under it too.
What people are actually building
8x8 says dozens of customers across more than fifteen industries are running hundreds of agents already. The patterns that keep coming up are less exotic than the demos suggest, and that is a good sign.
Reception and call routing
Intelligent call handling for a main line, multi-site reception and department routing. The agent works out who the caller needs, verifies who they are if it should, and puts them through. For a business where the phone rings all day and one person is trying to cover it, this is usually the first agent to build and it can be live in an afternoon.
Sales enquiry qualification
Inbound leads get answered every time, including at 7pm on a Friday. The agent captures the details, asks the qualifying questions you would ask, drops the lead into the CRM, and books or hands off the ones worth a human call. Salesforce is one of the integrations 8x8 has running in early deployments.
IT helpdesk and internal triage
Staff ring the helpdesk, the agent triages, raises the ticket with the right detail in it, and answers the questions that are genuinely answerable from documentation. Your team stops fielding password and how-do-I questions and gets on with the work that needs a human brain.
Connected workflows between systems
Voice-to-ticket creation, outbound appointment and booking confirmations, collections calls, bilingual agents, CRM updates. Anything where a conversation needs to end in two systems agreeing with each other, and where a person is currently the integration.
After-hours and overflow cover
The calls that currently hit voicemail. An agent can take the details, answer the straightforward ones, flag the urgent ones and leave you a proper record instead of a message saying 'call me back'. Cheaper than an answering service and it knows your business.
Pros and cons
What works well
- You build it by describing it. Plain English in, working agent out, tested and deployable the same day. No specialist developers needed for a first use case.
- It is built into the platform your calls already run on. Same numbers, same channels, same admin console, same governance. There is no middleware layer and no second vendor relationship to manage.
- Voice feeds the model directly, so conversations do not have that half-second lag and clipped delivery that makes callers hang up on phone bots.
- Every channel is already connected. Voice, digital, SMS and WhatsApp come with the platform, so you are not scoping an integration project before your first agent answers a call.
- Agents can take real actions. Raising a ticket, checking a booking, updating a CRM record mid-call is what separates this from a smarter IVR.
- Handover is a first-class feature, not an afterthought. You control when a human takes over and the context travels with the call.
- It is free to build and test during early availability, so you can prove it works on your own use case before committing to spend.
Where to be careful
- It is still early availability. AI Studio launched in April 2026 and is being rolled out to existing customers. Features, limits and regional availability are moving, so confirm what is switched on for an Australian tenant before you plan around it.
- Pricing is consumption-based once you are in production. That is fair, but it means your bill moves with call volume rather than sitting still like a per-seat licence. Model it before you put an agent on your busiest line.
- It only works as well as the information you give it. An agent pointed at out-of-date pricing, an old FAQ or a process nobody follows any more will confidently tell customers the wrong thing. The clean-up is the real project.Same problem we write about with Microsoft 365 Copilot: the AI is only as good as the mess underneath it.
- Giving an agent permission to act needs thinking through. An agent that can update records or raise tickets can also do so wrongly, at speed, hundreds of times. Start read-only, then grant actions one at a time.
- Voice AI lives on your internet connection just like the calls do. Bad bandwidth, no QoS, and the agent sounds as bad as your worst phone call.Our managed internet and firewall work is where this gets fixed.
- Some customers simply do not want to talk to a machine. Be honest with yourself about which calls those are, and always leave an obvious, quick path to a person. An agent that traps people is worse than no agent.
- You need to be on the 8x8 platform. If you run Teams Phone or another provider, AI Studio is not a bolt-on you can buy separately, it is a reason to look at 8x8 as a whole.
Pricing, honestly
There is no public per-seat price for AI Studio, and that is not evasion, it genuinely is not priced that way. It launched in early availability with no additional licence fee for existing 8x8 customers, a free tier for building and testing, and consumption-based fees once agents are handling production traffic.
| Stage | What it costs | What that means for you |
|---|---|---|
| Building and testing | Free tier | Design agents, wire up tools, run them against test conversations and prove the use case before anything touches a customer. This is the part to use properly. |
| Early availability for existing 8x8 customers | No extra licence fee | During the early availability period 8x8 has not charged an additional licence for AI Studio access on top of the platform you already pay for. |
| Production traffic | Consumption-based fees | Once an agent is handling real interactions, cost tracks usage rather than seats. The lever to model is expected call and message volume, not headcount. |
Consumption pricing cuts both ways. It is kind to a business testing one agent on a quiet line, and it needs watching if you put an agent on your busiest number and volume doubles. The number to work out before you commit is not the licence, it is how many interactions a month you expect the agent to handle. We will model that with you.
Pricing structure as described by 8x8 during early availability, current as of September 2026 and subject to change. AI Studio is not sold at a published per-user list price, so confirm current terms and get a quote when scoping a deployment.
What happens with and without an AI agent
These are the everyday moments where a phone either earns you money or quietly costs you some. They happen in every business. The question is which column you are in when they do.
| Situation | People and voicemail only | With an AI Studio agent |
|---|---|---|
| Reception is on lunch and the phone rings | It rings out to voicemail, or three people look at each other and nobody picks up. The caller tries a competitor. | An agent answers on the first ring, works out what they need, and either handles it or puts them through to whoever is actually free. |
| A new enquiry comes in at 8pm | Voicemail. You return the call at 10am the next day, by which point they have already spoken to someone else. | The agent qualifies the enquiry, captures the details into your CRM, and books the follow-up. You start the morning with a warm lead instead of a message. |
| A staff member rings the helpdesk about a password | A technician stops what they are doing, spends four minutes on it, and loses their train of thought on the job that mattered. | The agent handles it, or raises a properly detailed ticket. The technician sees it with the context already filled in. |
| Twenty people ring at once after an outage notice goes out | The queue blows out, hold times triple, and half the callers hang up before anyone can tell them the same thing you told the other half. | The agent takes them all at once, gives the same accurate answer, logs each one, and escalates only the callers who need something different. |
| You want to know what customers keep asking about | You guess, or you ask the team what they think. Nobody has the numbers, so the decision gets made on the loudest anecdote. | Every interaction lands in one data layer. You can see the actual themes and fix the thing causing the calls. |
Reception is on lunch and the phone rings
People and voicemail only
It rings out to voicemail, or three people look at each other and nobody picks up. The caller tries a competitor.
With an agent
An agent answers on the first ring, works out what they need, and either handles it or puts them through to whoever is actually free.
A new enquiry comes in at 8pm
People and voicemail only
Voicemail. You return the call at 10am the next day, by which point they have already spoken to someone else.
With an agent
The agent qualifies the enquiry, captures the details into your CRM, and books the follow-up. You start the morning with a warm lead instead of a message.
A staff member rings the helpdesk about a password
People and voicemail only
A technician stops what they are doing, spends four minutes on it, and loses their train of thought on the job that mattered.
With an agent
The agent handles it, or raises a properly detailed ticket. The technician sees it with the context already filled in.
Twenty people ring at once after an outage notice goes out
People and voicemail only
The queue blows out, hold times triple, and half the callers hang up before anyone can tell them the same thing you told the other half.
With an agent
The agent takes them all at once, gives the same accurate answer, logs each one, and escalates only the callers who need something different.
You want to know what customers keep asking about
People and voicemail only
You guess, or you ask the team what they think. Nobody has the numbers, so the decision gets made on the loudest anecdote.
With an agent
Every interaction lands in one data layer. You can see the actual themes and fix the thing causing the calls.
What it looks like day-to-day
A good agent is mostly invisible. Customers get answered faster, your team gets fewer interruptions, and the only person who thinks about the agent is whoever tunes it.
| Who | Their experience |
|---|---|
| A customer ringing your main number | Talks to something that answers straight away, understands them, and either helps or transfers them to a person who already knows why they rang. It does not feel like an IVR. |
| A staff member on the phones | Gets fewer of the calls anyone could have handled, and the ones they do get arrive with context attached. More time on the conversations that need judgement. |
| A team leader or manager | Sees what the agent handled, what it escalated and why, plus the themes across every interaction. Tunes the agent's instructions when something reads wrong. |
| Whoever builds the agent | Works in the AI Studio Builder inside the 8x8 Admin Console. Describes changes in plain English, tests them, then publishes. No code unless the integration genuinely needs it. |
| StartCloud (as an 8x8 Partner) | Picks the first use case with you, gets the source information straight, wires the integrations up safely, tests the handovers and exception paths, and stays on it after go-live. |
One agent, done properly, then the next
The order matters. Pick the right job, fix the information it will answer from, set the boundaries, then test the calls that go wrong. Businesses that do it the other way round end up with an impressive demo and a customer complaint.
Step 1. Pick one job worth automating
We look at where your calls actually go and find the single highest-volume, lowest-risk job. Usually reception routing, after-hours enquiries or helpdesk triage. One narrow agent that works beats an ambitious one that half works.
Step 2. Get the source information straight
The agent answers from what you give it. So before we build anything, we check the opening hours, the pricing, the process and the FAQ are actually correct and current. This is the least glamorous step and the one that decides whether the agent is trusted.
Step 3. Build and set the boundaries
We describe the agent in the Builder, then spend the real time on the limits: what it is authorised to do, when it must ask for more information, and exactly when it hands to a person. Read-only first, actions added deliberately.
Step 4. Connect it to your systems, carefully
CRM, ticketing, booking, whatever it needs to reach, through integrations, webhooks or APIs. We scope permissions tightly so an agent that can raise a ticket cannot also do something you never intended.
Step 5. Test the awkward calls, not the easy ones
Anyone can demo the happy path. We test the accent it struggles with, the caller who changes their mind, the request that is out of scope, the system that times out, and every handover. If a path is not tested, assume it is broken.
Step 6. Go live small, then widen
Start on one number, one channel or one time window, watch real interactions closely, tune the instructions, then extend to more channels. After go-live we manage it, the network under it, and the changes you want.
An AI agent on your phone line is a communications project and an AI project at the same time. Our business communications work covers the platform and the numbers, our managed internet and firewall work keeps the voice quality where it needs to be, and our AI agent builds and AI governance work covers the part where an agent is allowed to act on your behalf.
Common pitfalls
-
Starting with your hardest use case. The instinct is to automate the thing that annoys you most, which is usually the most complicated. Start with the boring high-volume job, get a win, build trust, then move up.
-
Pointing the agent at information nobody has checked in two years. It will repeat old prices and retired processes with total confidence. Fix the source material first.
-
Hiding the way through to a human. Customers forgive an AI that cannot help. They do not forgive one that will not let them past. Make the escape hatch obvious and quick.
-
Granting actions before you trust the conversation. Let the agent talk and look things up for a while before you let it write to anything. Permissions are easy to add later and awkward to explain after a bad day.
-
Testing only the happy path. The value is in how it behaves when a call goes sideways: unclear requests, background noise, a system that does not answer, a customer getting frustrated. Test those.
-
Nobody owning it after go-live. An agent is not a project that finishes. Someone needs to read what it handled, notice where it got things wrong, and tune it. Left alone, it drifts out of step with the business.
Is 8x8 AI Studio right for your business?
Yes, if you are on 8x8 and your phone is busier than the people answering it. Calls going to voicemail, reception drowning, enquiries arriving after hours, a helpdesk fielding questions a document could answer. AI Studio is the shortest route we have seen from that problem to something working, because the numbers, channels and governance are already in place.
Not yet, if your information is a mess or your volume is low. An agent answering from out-of-date material will damage more trust than it saves time. And if you take twenty calls a week, a well-set-up phone system and a good process will do more for you than an AI agent will.
The honest summary: the technology has got good enough that the hard part is no longer building the agent, it is deciding what it should do, what it must never do, and when it must get out of the way. That is a business conversation, not a technical one, and it is the conversation we would rather have with you first.
Where StartCloud fits in
The Builder is the easy part. Everything around it is where we come in.
Scope and build your first agent
AI agent builds
Set the rules for what AI may do
AI governance
Get the phone platform right underneath it
Business communications
Keep voice quality where it should be
Managed internet and firewall
Connect it to the systems you already run
System integrations
Talk to the Perth team
Get in touch
8x8 AI Studio: the questions we get asked
Is 8x8 AI Studio a separate product we have to buy?
No. It is built into the 8x8 platform and appears in the 8x8 App Store inside your Admin Console, so it uses the same numbers, channels, routing and admin controls you already run. There is no second vendor, no separate login and no middleware to sit between your phone system and the AI.
What is the difference between conversational AI and agentic AI?
A conversational agent talks. It answers questions, understands what the caller wants and passes them to the right place. An agentic agent acts. It calls out to your CRM or ticketing system, checks a booking, raises a ticket, updates a record, then carries on with the conversation. AI Studio does both, and most businesses start conversational and add actions once they trust it.
Do we need developers to build an agent?
Not for a first agent. You describe what you want to the AI Builder and it writes the prompt, sets up the tools and webhooks, and walks you through the configuration. Where you do want technical help is deciding what the agent is allowed to do, wiring it into a line-of-business system safely, and testing the edge cases before real customers hit it.
How is 8x8 AI Studio priced?
It has been in early availability since April 2026 with no extra licence fee for existing 8x8 customers, a free tier for building and testing, and consumption-based fees once an agent is handling real traffic in production. That means the cost tracks how much the agent is used rather than a flat per-seat licence, so the honest answer for any specific business is to model expected call volume and get a quote.
What stops an AI agent from saying the wrong thing to a customer?
Three things you set up front: what the agent is authorised to do, when it must ask for more information, and when it has to hand off to a person. Escalation carries the full conversation context across to the human, so the customer does not repeat themselves. The part worth spending real time on is testing the handovers and the exception paths, not the happy path.
Can an AI Studio agent work in more than one language?
Yes, bilingual agents are one of the early use cases 8x8 highlights. That matters more than people expect for Australian businesses with customers who would rather explain a problem in their first language, and it removes the awkward transfer to whoever in the office happens to speak it.
Do we need 8x8 Contact Center to use AI Studio?
You need the 8x8 platform, and the agent needs a channel to answer on, which usually means a voice number or a messaging channel already configured. A business running 8x8 Work with a main line can put an agent on reception duty. A support team with a real queue gets more out of it, because the agent can triage before a caller ever reaches a person.
See an agent get built, live
Reading about it only goes so far. We run live demo sessions where you watch an agent get described, built and put on a phone line in real time, using your business as the example rather than a tidy fake one. Bring the awkward questions.
One of your real calls
We take a job your team actually does on the phone and build for that.
Built in front of you
No slides. You see the Builder, the testing and the handover to a person.
An honest answer
Including whether your volume and information are ready for it yet.
Hosted by StartCloud as an 8x8 Partner. Tell us the job you want automated and we will book a time. No obligation, Perth-based team.
References
- 8x8 AI Studio product page (official)
https://www.8x8.com/products/ai-studio - 8x8 AI Studio product page, Australia
https://www.8x8.com/en-au/products/ai-studio - 8x8 AI Studio documentation: About
https://docs.8x8.com/8x8WebHelp/8x8-ai-studio/Content/AI-studio/about.htm - 8x8 brings agentic AI natively to the 8x8 platform for CX (8x8 investor news release, April 2026)
https://www.investors.8x8.com/news-releases/news-release-details/8x8-brings-agentic-ai-natively-8x8-platform-cx - 8x8 launches AI Studio, lets any CX team build their own AI agents (CX Today, 2026)
https://www.cxtoday.com/contact-center/8x8-ai-studio-cx-ai-agents/ - The user interface of the future is your voice: inside 8x8's AI Studio (SiliconANGLE, April 2026)
https://siliconangle.com/2026/04/14/user-interface-future-voice-inside-8x8s-ai-studio/
Document prepared September 2026 by StartCloud (Start Technologies Pty Ltd). 8x8 AI Studio is in early availability and its features, limits, regional availability and commercial terms are subject to change, so confirm current details when scoping a deployment. 8x8 and 8x8 AI Studio are products of 8x8, Inc. StartCloud is an 8x8 Partner and is not otherwise affiliated with 8x8, Inc.