AI agents for support, sales and internal operations
Agents that complete defined tasks inside your systems, with people on the decisions that matter.
An AI agent carries out a defined task inside your systems: answering a routine support question, preparing a sales summary, or completing an admin step it is allowed to take. It works within limits you set. It is not a general chatbot, and it is not a fixed workflow with a model attached.
Who this is for
Operations that can name the task, the system that holds the information, and what the agent must not do. 23Labs is based in Melbourne and works remotely with businesses in Australia and New Zealand. We build agents as part of how a business operates, not as an AI product looking for a use.
Support answers the same questions by searching
A person looks up a status, a policy or a record and replies. The path is similar every time, and the wording of the question is not.
A task needs several lookups
Preparing an account summary, a handoff or an internal answer means opening more than one system. A single fixed rule does not cover it.
A person should still approve the commitment
You want the preparation done, and you want a person on anything that changes a record, spends money, or promises a customer something.
The tasks people are still chasing through systems
An agent is for work that starts as a question or a request in plain language, and only becomes a next step once someone has retrieved the information.
Repetitive support
The same operational questions take a person off other work, because answering means searching a system rather than applying a template.
Answers assembled by hand
Sales or account support pulls a summary together from records before a conversation. The assembly is the task. The conversation still needs a person.
Admin that is a search, not a rule
Internal requests arrive in different words, and someone has to find the right record before they can act. The action itself may be simple once the record is found.
What an agent is allowed to do
We build task-based agents for support, sales support and internal admin. Voice is in scope when the task is a phone conversation and the limits are the same. The agent only reaches the systems and actions agreed up front.
When this is the right service
An agent takes a task, retrieves what it needs, and carries out the actions it is allowed to take. Classifying a document or extracting fields is AI workflow automation. A predictable sequence with no judgement is business process automation. We recommend those when they fit the work better. A new screen or portal for the team is custom software.
How an AI agent engagement works
The design is the task and the boundary. We do not start from a demo agent and look for somewhere to put it. There is no published timeline or price.
Name the task, the source of truth and the limits
We define what the agent is for, which systems or documents it may read, and which actions it may take. Anything outside that list is out of scope.
Decide what a person still approves
Routine answers can go through. A change to a record, a commitment to a customer, or anything ambiguous waits for a person.
Build it inside those limits
We connect it only to what the task needs, and we test it on real questions and records.
Review the first real tasks
We look at what it did, where it handed over, and where the limit was too loose or too tight. It does not gain new permissions on its own.
Where an agent fits
Illustrative tasks in transport and logistics, recruitment, and professional services. There is no published AI agent project on this site. The linked industry pages describe the operational context, not a deployed agent.
Customers ask where a job is, and someone looks it up to answer. An agent can handle the lookups it is allowed to make and hand over anything it cannot confirm. Reading a delivery document into fields is AI workflow automation, not this.
A consultant needs a short account summary before a client call, and someone assembles it from the recruitment system for them to approve. That is an agent task. The Huntress Recruitment project was an integration that moved vacancy data between Bullhorn, SEEK and the website. It was not an agent.
Clients ask what they need to send, or where a matter sits, and a person answers from agreed material. An agent can give that answer and pass anything commercial or unclear to the team. It does not replace an onboarding workflow.
Published work, and what it is not
There is no AI agent case study on this site. The projects below show how we treat operational work. They are workflows and an integration. They are not agent deployments.
CorIT Tech
Onboarding created projects and tasks from the CRM by rule. A person was not asking an assistant to decide the next step. CorIT reported saving more than 10 hours of administrative work a month.
Read the projectHuntress Recruitment
Vacancy information moved between Bullhorn, SEEK and the website, with public-facing fields kept separate from the internal record. That is a connection between systems, not an agent acting on a recruiter's behalf.
Read the projectFrequently asked questions
What is an AI agent?
Software that can take a defined task, look up information in your systems or documents, and carry out the steps you have allowed. That might be a routine support reply, an account summary, or an admin update. It is not a general chatbot, and it is not the same thing as automating a fixed workflow.
How is an AI agent different from AI workflow automation?
AI workflow automation handles one interpretive step, usually a document or a classification, then passes the result on. An agent is responsible for a task that may involve several lookups and actions. Because it can change records or reply to a customer, the limits and the approvals are part of the design.
How is an agent different from business process automation?
Business process automation follows rules you can write down. An agent is for requests that arrive in plain language and need information retrieved before anyone knows the next step. If the path is already predictable, an agent is the wrong tool.
What oversight does the agent have?
It only reaches the systems and actions agreed up front. Routine replies can be handled directly. Anything that changes a record, makes a commitment, or sits outside the defined task is handed to a person. That line is set in the design.
Do you publish an AI agent case study?
No. Published work covers onboarding automation, a recruitment system integration, a real estate enquiry workflow and a website rebuild. An agent engagement starts from a specific task and the systems it would need to touch.
How much does an AI agent cost, and can it use our current systems?
Custom software and automation projects typically start from A$2,500. Final pricing depends on the systems involved, complexity and scope. For a larger or more complex project, book an initial consultation. Cost also follows the task, the systems it may touch, and how much a person still approves. It can use a CRM or an inbox where that system allows a controlled connection. We do not publish a list of supported products, and we name a platform only once that connection is part of the project.
Do you build AI agents for New Zealand businesses?
Yes, remotely. 23Labs is based in Melbourne, serves Australia and New Zealand, and has no New Zealand office.
Tell us the task, and what the agent must not do.
Name the question or the admin task, the system that holds the answer, and the actions that need a person. That is enough to tell whether an agent fits, or whether a simpler workflow does.
Custom software and automation projects typically start from A$2,500. Final pricing depends on the systems involved, complexity and scope. For a larger or more complex project, book an initial consultation.