AI agents for defined, repeatable work, behind approval gates you control.
We design and build AI agents that take repetitive, rules-based work off your staff: sorting requests, drafting replies, assembling reports and checking documents. They work from your knowledge base, inside the limits you set, and a person approves anything that matters. It is how we run Jundra’s own projects.
Opens a prepared email to admin@jundra.com. The first consultation is free; discovery is a paid stage, scoped in writing. How it works
Teams buried in repetitive, rules-based work.
Operations, customer service, finance, compliance and program teams in larger organisations, where skilled people spend hours on tasks that follow the same steps every time.
Signs you need it
- Staff copy and paste the same information between systems.
- Requests arrive in shared inboxes and wait to be sorted and routed.
- Weekly reports are assembled by hand from several sources.
- Documents are checked against the same checklist again and again.
Agents that finish defined tasks end to end, inside the limits you set.
The exact deliverables are agreed in your written proposal.
- A task map and baseline
The tasks worth automating, how they run today, and the ones better left to people. - A custom agent team
AI agents with defined roles that read your knowledge base first, work through tracked tasks and write up what they did. - Approval gates
A person says yes before anything is sent, spent, published or changed in production. - Least-privilege access
Agents get only the system access each task needs, through separate accounts where possible. - Guardrails and red-teaming
Tested against agreed test tasks, plus prompt injection, bad inputs and out-of-scope requests, with results shown to you before go-live and re-run when models change. - Audit trail and pause switch
Agent actions recorded with their inputs and approver where your systems allow, and a way for your team to pause any agent. - Runbook and handover
How to run, pause and change the agents, handed over to your team.
Read this diagram as a list
- A task arrives: a request in a shared inbox, a report that is due or a document to check.
- The agent reads your knowledge base first, checks your rules and works through a tracked task, with only the system access it needs.
- Approval gate: nothing is sent, spent, published or changed in production until a person says yes. If it is not approved, the work goes back to the agent.
- The action goes ahead, the agent writes up what it did, and actions are logged where your systems allow.
- Routine steps run within the limits you agree, and your team can pause any agent; pausing stops new actions, and anything already approved is handled through your normal process.
Six steps, from one task to a wider rollout.
We automate one well-defined task first, test it against agreed results, then expand.
STEP 01
Free consultation
30 minutes, free30 minutes on the tasks that eat your team’s time and whether agents are the right answer. No charge, no obligation.
STEP 02
Discovery (paid)
paid, scoped in writingWe map the chosen tasks with the people who do them, the systems involved and the rules that apply. A paid stage, scoped and priced in writing before it starts.
STEP 03
Written proposal
scope, steps, priceScope, approval gates, test tasks, timeframe and price, in writing.
STEP 04
Pilot or MVP
fixed deliverablesAgents take on one task with a person approving their work, tested against the agreed results.
STEP 05
Rollout
scale what worksMore tasks and teams, once the first one works for you.
STEP 06
Handover and support
your team, your filesRunbook, training, 60-day defect fixes and optional support, including re-testing when models change.
We run Jundra’s projects with an AI agent team.
Jundra’s AI agents have defined roles, read our LLM wiki before acting, work through tracked tasks and write up what they did. Hard approval gates mean nothing is sent, spent, published or changed in production without a person saying yes.
We use the same tools and the same discipline on your work, and bring in the specialist developers and engineers each project needs.
Want to test it on one task? Book a free 30-minute consultation →
What you can count on.
Free first conversation
A 30-minute consultation costs nothing and commits you to nothing. Discovery is a paid stage, scoped and priced in writing, and the pilot or MVP has fixed deliverables.
60-day defect fixes
We fix defects in our work for 60 days after delivery, at no charge, on top of your rights under the Australian Consumer Law.
People stay in charge
A person approves anything sent, spent, published or changed in production, and your team can pause any agent from its controls.
Prequalified and insured
A prequalified supplier under the NSW Government ICT Services Scheme (SCM0020), with professional indemnity, public and products liability cover.
Built to fit Australian AI guidance.
We design to the National AI Centre’s Guidance for AI Adoption, and prepare the paperwork your own review needs: an AI register entry, a risk and impact summary, test results and the points where a person approves. That gives your team ready-made evidence for:
- the NSW AI Assessment Framework (NSW agencies);
- the Australian Government’s policy for the responsible use of AI in government (Commonwealth agencies);
- the National AI Centre’s Guidance for AI Adoption (any organisation);
- the OAIC’s privacy guidance on commercially available AI products.
Your privacy, legal and risk owners make the decisions; we give them the information to make them quickly.
Ownership of the work, data privacy and pricing are answered in the questions below.
Questions we get asked.
Is this just a chatbot?
No. A chatbot answers questions. An agent completes a task: it reads the request, checks your rules, does the work in your systems and records what it did, asking a person to approve anything important.
Can’t we just build this ourselves in Copilot Studio?
Often you can, and we will tell you when that is the simplest path. Where we help is choosing the task worth automating, setting the access and approval rules, testing it against agreed results, and connecting it to systems outside Microsoft 365.
Will the agents act on their own?
Only within the limits you agree. Routine steps can run automatically; anything sent, spent, published or changed in production waits for a person to approve it.
Which tasks are a good fit?
Repetitive, rules-based tasks with a clear right answer and a way to check it, such as sorting and routing requests, drafting replies for review, assembling reports and checking documents against a checklist. We will tell you when a task is better left to people.
Do the agents need access to our systems?
Only what each task needs. We agree the access in writing, use separate accounts where possible, and log their actions where your systems allow, so you can see what happened.
More questions (6)
Do we need to update our privacy policy?
From 10 December 2026, if your organisation is covered by the Australian Privacy Principles and uses a computer program, such as an AI agent, to make or substantially and directly inform decisions that could significantly affect people’s rights or interests, your privacy policy must say so (APP 1.7–1.9). We document what each agent does with personal information so your privacy team can decide what to disclose.
What about data privacy and confidentiality?
We never put your data into free or consumer AI tools. Before we start, we agree in writing where your data is stored, where AI processing happens, which services may process it and who on our team can see it, to support your obligations under the Privacy Act and the Australian Privacy Principles. Where data needs to stay in Australia, we can use Australian cloud regions where the provider offers them, or run open-source models on your own servers. Where you want a commercial provider, we use business services under terms that do not allow your data to be used for training. We are happy to work under your confidentiality agreement, and the proposal says what happens to copies of your data at the end of the engagement.
You’re a small firm. Can you handle an organisation our size?
Yes. You work directly with Jundra’s project leads, Ken Taggart and Ali Khorsandian, and we bring in specialist developers, engineers and security testers as each project needs, named in your proposal. We start with one team or one task, so the first stage is sized to fit, and we have supported one critical rail monitoring system for more than a decade.
How can we buy from you?
Directly, or through the NSW Government ICT Services Scheme, under which Jundra is a prequalified supplier (SCM0020). Tell us how your organisation buys and we will confirm in writing what applies before discovery starts.
What does it cost?
The 30-minute consultation is free. Discovery is a paid stage, scoped and priced in writing before it starts. After that, scope and price depend on your organisation, so we do not publish package prices: you get a written proposal with the scope, steps, timeframe and price before the pilot starts, and the pilot or MVP has fixed deliverables.
Who owns the work?
You do. You own the project work you pay for, handed over into accounts you own. Pre-existing tools and third-party components, including commercial AI models, stay under their own licences, listed in your proposal. Every project includes a 60-day defect-fix period after delivery, in addition to your rights under the Australian Consumer Law.
Related AI services
Book a free 30-minute consultation.
Bring one task your team does every week. We will tell you whether an agent should do it.
You’ll talk with Ken Taggart or Ali Khorsandian, who lead Jundra’s projects. In 30 minutes we will:
- look at one task your team does every week;
- tell you whether an agent should do it, or whether something simpler would work;
- name what the agent would need access to, and what a person must still approve;
- outline what a paid discovery would cover.
No charge and no obligation. If you go ahead, discovery is a paid stage, scoped and priced in writing. Please do not include confidential information in your email.
How this button works
It opens a prepared email to admin@jundra.com in your email app; nothing is booked or sent automatically. Press Send and we reply within one business day to agree a time: a video call, a phone call or a visit to your site. Using webmail? Copy the address into a new email. How we use enquiry details.