Maku

Agents that do the work, not just describe it.

Software that takes real action inside your systems, your CRM, your inbox, your tickets, your spreadsheets, with the permissions and approvals to do it safely.

Start a project
How we work

The problem

Chatbots answer questions. Agents finish tasks.

A chatbot answers a question and stops there. An automation executes the same fixed sequence every time, regardless of what it encounters along the way. An agent is different again: it can assess a situation, decide what to do, take an action inside a real system, and adjust course if the first attempt doesn’t work.

That’s a genuinely useful capability, and it’s also a bigger responsibility than a chatbot ever carries. An agent that can update a record, send an email or close a ticket needs the right permissions, a clear audit trail, and a human checkpoint wherever the cost of a mistake is high. We build all three in from day one, never bolted on after something goes wrong.

If what you actually need is a public-facing assistant rather than something acting inside your own systems, have a look at our AI chatbots page instead. Many clients end up wanting both, usually in that order: a chatbot to handle the front door, then an agent behind it to act on what comes in.

There’s a real cost to getting the boundary wrong. Give an agent too much freedom too early and a small error compounds fast across dozens of records before anyone notices. Give it too little and it just becomes an expensive way to ask a model questions nobody acts on. We spend real time in scoping working out exactly where that line sits for your business, before we write any code that touches a live system.

What you get

Where agents pay off, and where they don’t yet.

Triage and routing

Reading incoming tickets, emails or leads, sorting them, and getting the right one to the right person faster than a human queue manager could.

Research and drafting

Pulling information together from several systems and producing a first draft, a summary or a briefing document for a human to check and send.

Reconciliation

Comparing records across two systems, spreadsheets, invoices or orders, flagging mismatches instead of a person doing it by hand each week.

Monitoring and alerts

Watching logs, metrics or inboxes for the pattern that matters, and raising it before it becomes a bigger problem.

Permissions and approvals

Every action scoped to what it needs, with a human approval step wired in wherever a mistake would be expensive or hard to undo.

Audit logs

A record of every action the agent took, when, and why, so you can review its work the same way you’d review a new hire’s.

Where agents don’t pay off yet: tasks with no clear right answer, high-stakes decisions with no room for a wrong call, and anything where the cost of building proper guardrails outweighs the time it saves. We’ll tell you honestly if that’s what we’re looking at.

Process

How we build one.

Map the task

We walk through exactly what a human does today, step by step, including the judgement calls, before we write a line of code.

Design the permissions

We agree which systems the agent touches, which actions need approval, and what it’s never allowed to do.

Build and test

The agent is built against real examples from your own workflow, not generic test cases, and checked against edge cases that trip humans up too.

Launch with a human in the loop

Early runs are supervised closely. Autonomy increases as the track record earns it, not before.

Why Maku

Built by people who also build the systems it touches.

A lot of agent projects are built by people who understand the model but not the system it’s meant to act inside. We do both. Our team does deep custom WordPress development and real AI engineering, so when an agent needs to talk to a CRM, a database or an internal tool, we understand what’s actually on the other end of that connection.

  • Permissions, approvals and audit logs designed in from the first sketch, not added after a scare.
  • Human-in-the-loop by default. Full autonomy is earned, never assumed.
  • We use AI tools ourselves daily, so we know where agents genuinely help and where they just add risk.
  • You own the code. No proprietary platform, no lock-in, and nothing you can’t hand to another team later.
  • Unsure whether an agent, an automation, or a simpler fix is the right call? AI strategy is built for exactly that question.

FAQ

Questions we get asked

What’s the actual difference between an agent and automation?

A normal automation follows a fixed path: if this happens, do that. An agent looks at the situation, decides what to do next, and can change course based on what it finds. That makes it more useful for messy tasks, and it means you need more care around what it’s allowed to touch.

Can an agent break something in our systems?

Any tool with write access can cause damage if it’s built badly, and an agent is no different. We scope permissions tightly, add approval steps for anything risky or irreversible, and log every action so you can see exactly what happened and why.

How do we stay in control of what it does?

You set the boundaries up front: which systems it can touch, which actions need a human to approve first, and which it can just do. Most agents we build start cautious, with more actions handed over once the track record earns it.

What does an agent cost to run day to day?

Running cost is mostly model usage, which scales with how often the agent is triggered and how much it reads each time. We design for cost during the build and show you the expected range before launch, not after the first invoice.

Do we need our own servers or infrastructure?

Usually not. Most agents run on standard cloud infrastructure we set up and manage, connecting out to your existing tools through their APIs. If your systems have hard constraints on where code can run, we work within them.

How do you measure whether it’s actually working?

We define success before we build: time saved, error rate, tasks completed without a human step-in. We track that from day one, so you can see the real number rather than take our word for it.

AI

More AI services

All AI services →

Engage

Building something ambitious?

Send us a few honest paragraphs about what you’re building, who it’s for and when you need it live. We’ll reply with a straight answer.

Start a project

support@maku.au