AI development is the work of designing, building, and deploying custom AI agents and automations inside a business's existing systems. It covers the build itself, LLM application development, and the less glamorous half, implementation: wiring the thing into the tools your team already uses and proving it holds up under real work. The deliverable is a working workflow, not a report.
What a build actually involves
Every real build has three layers. Integration: the agent gets read and write access to your systems of record, Xero or MYOB, your CRM, your inbox, through each platform's supported API with scoped credentials. Grounding: the agent answers from your documents and procedures, so its outputs are checkable, because an invented figure written into a live record is worse than no automation at all. Governance: explicit boundaries between what the agent does alone and what waits for a person, with every action landing in an audit log.
Integration is where I spend most of my build time, and it's where the results come from. The model is the cheap part. The plumbing is the product.
Choosing your route
| Route | What you get | Where it breaks |
|---|---|---|
| Off-the-shelf tool | A vendor's workflow, fast | The job spans systems the vendor doesn't reach |
| No-code platform | DIY automations on subscriptions | Your team owns the maintenance burden forever |
| Custom agency build | Your workflow, in your stack, with handover | Costs more than a subscription up front |
| In-house hire | Full control | One salary buys one person's bandwidth |
The honest comparison is not tool versus tool, it's who carries the delivery risk. A fixed-price build against a defined workflow puts that risk on the provider. An open-ended hourly engagement puts it on you. We price per agent, against outcomes, and I'd walk from anyone who can't tell you what the agent will save before they tell you what it costs.
Where to start
Not with the model. Pick one repetitive, high-volume process, price the manual hours honestly, and scope a build against it. That is the whole logic of our audit: interview the team, map every candidate workflow, rank them by payback, and spec the top ones. If you want the conceptual grounding first, start with what an AI agent actually is, and if a role-shaped agent fits better than a workflow-shaped one, the AI roles section maps those. For the delivery model where your team ships its own automations after a workshop day, see automation services.