AI agents for business: how the good ones work
How AI agents for business work, what custom agents cost, and the build sequence that separates shipped systems from stalled pilots.
AI agents for business execute work. They do not only answer questions about work. The distinction matters: a chatbot tells your team where the job note is; an agent reads the job note, checks the customer record, drafts the update, and routes anything unusual to a person. Companies in this space vary enormously in whether they can actually deliver the second thing, and this guide is how to tell.
What good agents look like
The agents worth paying for share four properties:
- They work a real queue. Tickets, quotes, invoices, dispatch boards, close-out packs. Not a demo environment, the actual system your team logs into.
- They use source records. The agent looks its answers up in your documents and records rather than answering from memory, so every important action carries a source.
- They are checked before handover. A quality-check set of real cases from your workflow runs before every change. Ask a vendor what cases backed their last deployment; silence is your answer.
- They know what to hand to a human. The repeat path is theirs; ambiguity, exceptions, approvals, and judgment route to your people with context attached, not dropped on the floor.
The build sequence that works
Every stalled pilot story we hear skipped a step in this sequence. Map the process first, with the operators who run it. Name the source records and tool connections second. Build the agent third, on the workflow the diagnostic ranked first, not the one that demos best. Then operate it with monitoring your own team can read. Agencies that start at step three produce demos; the queue teaches them your edge cases live, at your expense.
What it costs, honestly
Custom agent systems move with scope: how many tools are touched, how clean the source records are, what the exception path needs, and how acceptance will be measured. Svolta starts with Workflow Scoping from $2,995 AUD ex GST per workflow. Workflow Launch is from $14,995 AUD ex GST per workflow and includes all scoping expenses. AI Accelerator is from $995/m AUD ex GST per workflow, with new features and material changes separately scoped. Bigger builds cost more.
Shortlisting: three artefacts to demand
Ask every company for a redacted Baseline Pack, a redacted quality-check summary, and the production support boundary from a past build. The first shows whether they measure before promising. The second shows whether “it works” means anything. The third shows who restores the accepted system and how changes are classified after launch. A free 30-minute fit call is where we tell you whether your workflow has the volume, access, and ownership to justify Workflow Scoping.
Start here
Bring one workflow to the fit call.
Bring one workflow, its volume, and the systems it touches. The call checks whether the work is defined well enough for a useful next step.
We use the call to check whether Workflow Launch fits the workflow, evidence, and decision boundary. The service page remains the source for scope and pricing.
If the fit is unclear, we say so on the call. There is no required follow-on sequence.
Go deeper on each part.
The spokes of this guide. Each one takes a single question from the topic above and answers it properly.
AI agent implementation checklist for production
A production AI agent needs a bounded job, accountable owner, controlled data and tool access, test evidence, human stops, monitoring, and a rollback path. Use this checklist before increasing its autonomy.
Read AI agent implementation checklist for productionAgentsCustom AI agent development services: what 'custom' should mean
Every vendor sells 'custom' AI agents. The word should mean your workflow, your systems, your data, your evals, and your ownership at handover. A checklist for reading proposals, and the parts that should never be custom.
Read Custom AI agent development services: what 'custom' should meanThe questions that come up.
Answered plainly. If yours isn't here, a free Consultation is the fastest way to get a real answer for your operation.
A chatbot answers questions about work; an agent executes work. A chatbot tells your claims handler where the policy document is. An agent reads the claim, checks coverage, assembles the file, and routes it with citations, leaving the coverage decision to the human. If a proposal's screenshots are all chat windows, you are being sold a chatbot.
Custom agent systems move with scope: the systems touched, data readiness, exception path, review points, and evidence needed for acceptance. Svolta's per-workflow prices are from $2,995 AUD ex GST for Workflow Scoping, from $14,995 AUD ex GST for Workflow Launch including all scoping expenses, and from $995/m AUD ex GST for AI Accelerator. Bigger builds cost more.
Workflow Scoping is usually 1 to 2 weeks. Launch timing then depends on the signed integrations, data, exception path, and acceptance boundary. The target is a first measured improvement 14 to 30 days after the production thin slice goes live, not a blanket promise that every complete build ships in that window.
Book a free consultation.
Bring one workflow this guide made you think about. A 30-minute call checks whether Workflow Launch is the right next step. The service page remains the source for scope and pricing.