Every job platform now has an AI announcement. Almost none of them change how a trade business runs. This post is a field guide for owners: what the current AI features actually are, what a genuinely useful AI on the tools looks like, and the five questions that expose the difference in any sales demo.
The hype: AI as a writing shortcut
Most AI in trade software today is text generation attached to a form field. It rewrites your email to sound friendlier, pads a two-line note into quote line items, or summarises a document you attach. These are real conveniences and they demo brilliantly, but notice what they have in common: the AI never touches the business. It cannot tell you what is on today, because it cannot see the schedule. It cannot tell you which invoices are overdue, because it cannot see the ledger. It is a typing aid wearing an AI badge.
The real thing: AI connected to the working system
The structural difference arrives when the AI is connected to the system of record itself. ONYX1 does this over MCP, an open connector standard supported by the major AI apps, which means the Claude or ChatGPT you already pay for plugs directly into your business. You ask what is on today and it reads the actual schedule. You ask who is working right now and it reads the actual clock-ins. You tell it to draft the quote for Hill St and a real quote appears in the system, numbered and GST-correct, waiting for review. The AI stops being a writing aid and becomes a hands-free way to run the day.
Two properties make that trustworthy rather than terrifying. First, the AI only narrates numbers the system computed; it never invents a total, because totals are recomputed on the server. Second, and more important, everything it can see and do follows the role of the person who connected it, enforced in the database rather than by the prompt. A tech's AI cannot read a price. Not because a prompt asks it nicely not to, but because the permission does not exist where the data lives.
The five questions to ask any vendor
- Who does the AI run as? If the connector runs as an administrator or a service account, everyone using it effectively has admin rights. The AI should inherit the exact role of the person using it.
- Where are the permissions enforced? 'The AI is instructed not to' is a prompt, not a control. Enforcement belongs in the database, where no clever wording can reach.
- What happens on a money move? The standard is a preview plus an explicit confirmation bound to that exact action, so a confirmation for a small invoice cannot be replayed against a large one.
- Is there an audit trail? Every tool call and every confirmation should land in a log you can read. If retries can double up an action, the writes are not idempotent and the bookkeeping will eventually show it.
- Whose AI is it, and whose terms apply? If you connect your own Claude or ChatGPT, your data flows under your account and that provider's terms, and you can switch AI apps without switching job systems. If the vendor bundles their own model, ask what happens to your data and your workflow when they change it.
Why an open standard matters
AI models improve every few months, and the best app today may not be the best next year. Building on MCP means the job system and the AI are separable: when better AI arrives, you switch apps, not job systems. A proprietary chatbot welds the two together, and history is not kind to welded software.
What this looks like in a working day
The owner asks what is on today over breakfast and gets the real schedule. The office tells the AI to draft a quote from the site notes and reviews it before it goes anywhere. On the drive home the owner asks which invoices are overdue, gets the real aging, and tells the AI to mark one paid, confirming the preview it returns. None of that required opening a laptop, and none of it happened outside the permission system. That is the bar. ONYX1's version of it is documented at onyx1.co.nz/ai, with the security architecture at onyx1.co.nz/security, and it is included from the Pro plan with every trial running the full Pro feature set.
Keep reading
- Two-way Xero sync is live on ProQuotes and invoices now sync over Xero's API for connected Pro organisations, with a direction switch per document type and nothing moving until you choose.
- The disruption potential of ONYX1Flat pricing instead of per-seat penalties, AI with the rules enforced in the database, and one lean loop instead of feature creep. How ONYX1 aims to shift the way NZ trades buy and use software.