What "AI in field service" actually means in 2026 — and what's marketing copy

Every field-service platform has shipped an “AI” feature in the last 12 months. Almost none of them ship artefacts. Most of them ship a chat panel. Here's how to tell the difference before you buy — and why it matters.

The chat-panel pattern

Open most major field-service apps in 2026 and you'll find a sparkle icon in the top-right corner. Click it; a chat panel slides out. Type a question. The AI answers. The answer is usually a paragraph of text — sometimes useful, often generic, occasionally hallucinated. You then have to take that text and do something with it: copy a line into a quote, retype a schedule suggestion, paraphrase a customer message you might send.

This is “AI” in the same sense that a search box is search. It's a generic affordance bolted onto an existing product, with the cognitive lift of turning information into action staying entirely on you.

The artefact pattern

AI features that ship artefacts produce structured outputs that drop directly into the workflow. Examples from Stelid:

• Photo-to-Quote: take 5 photos of a job; the output is a quote with line items, quantities, and unit prices pulled from your pricebook. Not a paragraph describing what the quote should be — the actual quote, dropped into the editor for review.
• Voice-to-Job: speak a job description; the output is a structured job with type, priority, scheduled time, and required materials. Not a chat reply — a job, ready to dispatch.
• Smart Scheduling: ask for a slot; the output is a calendar entry with the time, the assigned crew, and the travel-time accounting against the surrounding jobs. Not “I suggest Tuesday at 2pm” — Tuesday at 2pm, in the schedule.
• Customer messages: one-click follow-up; the output is a fully-drafted message ready to send. Not a template — your message, in your voice, customer-name-substituted, ready to fire.
• AI Receptionist: a chat widget on your site; the output is real bookings, in your real schedule, with full transcripts and audit rows. Not a “live chat” theatre — live booking, with real money attached.

Why this matters

Chat-panel AI is a productivity ceiling: at best it shaves a few seconds off every task by answering a question faster than a search engine. Artefact AI is a productivity step-change: it removes whole tasks from the workflow.

Quoting a job from a chat suggestion still requires you to parse the suggestion, decide what to keep, type it into the quote editor, and send. Quoting a job from Photo-to-Quote requires you to review what the AI wrote and click Send. The difference compounds across hundreds of jobs a year.

How to tell what you're looking at

Three questions for any AI feature:

1. What does the output look like? A paragraph of text? It's chat. A structured object inside the product? It's an artefact.
2. What do you do next? Copy/paste/retype something? It's chat. Click “Send” or “Approve”? It's an artefact.
3. Where's the audit trail? A chat panel that doesn't log anything is also a chat panel that can't be held accountable. Stelid writes every AI call to ai_audit_log with the model, provider, token usage, and the org-wide kill switch. EU AI Act Article 50 isn't optional; the audit isn't optional either.

The honest version

AI as a chat panel is easy to ship. AI as artefacts is harder — you need product design (what shape is a quote? a job? a schedule entry?), prompt engineering tuned to that shape, structured outputs validated server-side, an editor that lets the human override anything, an audit trail per call, an org-level kill switch, and a compliance disclosure on every AI surface. Two engineering quarters per feature, not two weeks.

Most platforms shipped chat first because chat is what their leadership team saw in ChatGPT. Some are now retrofitting artefact features. We started with artefacts, and we're going to keep starting there.

See the 7 AI features in Stelid — every one of them produces an artefact. Or read the trust page for how the audit trail works.