Concept work

Synq

automation you can read.

synqLIVENeeds your reviewINVOICE 2026-0418Lumen d.o.o. · €6,240.00Due date unclear — two dates found.ApproveEditInvoice intake214 runsContract summary38 runsLead routingPausedsynqOverviewWorkflowsData sourcesAgentsInvoice intakeLIVETRIGGER
Invoice receivedAI STEP
Extract fieldsDECISION
Over €5,000?RUNS TODAY214NEEDS REVIEW6synqOverviewWorkflowsData sourcesAgentsAudit logSettingsInvoice intakeLIVECanvasRunsVersionsTRIGGER
Invoice receivedAI STEP
Extract fieldsDECISION
Total over €5,000?YES
Ask for approvalNO
Post to ERPINSPECTORExtract fieldsConfidence 0.94VendorLumen d.o.o.Invoice2026-0418Total€6,240.00DueReviewShow reasoning →ApproveOverride

A concept AI workflow platform: a typed canvas, dashboards on a strict density grid, and AI steps that always show their reasoning and a way to override.

Category: AI · Enterprise SaaS
Year: 2026
Disciplines: Product design, AI UX, Design system
Typography: IBM Plex Sans × IBM Plex Mono
Concept work: A self-initiated concept. There is no client, and no results are claimed.

The brief we set ourselves

AI automation is often sold as magic: a box that does things you cannot see. We wanted the opposite — automation you can read, where every AI step shows its reasoning and a person can always override it.

The idea

Synq is a fictional workflow platform for finance teams. Workflows are typed: every step declares what it takes in and what it produces. Dashboards sit on a strict four-point density grid. When the model is unsure, the work is routed to a human with the evidence attached.

Typography and colour

IBM Plex Sans and Plex Mono, chosen for density and trust. The ground is deep slate; green means running or approved, amber means a person is needed. Nothing else is coloured.

Typography · DisplayIBM Plex Sans

Typography · TextIBM Plex Mono — dense, trustworthy

Depth#0E1416 · Ground
Panel#172024 · Nodes
Run#6FD3B5 · Live and approved
Review#F2B84B · Needs a human

Signature interaction

Select a step on the canvas. Open the AI step's reasoning, then approve the result or override it — the canvas records who decided.

Signature interactionLive — try it

Invoice intakeLIVE

INSPECTORExtract fields

Reads vendor, number, total and due date. Input: document. Output: typed invoice.

Vendor: Lumen d.o.o.
Total: €6,240.00
Due: 2026-11-18
Confidence: 0.71

What we would test next

Whether reviewers read the reasoning or skip it, what confidence threshold sends the right amount of work to people, and how to show an override so the model and the team both learn from it.