Harnessed UX
Designing interfaces that survive the AI agent.
What it is
VOID's offer dedicated to AI products.
For 10+ years, we've designed the digital interfaces of the largest banking, financial and insurance institutions in Morocco and Africa. With AI agents landing in customer portals, chatbots, business copilots and internal assistants, classic UX no longer cuts it. Harnessed UX integrates the technical harness constraints — guardrails, fallbacks, observability — from the scoping phase.
Useful design
Four promises. Nothing more.
- Promise 01
Protect user trust against AI agent drift.
- Promise 02
Smooth human-agent handoffs, even in degraded mode.
- Promise 03
Surface what the agent knows, ignores, and proposes.
- Promise 04
Make auditable every agent decision for regulated environments.
Services
Four ways in.
UX audit of existing AI interfaces
Chatbots, copilots, assistants. Diagnosis of risk zones, missing fallbacks, absent trust patterns.
Design of new AI experiences
End-to-end design via the Harnessed UX methodology, from user research to AI design system.
Joint UX + technical harness scoping
Mixed sessions with your tech / data teams. Latency, availability and degraded-mode constraints embedded from day one.
Design & product team training
On AI constraints: trust patterns, explainability, regulated compliance. Workshop format, 1 to 3 days.
UX audit of existing AI interfaces
Chatbots, copilots, assistants. Diagnosis of risk zones, missing fallbacks, absent trust patterns.
Design of new AI experiences
End-to-end design via the Harnessed UX methodology, from user research to AI design system.
Joint UX + technical harness scoping
Mixed sessions with your tech / data teams. Latency, availability and degraded-mode constraints embedded from day one.
Design & product team training
On AI constraints: trust patterns, explainability, regulated compliance. Workshop format, 1 to 3 days.
Methodology
The 4H approach.
Harmonize
Environment, stakes, harness constraints. Joint business · product · tech scoping. AI risk zones identified.
Hypothesize
Ideation, scenarios, fallbacks. We design both nominal cases and the cases where the agent fails, doesn't know, or goes offline.
Harness
Prototype, interface, interaction. Confidence indicators, reasoning transparency, human-agent handoffs.
Honor
User testing, interviews, focus groups. Validation against the 12-point Harnessed UX evaluation grid.
The team
An expert and award-winning team.
Pragmatic
We talk about what we know how to do.
Curious
Daily watch on models, patterns, frameworks.
Rigorous
Every design choice is challengeable by a CIO.
Committed
15 years serving banking, insurance, regulated.
Awards & recognition
VOID has been awarded for over a decade at major design & tech competitions — Awwwards, CSS Design Awards, Drupal Splash Awards, FWA.
The proof
2 real cases. 2 methodological assets.
We own it: Harnessed UX is a young offering, backed by technical choices tested internally before being sold.
Autogit × Harness Engineering
Sovereign AI agent for security remediation, 100% on-premise, deployed on our internal projects. The PR review interface was designed following the complete 4H approach. Detailed zoom below.
Read the Harness Engineering articleCompliance documentary copilot
Supporting a compliance team on a regulatory documentary search copilot. Joint UX + harness scoping, source citation patterns, fallback to a human legal expert.
Harnessed UX grid — 12 points
AI interface evaluation grid: trust, transparency, handoff, fallback, auditability, reversibility, privacy, feedback loop. Standalone deliverable, reusable.
Pattern library — banking AI interfaces
Pattern library dedicated to banking / insurance: confidence indicators, human-agent handoff, regulatory citation, revocation, consent.
Flagship case zoom
Autogit × Harness Engineering.
A sovereign AI agent for security remediation — 100% on-premise, with zero data leaving the infrastructure. The review interface was designed following the complete 4H approach.
on-premise. Zero data exfiltration.
evening critical CVE → morning PR with green tests.
projects covered by the pilot.
full approach applied end-to-end.
Four design angles.
01. Joint UX + tech scoping
Mixed designer + infra engineer sessions from day one. Risk zone mapping (breaking change, transitive dependency, ambiguous fix). Latency and degraded-mode constraints embedded in wireframes.
02. Agent PR review pattern
Every agentic PR surfaces the CVE addressed, the confidence on the fix, the decision trace, the tests run. The human reviewer approves or rejects at a glance.
03. Human handoff fallback
On major breaking changes (v2 → v3), the agent doesn't attempt: it documents what it saw, lays out the context, and hands off to a developer with a pre-written briefing. Zero auto-merge.
04. UX observability
Reviewer dashboard: acceptance rate, average latency, CVEs closed, agent fatigue signals (stale, drift). The UX itself is instrumented to feed self-improvement.
For full details (stack, technical choices, pitfalls avoided), read the article Harness Engineering: making AI agents reliable in production.
You're deploying an AI agent.
We design the interface that makes it defensible.
Customer chatbot, business copilot, internal assistant. Audit, design, joint scoping, training. Let's discuss your context in 45 minutes.