Solve AxisWork with me
AI-native product design · case study

A modelling product, organised around how analysts actually reason about a solve.

Mature enterprise tools accrete bespoke run-forms because they were never built around the user’s question. Solve Axis starts clean — with AI as connective tissue and a designed accountability layer, not a feature bolt-on.

Abstracted from real enterprise work · anonymised throughout · a fictional product.
The signature idea
SOLVER CONFIDENCEmaths
0.4%optimality gap
Proven near-optimal. A property of the maths — not a guess.
AI-SELECTION CONFIDENCEjudgement
Moderate
The AI's estimate that it chose the right solve. Not a guarantee.
IBM Carbon design systemSource Serif 4 · DM SansThe two-confidence modelProjected, instrumented metricsPressure-tested by 4 AI personas
The design tenets

Four principles the whole product obeys.

01

Configuration is an expression of a model

Not a pile of bespoke run-forms. A few underlying dimensions organise every solve, so the user navigates by intent — not by hunting for the right named form.

02

Two kinds of confidence, never one

Solver optimality — defensible maths — and AI-selection confidence — advisory judgement — are shown apart. A single blended number is worse than nothing.

03

Override is a recorded decision

Pushing back means changing the question, not nudging sliders. Every choice — change or keep — is written to a sealed, signed, immutable account.

04

Show the boundary, don’t hide it

Some questions can’t be run at scale. The product renders where it ends — and names what’s a design problem versus a platform decision.

The workflow

Ask Review Record.

One run, end to end — and the rule that holds throughout: nothing executes until a human has reviewed it.

90-second narrated walkthrough
The walkthrough, looping — sound on for the narration.
Ask01

Goal-Led Entry

Describe the decision in plain language. The AI proposes the right solve — with a recognizable anchor — and keeps the model’s dimensions backstage.

Review02

The Accountability Moment

Before anything runs: two confidences, the one load-bearing assumption pulled forward, and override expressed as changing the question.

Record03

The Decision Record

Run & record seals a tamper-evident artifact — assumptions, confidence basis, overrides and a human sign-off — exportable for audit.

The craft

Six real screens carry the argument.

Pulled straight from the design file. Click any screen to open it full-size — representative solves at different corners of the model, plus the wider platform they live in.

The framework

A model, then AI as its connective tissue.

The organising dimensions locate any solve, so the same surface absorbs the next run type the market invents. AI doesn’t sit beside the model — it expresses it: proposing a solve from a plain-language goal, and carrying a designed accountability layer that makes it safe to trust on a high-stakes decision.

01Goal-led configuration — the model made conversational
02Two-confidence trust — solver maths vs AI judgement
03Override that writes to an immutable record
04An honest boundary where design meets the platform
The honest edge

Not every question can be run.

The same dimensions that organise a solve also map its cost. Some corners are cheap, some expensive, some beyond reach today — and the product says so, plotting where a question sits and offering the honest response.

Green and amber are a design problem. Red is a platform decision. Knowing which is which — and never blurring the two — is the product.

RUNNABLEEXPENSIVEBEYOND REACH
How it was pressure-tested

Four AI personas, run as concurrent agents.

Each persona reviewed the work twice — a pre-read round that shaped the brief, and a reaction round against the finished screens. Independent lenses; the verdicts moved.

MK
Maya KesslerSupply-chain analyst
Not yetUse it daily
DR
Devin RaoSolver / platform engineer
No sign-offBoundary shown
PN
Priya NairSenior design leader
BorderlineInterview
RT
Ron TaseEnterprise buyer
BorderlinePilot
Work with me

This case study is the demo.

It’s a live demonstration of the AI Product Clarity Audit — the same lens applied to your product: where AI genuinely helps, where it’s risk theatre, and how design makes it trustworthy.