Running at A2MAC1, the benchmarking platform of all major OEMs

Your senior specialist for every design review.

Meet your customers' quality expectations and hit your cost targets. Agent Q - Design and Perceived Quality scores craftsmanship the way your best people do, and weighs every issue by customer impression in Europe, the US and China.

  • Where quality must rise
  • Where cost can come out unnoticed
  • Why, for every issue

A top-3 Japanese OEM asked: can Agent Q think like our best people?

86%agreement with a panel of senior specialists.

Measured against the common judgement of the panel. Independent benchmark, 2026. 126 real cases, blind test.

Customer impression · interiorLessMore

1Instrument panel

Flush

The upper dashboard panel sits out of flush where it meets the A-pillar trim and the edges of the instrument panel, with visible steps.

Craftsmanship 5.3 / 10

Customer impression 27 / 50

Customer impression is an industry-wide objective benchmark. Shown here of 50, from 1,005 interior buyers in Germany and the US. Findings: real Agent Q results on a benchmark car, name hidden. Rendered car.
  • Today

    Agent Q - Design and Perceived QualityKnow what your best specialist would flag, on every car you review.

  • Next

    Agent C - Customer Acceptance IndexKnow what customers will accept, before you build it.

Partners and programmes

  • A2MAC1
  • Vinnova
  • NVIDIA Inception Program
  • Almi
  • VGR, Västra Götalandsregionen
  • AI Gothenburg
  • in collaboration with CampX by Volvo Group

Running today

Agent Q - Design and Perceived Quality

Agent Q runs at A2MAC1, our first licence: the benchmarking platform used by all major OEMs and 200+ suppliers. Serving giants like Stellantis and Audi. We are an official Volvo Group supplier.

From photos to answers in four moves.

Try Agent Q
q.intendedfuture.aiReplay of a real analysis
Front, Back and SidesExterior · Front, Back, Sides3 photosGlobal marketExport
  1. Looking at the photos
  2. Finding issues and drawing them
  3. Scoring the attributes
  4. What a customer would feel

Drop photos here

front, back, sides, interior: any angle, any light

  • A dark grey SUV's front fender and hood shut-line in a car park.
  • The lower rear bumper of a dark grey SUV with a silver trim strip.
  • The lower edge of a dark grey car door with its side molding.

3 photos

Three photos. Let me look.

Photos dropped.

Now try it on your own car. Drop the photos from your next design review and get the findings, the scores and customer impression for every issue.

Try Agent Q

A top-3 Japanese OEM asked:

Can Agent Q think like our best people?

86% agreement with a panel of senior specialists.

Measured against the common judgement of the panel. Independent benchmark, 2026. 126 real cases, blind test.

One shape = 1% of judgements.

It works like a second senior specialist, reviewing the same evidence from another perspective.

Three quality specialists lean over a car seat, checking the stitching and taking notes.

Your standards. Your customers.

Agent Q adapts to your discipline instead of imposing one. Set the rules once, then apply them the same way in every review.

  • Custom profiles

    Make Agent Q think like your quality team. Create instruction profiles for an organisation, a department or one person, with your criteria, your jargon and your language.

    Profile“Act as our exterior quality engineer. Apply our gap and flush standards. Flag anything a premium buyer would notice.”

  • Customer-weighted priorities

    Weight findings by customer segment, so you fix first what drivers notice first. US and DE datasets today, other markets on request.

  • You stay in control

    Everything on the Canvas is editable before export. Outputs are probabilistic, so treat them as decision support, the way you would treat a panel of expert reviewers.

Self-serve

Start in your browser today.

Instant, repeatable perceived-quality reports from Agent Q, with custom profiles and editable exports.

Try Agent Q

With our team

Start with our specialists.

PQ Audit, PQ Sprint or PQ Program. The same logic, run with people first.

See the engagements

What a Skill Model is

A Skill Model is an LLM trained on how the best human experts do one job.

Expertise, finally reproducible.

What Agent Q learns trains the first Skill Model, for Perceived Quality. We are building it now. It ships early 2027.

Layer 05 · License

The Skill Model

An LLM trained on how the best human experts do one job. The first, for Perceived Quality, ships early 2027.

Three steps from human expert to Skill Model.

  1. Our self-learning agents work next to human experts and take over tasks from day one.

    Layers 01 and 02

  2. They learn from every decision the human experts make, exactly like humans do. The experts get stronger. The agent grows.

    Layers 03 and 04

  3. What they learn trains the Skill Model. We are building it to work as software first, and inside a robot tomorrow.

    Layers 05 and 06

The Intended Future stack, built from the human expert up.

Agents in production today. The first Skill Model ships early 2027.

Today our agents run on what they have learned and a general model.

Working with our agents is like hiring a junior specialist who grows. Very fast.

Why

Humans landed on the Moon in 1969. Going back is hard.

The skill lived in humans, not in the papers.

An older man with grey curly hair looks up from his laptop. Behind him, a long table of people at work.

Same in every factory. Same in every design studio. The senior human expert knows. The documents do not.

  • The world is short of experts.
  • Ten years to grow one.
  • A six-figure salary to keep one, anywhere in the world.
  • You cannot hire them faster. You cannot copy them.

Same engine, any job

A new skill used to take 3 years. Now it takes seven days.

Our first skill took human researchers three years. Now our agents create a new skill in seven days. Training the model takes hours.

Skills built with the same engine
SkillBuilt byStatus
Perceived quality of carsHuman researchers, 3 years86% blind test*. In production at A2MAC1
Market research for product designersSelf-learning agent, 7 daysAgentic version launching now
Counter-drone watchSelf-learning agent, 7 daysSuccessful field trial at FMV Battleweek
Decision support for supply chain resilience and preparednessSelf-learning agent, 7 daysIn development, for the medical supply chain

* 86% agreement with a panel of senior specialists. Measured against the common judgement of the panel. Independent benchmark, 2026. 126 real cases, blind test.

Every customer makes the model better.

  1. More human experts make a better model.
  2. A better model wins more customers.
  3. More customers bring more human experts.

Perceived Quality and Agent C

Prefer to start with people, not agents? Begin with our team.

Perceived quality is how a customer's brain judges a car in context, using past experiences and expectations, before and while they use it. It is what eyes, ears and hands pick up in the real world, not what the CAD file promises.

  1. PQ Audit

    Quick check

  2. PQ Sprint

    2 to 3 weeks

  3. PQ Program

    Ongoing support

What we assess

Appearance and geometrical quality, joining, illumination, material and paint, sound, smell and solidity: our full set of sensory modalities. We score deviations from design intent and explain why they matter.

How we work

  1. Assessment. Structured attribute scoring across the vehicle.
  2. Prioritisation. With the optional Customer Acceptance Index (CAI).
  3. Action plan. Fixes, trade-offs and what really matters to the customer.

What you get

  • PQ Quadrants and Action List. What customers notice, against the effort to fix.
  • Attribute Scorecards. Clear criteria and target ranges.
  • Deviation Map. Where issues cluster, with likely causes.
  • Executive Summary and Data Package. Decisions you can defend and re-use.

Keep the data. When you are ready, the same logic powers Agent Q for instant, repeatable reports.

Schedule a demo

Data collection

Real customers, in your key markets.

Our Customer Acceptance Index™ quantifies customer acceptance for everything from take rates and trim levels to new design directions.

The PQI 500 couples craftsmanship analysis with customers' impressions and refreshes market data, typically around 1,000 respondents per market, so priorities stay current. You receive AI-ready data packages.

Agent C

Customer acceptance intelligence.

Predicts what customers will accept before you build it. Consumer studies in minutes instead of two weeks.

Agent C study titled Which design will people pick. Eighteen desk lamp designs are ranked by preference score, with the leading design shown and a 95% chance that it is the best of the set.
A study from the CAI app. Every shopper in it is a synthetic persona, not a real person.

Defense // MIMIR · Sovereign surveillance, edge-deployed

The system that learns an operator's craft, and takes it over.

MIMIR is a decision layer for sensor-based area surveillance and counter-UAS. It reads the sensor picture, reasons about what is in it, and presents ranked alarms, each carrying its evidence, its provenance and a recommended action. The operator decides. Then it learns from it.

  • Offline
  • Air-gapped
  • No cloud
  • No foreign dependency
The sensor tower on the range at Älvdalen.

Live tests // FMV Battleweek, Älvdalen, September 2026Three days live on a real sensor tower: FPV drones, ground vehicles and a Shahed-type fixed-wing drone, with MIMIR reading the sensor picture beside the operators.

Operator console // Friday 13:16, 2× speed · plan view · bearing waterfall · alerts · tower camera · ranked tracks

Handover // Five stages, statuses stated honestly

Autonomy is earned, not promised.

  1. 1

    WatchEvery prediction joined to the operator's judgement in a protected corpus; second opinions captured.

    Shipped
  2. 2

    MeasureIntra-operator consistency per class of contact. The profile is surfaced to the operator: their measure is theirs to see.

    Shipped
  3. 3

    ProposeRules drafted from counted behaviour, bounded to noise classes, staged for review.

    Shipped
  4. 4

    CalibrateDual Krippendorff α: agreement within and between operators; confidence calibration.

    Designed, not built
  5. 5

    HandoverA named class of judgement runs without a click, above an agreed threshold, reversible, audited.

    Not built. No such path exists in the code

Autonomous in judgement. Never in action. There is no weapons control and no autonomous outbound action: a structural invariant, enforced in the build system rather than by configuration.

The MIMIR edge unit, an industrial design concept, on dark rock.
One rugged edge computer // Industrial design concept
Hardware
NVIDIA Jetson AGX Thor
Network
None required for operation
Cloud
None. All inference is local
Install
Single container stack, one command
Rollback
Three commands. Decision history survives a downgrade
Shape
One box per watch position. Escalations shared across consoles

Designed in Gothenburg // Made in Sweden

Research and method

Science trusts a result only when it can be reproduced.

Expertise never met that bar. One senior knows. Nobody can repeat it.

We change that. Our agents study the best people of a profession until many of them agree on what good looks like. That agreement becomes the Skill Model. Once reproducible, it is unlimited. Every team can run it. Every robot can carry it.

Two older men graft vines on a terraced hillside by the sea. A humanoid robot works behind them.
  • Chalmers

    Ten years of research at Chalmers University of Technology and executive positions at global car OEMs. Our Perceived Quality framework is used by the majority of car OEMs.

  • Cognitive inference

    We infer how human experts think from what they do.

  • Yin and Krippendorff

    Our research agents follow Yin's Case Study Research Methodology and Krippendorff's Content Analysis Methodology.

  • SOΦIA

    The AI memory we invented. It keeps what matters and lets the rest fade. Patent applications in preparation.

Publications

Company

The Design Science Expert and the Data Expert.

Portrait of Dr. Kostas Stylidis

Dr. Kostas Stylidis

CEO and Co-founder · The Design Science Expert

10+ years research at Chalmers University of Technology. Ex-Director of Design Research at Geely Design Global. Shaped design strategy for Zeekr and Lynk&Co launches.

Portrait of Dr. Julia Orlovska

Dr. Julia Orlovska

CTO and Co-founder · The Data Expert

Built the research protocols the agents run. Developed ADAS systems at Volvo Cars. Senior Data Analyst at Sahlgrenska University Hospital. Published in top journals.

Both founders hold multiple grants from Vinnova, Sweden's Innovation Agency. Advisors from Volvo, Ford, SKF, Stena, and Avanza.

Company
Intended Future AB
Based in
Gothenburg, Sweden
Origin
Spun out of Chalmers University of Technology
Funding
Our projects are funded by Vinnova, Sweden's Innovation Agency.
Programmes
Member of the NVIDIA Inception Program. In collaboration with CampX by Volvo Group.

News

Latest from Intended Future

Future Insight on LinkedIn

News could not be loaded right now. Read our latest thinking in Future Insight.

FAQ

Questions we hear most.

Anything else? Write to agentq@intendedfuture.ai.

What is a Skill Model?

An LLM trained on how the best human experts do one job. Our self-learning agents work next to human experts and learn from every decision they make. What they learn trains the Skill Model.

Can I license a Skill Model today?

Not yet. We are building them. The first Skill Model, for Perceived Quality, ships early 2027. Agent Q is in production today, and what it learns trains that Skill Model.

What is Agent Q?

Agent Q - Design and Perceived Quality is a self-learning agent. It looks at photos of a car and judges each detail the way an experienced specialist would: craftsmanship scores, customer impression and two fixes for every finding. It runs at A2MAC1.

How do I start?

Self-serve at q.intendedfuture.ai, or with our team through a PQ Audit, Sprint or Program. For a pilot, tell us your domain, timing and goals. We typically scope a 4 to 6 week MVP that combines your data with our agents.

Can Agent Q run on our own infrastructure?

Yes. We can deploy Agent Q in a private VPC or on-premises, with your identity provider and logging requirements.

Will it always give the same answer?

Not always. Agent Q uses vision-language models, which are probabilistic, so results can vary slightly between runs, images or model versions. We reduce variance with fixed settings and consistent image guidance. Treat its output as decision support, the way you would treat a panel of expert reviewers.

Where are you based?

Intended Future AB is headquartered in Gothenburg, Sweden, with collaborators across Europe and North America.

Agent Q - Design and Perceived Quality

Bring your senior specialist to the next design review.

Start in your browser today, or let our specialists run the first review with you.

  • Runs in your browser
  • Your standards, your customers
  • Editable exports

Contact

Tell us about the job your best people do.

Start with Agent Q, start with our team, or talk to us about the next Skill Model.

agentq@intendedfuture.ai

Intended Future AB · Gothenburg, Sweden