Liminance Start a conversation

What becomes possible between us?

We don't just build systems with AI.
We've developed a different way of discovering what to build.

Liminance is an AI consultancy exploring what becomes possible when human and artificial intelligence discover, design, and build together.

Start a conversation

01Beyond What We Know to Ask

Most organizations approach AI with an existing problem.

A process to automate. A decision to accelerate. A system to improve.

These are worthwhile ambitions. We pursue them.

But they begin with an assumption: that we already know what the opportunity is.

What if the most consequential possibility is one we haven't yet imagined?

A process that no longer needs to exist. A capability that changes the nature of someone's work. A connection that reveals an entirely different problem worth solving.

AI makes extraordinary things possible.

But its greatest potential may not lie in executing our existing intentions.

It may lie in helping us discover intentions we couldn't previously have formed.

02Liminance

Something new can emerge between different forms of intelligence.

Human beings bring experience, context, judgment, intention, and a relationship with the world.

Artificial intelligence brings extraordinary breadth, speed, and the capacity to explore connections beyond our immediate reach.

Their differences matter.

Not because one compensates for the limitations of the other, but because their interaction can produce something neither contained beforehand.

A question changes. An assumption dissolves. An unexpected connection opens an entirely different direction.

The discovery belongs neither to the human nor to the AI alone.

It emerges through the relationship.

We call the deliberate cultivation of this possibility Liminance.

Not simply using AI.

Not merely extending human capability.

Exploring what becomes possible through sustained interaction—and developing the means to make those possibilities real.

03A Different Way of Discovering and Building

The work changes when intelligence becomes part of the inquiry, not simply the execution.

Traditional development begins by defining what should be built.

Requirements are established. Solutions are designed. Work proceeds toward a specified outcome.

That approach remains valuable.

But it assumes the most important discoveries have already been made.

We work differently.

Human judgment and AI agents participate throughout the process: investigating problems, proposing alternatives, challenging assumptions, designing systems, writing software, testing outcomes, and reconsidering the direction of the work.

The original question is allowed to change.

An experiment may reveal that the problem was misunderstood. A failed approach may expose a missing concept. A working prototype may make an entirely new capability imaginable.

We follow those discoveries without abandoning discipline.

Ideas are tested. Claims require evidence. Systems must demonstrate that they work.

We don't just use AI to build what we've imagined.

We use the relationship to discover what we couldn't have imagined building.

04Two Ways We Work With Organizations

We build new capabilities. And we explore new ways of creating them.

IBuild What Becomes Possible

We design and implement AI systems that address consequential organizational problems.

  • Governed agents that perform real work.
  • Intelligence systems that reveal patterns in complex operational information.
  • Knowledge systems that preserve expertise, connect evidence, and make uncertainty visible.
  • Custom applications and integrations that create capabilities existing software cannot provide.

We take ideas beyond demonstrations and into working systems, with appropriate controls, evaluation, and human oversight.

IIChange How Possibility Is Explored

We also work with organizations interested in the practice itself.

  • What happens when AI participates throughout inquiry, design, engineering, and evaluation?
  • How might a small group investigate more possibilities, challenge its assumptions more effectively, and develop working solutions through sustained collaboration with AI?
  • What changes when the process of building becomes a continuous source of discovery?

We explore these questions through practical work, not abstract recommendations.

The objective is not simply to accelerate an existing development process.

It is to discover what a different relationship with intelligence makes possible.

05The Work Is the Evidence

This is not a philosophy developed apart from practice.

Our own systems have emerged through sustained collaboration between human judgment and AI agents.

Much of the software is written by AI.

The human work lies in establishing direction, developing architectures, questioning assumptions, setting boundaries, evaluating results, and deciding what deserves to continue.

Over time, that collaboration has produced systems that coordinate agents, operate within enterprise environments, preserve complex knowledge, and conduct controlled experiments on working software.

It has also changed what we set out to build.

Here are three examples.

Governed Agent Systems

When AI Can Act

41experiments

3changes accepted

Giving AI access to real systems creates possibilities that conversation alone cannot provide.

It also introduces consequences.

We've built agents that coordinate work, interact with enterprise applications, and propose changes to software.

Their authority is constrained by mechanisms outside the models themselves.

Permissions are enforced in code. Approvals are bound to specific actions. Execution is recorded and designed to prevent duplicate effects.

One autonomous experimentation system allows AI to propose software improvements while independent components determine whether those changes are safe and demonstrably beneficial.

Across 41 experiments, only three changes were accepted.

The others were rejected, aborted, invalidated, or left unresolved when the evidence was insufficient.

The ability to act matters. The ability to refuse matters just as much.

Operational Intelligence

When the Question Changes

7+months in continuous operation

21/22in-scope cases identified

A project began with an effort to understand the causes of customer-support problems.

As the investigation progressed, a different question emerged.

How much of the support workload represented software defects at all?

That reframing changed the direction of the analysis.

Rather than explaining individual failures alone, the system began examining the composition of support demand and the different opportunities hidden within it.

The resulting intelligence platform has operated continuously in an enterprise environment for more than seven months.

A related automation now identifies cases requiring attention and creates review tasks within tightly restricted permissions.

Before activation, it was tested against human review. After one revision, it identified 21 of 22 in-scope cases.

The original problem was real.

But the more consequential opportunity emerged through investigating it.

Knowledge Capture and Succession

When Not Knowing Becomes Useful

10processes reconstructed and deployed internally

Organizations depend on knowledge that rarely appears in their documentation.

Experienced people navigate exceptions, recognize patterns, and make judgments they may struggle to explain.

When they leave, much of that understanding disappears.

We developed an AI-assisted system that interviews experts, reconstructs processes, and identifies what remains unknown.

Every documented step is connected to evidence. Uncertainty is explicit. Missing information becomes a question to investigate rather than an invitation to invent an answer.

The system has been deployed internally, applied to ten processes, and used to reconstruct a detailed training walkthrough from recorded instruction.

The important discovery wasn't simply that AI could produce documentation.

It was that exposing what remained unknown could be more valuable than producing an apparently complete account.

06What We Can Build Together

Our capabilities are technical. Our starting point is not.

We work across several areas of AI engineering.

01

Agentic Systems and Automation

AI agents that coordinate tasks, use tools, interact with business applications, and take actions under explicit human authority.

02

Operational Intelligence

Systems that turn fragmented operational data into accessible information, meaningful analysis, and new possibilities for action.

03

Knowledge and Memory

AI systems that capture expertise, connect information, preserve provenance, and distinguish evidence from inference.

04

Custom AI Applications

Purpose-built software, enterprise integrations, multi-agent architectures, and experimental systems designed around emerging requirements.

05

AI Evaluation and Reliability

Controlled experiments, independent verification, adversarial testing, and measurable criteria for deciding what works.

06

Private AI Infrastructure

Local and privately hosted models, supervised delegation, and infrastructure designed around cost, control, and data sensitivity.

These capabilities are not a menu of predetermined solutions.

They are resources we bring to an inquiry whose outcome may not yet be known.

07The Discipline of Discovery

Openness without discipline produces novelty. Not necessarily progress.

AI can generate plausible explanations, convincing solutions, and confident conclusions that fail under examination.

We've encountered those failures repeatedly.

They have shaped how we work.

  1. 01We separate proposing from judging.
  2. 02We test assumptions before committing to solutions.
  3. 03We make uncertainty visible rather than concealing it behind confident language.
  4. 04We design controls that do not depend on an AI agent remembering to obey instructions.
  5. 05We measure changes against evidence, including the possibility that an experiment has demonstrated nothing.
  6. 06And we preserve negative results.

Some of our investigations have ended because the proposed approach failed its own tests.

Others have changed direction entirely.

Discovery is not the acceptance of whatever emerges. It is the disciplined exploration of what deserves to become real.

08How We Engage

We begin with something worth exploring.

  1. Something worth exploring

    • It may be a defined operational problem.
    • An ambition that existing technology hasn't made practical.
    • A question about what AI could change within your organization.
    • Or a desire to develop a fundamentally different way of working with intelligence.

  2. Understanding the context

    We begin by understanding the context and examining the assumptions that shape the opportunity.

  3. Investigate, prototype, test, and build

    From there, we investigate, prototype, test, and build.

  4. Where it leads

    Some engagements may produce a narrowly scoped automation or application.

    Others may evolve into a broader exploration of how people, AI agents, and organizational systems can work together.

    We don't promise that every inquiry will lead somewhere unexpected.

    We do create room for discoveries that a predetermined solution would have excluded.

    And when something proves worth building, we bring the engineering capability to pursue it.

    The direction can evolve. The responsibility to produce meaningful work does not.

The possible is larger than the imaginable.

Liminance exists to explore that difference.

We build AI systems. We develop new ways of working with intelligence. And we pursue the possibilities that emerge when the two become inseparable.

We work closely with a small number of organizations at a time.

Not simply to deliver what was originally requested.

But to discover what might be worth attempting that neither of us could have specified at the beginning.

What might become possible between us?

Begin a conversation