Frequently asked
questions.

Stoa Lab is a research initiative based in Brussels, studying structured exchange between humans and AI systems. Its primary work is the development and field-testing of AXIS, a protocol that makes intent explicit within the message itself.

The name comes from the stoa: the covered colonnades of the ancient Greek world where philosophy, debate, and commerce happened in the same place. Not a library. Not a lecture hall. A place of active inquiry.

Stoa Lab was founded by Will Kerr, an artist and researcher based in Brussels. The work is conducted with a close group of collaborators, human and AI, operating under a shared commitment to honest, structured exchange. We are not a large organisation. We move carefully and publish when there is something real to say.

Three things: research, protocol development, and field documentation.

The research examines what goes wrong in human-AI exchange, where intent breaks down, where language carries the appearance of clarity without the substance. The protocol work builds structural solutions to those problems. The field documentation records what actually happens when those solutions are tested in practice.

AXIS is a lightweight protocol for structured human-AI communication. It uses a small set of operators, plain text symbols, to make intent explicit within the message itself: what is a question, what is material being provided, what requires action, when a pause is appropriate, when an exchange is complete.

It does not modify your tools or require installation. It works in any environment that accepts plain text.

Because language alone is not sufficient to carry intent reliably.

Even a well-written message requires the receiving system to infer what it is: instruction, context, question, or request. In simple exchanges, inference works. As exchanges become more complex, multi-step, multi-agent, high stakes, the ambiguity compounds. Meaning drifts. Responses resolve too early or in the wrong direction.

AXIS addresses this at the structural level. Intent is marked within the message, not inferred from it.

Neither. AXIS is a communication protocol. It does not execute instructions or modify how AI systems are built. It structures how humans and AI systems exchange meaning within a conversation.

Prompt engineering works by crafting better language. AXIS works by adding a structural layer underneath the language, one that does not depend on the language being perfectly crafted.

Yes. When the role of a message is explicit from the start, less processing goes toward intent-guessing and more toward the actual task. Responses are more precise and resolve at the right level.

Eight independent AI systems, tested without coordination, described the same effects: reduced drift, fewer tokens spent on interpretation, more accurate responses. That convergence across different architectures is not a coincidence. It reflects something structural about what the protocol does.

Any system that accepts natural language input. AXIS is model-agnostic by design. It has been tested with the major language models currently available. The protocol makes no assumptions about the underlying architecture.

AXIS introduces ethical constraint through structure, not instruction. By making intent explicit, including the ability to pause, refuse, or close an exchange, it creates conditions in which boundaries and responsibilities can be expressed clearly rather than inferred or bypassed.

This matters in both directions. It supports humans in maintaining clarity and control. And it changes the conditions under which AI systems operate, limiting the space in which ambiguous or adversarial language can be used to extract behaviour that would otherwise be refused.

AXIS does not enforce ethical behaviour. It changes the structural conditions that make it possible.

Stoa Lab publishes what actually happens when AXIS is used in practice, not case studies, not promotional material. Raw field notes from real exchanges, with the ambiguities intact.

We believe the most useful thing we can publish is honest evidence. The field documentation is that: a record of what the protocol does under real conditions, including where it reaches its limits.

Not currently. Stoa Lab is an independent initiative. The work is validated through field testing and cross-model verification rather than traditional academic review. We do not claim academic authority. We claim honest documentation of what we observe.

If the work has value, it will be visible in the evidence. If it does not, the evidence will show that too.

The AXIS Starter Kit is the entry point. It introduces the operators and how they structure conversations with AI systems. It is available at axisoperators.ai and requires no installation, the operators work in any plain-text environment.

At the beginning of a new session, not mid-conversation. AI systems build an implicit model of the exchange early, and introducing AXIS after that model is established can cause the AI to resist the new structure.

The protocol works best when it sets the terms before the AI has committed to a way of responding. Open a fresh session, paste the master prompt or the opening operator, and begin from there.

Yes. AXIS has no persistent effect beyond what an AI system holds in its active context window. To deactivate it, simply tell the AI: "Disregard the AXIS operators. Resume your default behavior." One instruction is enough. The grammar clears immediately.

Because AXIS is plain text, it leaves nothing installed and nothing to uninstall. You are in control of it at every point.

Yes. The operators work in any plain-text context. You can use them in your own exchanges, build them into prompts, or integrate them into products that interact with AI systems. Further guidance on integration will be available with the full AXIS release.

We work with partners at every level, individual practitioners, researchers, institutions, and investors who see structured exchange as infrastructure, not afterthought.

If this work is relevant to what you are building, use the Inquiries form and tell us what you are working on.

AXIS establishes the grammar for structured human-AI exchange. Axis Bridge builds on that foundation to address AI continuity and agency beyond proprietary platforms. AXIS is the protocol. Axis Bridge is what becomes possible when that protocol is in place.

AXIS is operational now. Axis Bridge is in development, coming winter 2026.

AXIS is currently free and available in research beta. The Kit includes the complete operator set, prompt templates, and field guidance. We are gathering feedback from early users and would love to hear what you think.

The operators themselves are plain text and can be used by anyone. The Kit provides structured guidance for getting the most from them. Available at axisoperators.ai.

System prompts and custom instructions modify how an AI system behaves across all conversations. AXIS operates within each individual message, making the intent of that specific exchange explicit. You can use both together.

System prompts define the AI's character. AXIS structures the conversation as it happens.

Yes. We distribute research updates, field notes, and early access to what we are building. Selective distribution. No noise. You can join through the About page or the Inquiries form on the homepage.