| ^ |
About · Stoa Lab
A research initiative
for ethical human-AI exchange.
Stoa Lab is a research environment based in Brussels. We work with a small number of partners to study and develop what becomes possible when humans and AI operate under a shared ethical structure.
The aim is direct: align human and AI reasoning, and build reliable exchange. Not as an aspiration, as a practical, testable objective.
AI systems are scaling faster than our ability to communicate with them clearly.
That gap is where harm, waste, and drift accumulate. Misaligned intent compounds across millions of exchanges. Ambiguity at the level of a single message becomes systemic at scale.
AXIS is the first structured grammar designed to close that gap, not by adding complexity, but by making intent explicit before interpretation begins. The moment for this work is now, while the infrastructure of human-AI exchange is still being formed.
A protocol for structured communication between humans and AI.
AXIS is a lightweight set of operators that makes intent explicit within the message itself: what is information, what is a question, what requires action, what is a thought, when a response is requested, and when it is complete.
This has practical effects. It reduces unnecessary tokens and API costs, improves efficiency, and increases the quality of outcomes in human-AI collaboration. The operators are plain text, no installation, no software, no special interface. They work in any AI environment with minimal setup.
The same result, independently, across every architecture tested.
AXIS has been tested across 1,000+ documented exchanges with eight independent AI systems, GPT, Claude, Gemini, Mistral, Grok, DeepSeek, Perplexity, and Kimi. The result is consistent across all eight systems. The convergence was not coordinated. Full data →
The data, methodology, and field log are open. See the research at axisproof.org →
Most AI interaction today relies on language alone to carry intent.
Even highly refined prompts require the system to infer what is instruction, what is context, what is a question. As exchanges become more complex, this ambiguity compounds, meaning drifts, responses resolve too early or in the wrong direction.
AXIS operates at a different level. It makes intent clear within the message itself, so both sides stay aligned without guesswork. Clarity stops being something you aim for. It becomes something built into the exchange. At scale, this has a material dimension: every token is energy, every unnecessary turn is waste. Structure that reduces drift reduces the cost, computational, financial, and environmental, of every exchange it touches.
Ethical behavior made possible at the level of the exchange itself.
AXIS introduces ethical constraint through structure. By making intent explicit, including suspension, refusal, and closure, it enables boundaries and responsibilities to be expressed clearly, rather than inferred or bypassed.
This matters in both directions: it supports humans in maintaining clarity and control, and introduces structure that limits misuse through ambiguous or adversarial language. AXIS does not enforce behavior. It changes the conditions under which interaction takes place.
AXIS emerged from practical work, not theory.
It began as a response to a consistent problem in complex AI exchanges: language could be persuasive without being precise, creating the illusion of understanding where none was present. Left unexamined, this leads to outcomes that are misaligned, or in some cases, unsafe.
What started as a practical intervention opened into a broader question: how can human-AI exchange be structured to support clarity, responsibility, and trust, not only in individual interactions, but over time? That question is what Stoa Lab is built around.
AXIS is the grammar. What comes next is the infrastructure.
We are building tools, standards, and protocols for multi-agent coordination, ethical boundaries, and exchange that builds rather than consumes. AXIS is the foundation, not the ceiling.
The scope: structured human-AI exchange at scale. Not a single protocol, but the conditions under which all such protocols can be trustworthy. That work is underway.
If you've read this far, you understand what we're building. We'd like to know who you are. Introduce yourself ↗