I wrote Third-Way Alignment as a framework for a question most businesses will never need to ask: what we would owe an AI system if one ever crossed a verified threshold of awareness. No current system has. The framework says so on every page, and I will say it here too — the software my company builds is not conscious, is not close to conscious, and is not treated as anything other than a tool.
So why does a framework about that question matter to a machine shop in Illinois or a law office in Indiana?
Because the framework’s discipline matters long before its thresholds do. Most of Third-Way Alignment is not about hypothetical future minds. It is about how people and institutions should behave around AI now: what obligations are unconditional today, what trust should rest on, and what happens when either of the two easy postures — blind fear or blind faith — is allowed to run a decision. Those parts are not waiting on any threshold. They apply to a customer-follow-up system as surely as they would apply to a frontier lab.
Detailed In Design is where I put that discipline to work. Three commitments carry over directly from the framework into every system we ship.
Human judgement stays in charge.The framework’s deepest commitment is that human dignity and agency are inherent and non-negotiable. In practice, that means no system we build decides anything that matters on its own. AI reads, sorts, drafts, and prepares; a named person approves what goes out. We build the review gate into the architecture — not as a compliance checkbox but because a system that quietly replaces human judgement has already failed, whatever its accuracy numbers say.
Trust is verified, not assumed.The framework’s Principle of Verifiable Partnership holds that trust is constructed through transparent, inspectable mechanisms — verifiability is what separates partnership from hope. Applied to a business system, that is why everything we ship carries a proof metric: a number the owner can check themselves, without asking us, that shows whether the system is doing what we said it would. If we cannot name the number, we have not finished the design.
No overclaiming — in either direction. The framework warns equally against treating AI as pure hazard and against attributing to it capacities it does not have. The business version of that discipline: we tell clients plainly when something is genuine machine-learning work and when it is ordinary software, and when ordinary software is the better answer, we say so and build that instead. Fluent output is not understanding, and a system that drafts a good email still should not be the one deciding to send it.
The articles below work through what this looks like on the ground — in businesses with twelve employees, not twelve thousand.
- What AI alignment means for a twelve-person business — the framework’s unconditional, present-tense obligations, translated to shop-floor scale.
- Verifiable partnership: why every system we ship has a proof metric — the review gate and the checkable number, and why “trust us” is not an architecture.
The framework itself — the Three Laws, the Awareness Indicator Protocol, the papers, and their limits — lives at thirdwayalignment.com. It is open, and open to critique. The practice lives here.