The Human-AI Unbroken Loop
The dual-feedback loop between human and machine, and why it changes everything about the boundaries of thinking and of responsibility.
AI is not a tool we wield. It is a loop we participate in.
For twenty-five years, I sat on the vendor side of enterprise technology decisions. I watched buyers approve systems they could not later defend. I watched executives sign off on deployments whose outcomes belonged to no one. None of those decisions felt small in the room. All of them did, six months later, when the question came back.
That distinction — tool versus loop — is important. It manifests the new power that knowledge brings to humans, who in turn must use it responsibly in decision making. In real life, when in business the output is wrong, there must be someone who decides, carries the impact of such decisions, and defends them in front of a board.
For two years now, the dominant framing in serious commentary has been that AI is a powerful general-purpose technology — like other breakthrough technologies that each brought their own social and scientific value in human history.
I disagree.
I want to be clear about what this essay is and what it is not. The dual-feedback loop is not a deployment pattern I can show in production. It is a design proposal. What I can show is twenty-five years of deployments where its absence was the cause of the failure. The argument runs the other way. We do not yet have a loop. We have transactions wrapped in tools, and we are surprised when they break.
AI impacts every aspect of human life and business. Daily practices change at scale. A technology that reorders how lawyers research, how analysts model, how procurement teams write requests, how executives draft strategy, is by definition disruptive — it produces a discontinuity in a way that does not allow humans to step into it in a gradient manner.
Someone can fairly say that in the past there were other disruptive technologies that contributed to the evolution of humankind. That is partially correct. What makes AI different is that it is the first technology that requires both sides of human judgement to produce a usable output. Critical thinking, to evaluate what the machine returns. Emotional intelligence, to decide whether the answer should be acted on, softened, or refused. Earlier disruptive technologies could be wielded with one hand. AI cannot.
That is why the model of the human wields the tool breaks here. The cycle is not one way. The human feeds the machine. The machine produces output. The output feeds back into the human’s next input. The human’s quality of input shapes the machine’s quality of output. The machine’s quality of output shapes the human’s next decision. Around again.
I call this the feedback loop. This is where the value is created, where accountability is shaped, where consequences become evident.
The operative question changes shape. The question is no longer is the AI good? The question is is the loop good?
And a loop has structure. There are four humans inside it.
The engineer builds the model. The buyer scopes and procures the system. The executive funds and approves the deployment. The user chooses to follow the output or override it. Four humans, four decisions, each one logged somewhere. Each one can be defended or it cannot.
Policy mechanisms will arrive. Regulation will arrive. Institutional norms will arrive. The real question is not whether they will arrive. The real question is whether they will be respected once they do.
Who respects them? Humans. Specific humans, with names, signing specific approvals, drawing on the intelligence built at every point in the loop.
There is no clean handoff point where human responsibility ends and machine responsibility begins. What flows in shapes what flows back. What flows back shapes what flows in. The two sides redesign each other in motion. The loop is the place where human judgement and machine output have to balance — not after the fact, in the contract clause or the regulatory filing, but in the moment, every time.
If I now apply the feedback loop concept to the business operating model, four commercial documents change.
Vendor selection stops being a capability comparison. It becomes a loop-quality comparison: who designs for input discipline, who designs for output traceability, who designs for the human’s next decision rather than for the machine’s next answer.
Procurement design stops being a specification of features. It becomes a specification of the participation contract: who feeds what, who reviews what, who owns the loop end-to-end.
Governance architecture stops being a compliance overlay. It becomes a description of who sits in the loop and who does not, with named owners at each touchpoint.
Pricing models stop being a function of usage. They become a function of outcome under the loop — because output quality is now jointly produced, not vendor-produced. A vendor who prices as if the output is theirs alone is mispricing the relationship.
Here is what the absence of the loop looks like in practice.
In a Tier-1 telco deployment of a cutting-edge technology, vendor and buyer specified the architecture, the technical specs, the SLA, the customisations for integration, the audit cadence, and the price. The contract did not specify who on each side owned the input quality. It did not specify who on each side owned the output validation. It did not specify who on either side owned the loop end-to-end.
Six months in, when the deployment underperformed, the impact became a ping-pong table for every party involved. Blame moved toward the vendor. The vendor pushed responsibility back to the system integrator and the buyer. The steering committee turned into a battlefield of responsibility allocation, with no decision being made, while middle managers on the buyer side ran their own internal game.
The project took three years to implement. It cost the vendor brand reputation. It cost the buyer revenue. It exposed the sponsors of the programme on both sides. Everyone involved was responsible. None of them was accountable.
That is what a tool wrapped in a contract looks like, six months after signing.
The loop is the asset. The loop is the power. The loop is the safety net.
Named Risks » Funded Response » Defensible Decision
Commercial judgment from the inside
.