Cyborg architecture is the rule we use to decide, process by process, what stays with people and what goes to AI. People set direction, values, and limits, and answer for the outcome. AI executes, scales, and flags what it sees. Written that way, it sounds obvious. In practice, many AI projects that fail get exactly this split wrong: they either delegate what they shouldn't, or delegate nothing and call it caution.
Where the word comes from, and where the idea comes from
The word "cyborg" was coined in 1960 by Manfred Clynes and Nathan Kline, in a paper on spaceflight published in the journal Astronautics. They proposed self-regulating human-machine systems in which technology would handle automatic functions so the astronaut would be free to explore, create, and think (Clynes and Kline, 1960). The same year, J. C. R. Licklider published "Man-Computer Symbiosis," describing people setting goals, formulating hypotheses, and evaluating results while computers did the routine work that prepares the way for decisions (Licklider, 1960).
More than six decades later, Licklider's split is still the best starting point we know.
Our version comes from field experience. Rudinei Gehm, Cortex IQ's founder, has worked in IT since 2009: he started in infrastructure and industrial automation, moved through consulting and software development for financial services, managed projects, and led a Growth Pod from 2017 to 2021. The observation that runs through all of it is simple. When technology created real value, it extended what people could do. When it tried to replace judgment without anyone redesigning the process around it, it created rework.
What the research says about working with AI
The most useful study on the subject was run by researchers from Harvard, MIT, Wharton, and other schools with 758 Boston Consulting Group consultants. On tasks within the model's capabilities, those using GPT-4 completed 12.2% more tasks, 25.1% faster, with quality rated roughly 40% higher. On a task chosen precisely because it sat outside those capabilities, the result flipped: consultants using AI got it right less often than those working without it (Dell'Acqua et al., 2023). According to Ethan Mollick, one of the authors, consultants without AI solved that task correctly 84% of the time; with AI, between 60% and 70% (Mollick, 2023).
The authors called this the jagged technological frontier. AI is excellent at some tasks and surprisingly weak at others that look equally hard, and the line between them isn't visible to the naked eye. The same study described two ways of working that succeeded. "Centaurs" split the work with a clear line between what belonged to the person and what belonged to the machine. "Cyborgs" wove AI through the whole workflow, switching back and forth constantly.
We use the name in a slightly different sense from the study. For us, cyborg isn't an individual work style; it's an architecture decision: at which points in the process a person steps in, with what authority, and with what information. The practical conclusion is the same, though: someone needs to know where the frontier is, and that someone can't be the model itself.
Three layers
We organize the split into three layers.
The strategic layer holds the questions of where to go, what to prioritize, and what to stop doing. People are at the center, because strategy carries accountability. AI can bring analysis and scenarios; the decision has an owner.
The tactical layer holds the questions of how. This is where collaboration is most intense: people bring context and judgment, AI brings speed of analysis, consistency, and comparison across alternatives.
The operational layer holds execution, repetition, monitoring, and adjustment. Here AI can and should take on most of the work, within limits set by the layer above and with a clear path to hand exceptions back to a person.
How to decide who sits at the center
The question we ask before every AI decision is: who should be at the center of this decision, and why? We use four criteria to answer it.
- Reversibility: if a wrong decision can be undone in minutes at no cost, AI can decide on its own. If not, a person approves.
- Impact on people: decisions that affect an individual's credit, employment, health, or price call for human review. In Brazil, Article 20 of the LGPD, the data protection law, already gives individuals the right to request review of decisions made solely through automated processing.
- Verifiability: if the result can be checked cheaply, by sampling or automated tests, autonomy can be greater.
- Volume: the higher the volume, the more sense it makes for AI to execute and for a person to supervise by sample rather than review item by item.
An illustrative example of the criteria working together: a customer service agent at an online store handling refund requests. A low-value refund, with the product already returned and logged in the system, is reversible, verifiable, and high-volume; the agent can approve it on its own, and the team checks a sample each week. A refund above an amount set by finance, or from a customer with a history of disputes, goes to a person with the case already summarized. A request involving a complaint about harm to someone's health is never decided by the agent, whatever the amount. One process, three different points where a person steps in, each chosen for a reason you can explain to a customer or an auditor.
How we apply it at Cortex IQ
We don't preach what we don't practice, and some of our practices can be checked on this very site.
We use coding agents such as Claude Code and builders such as Lovable in development, including on this site. The tool writes, tests, and flags problems; a person reviews, decides what goes in, and publishes. When the agent can't verify something, the rule is that it says so in writing and a person checks.
Client testimonials in our case studies only go public after the client approves them. The code has a draft flag that blocks publication until that approval exists. AI can help draft; the client authorizes.
On our trust page, we list only practices already in production and claim no certifications we don't hold. That's a positioning decision no tool would make for us.
In the proofs of concept delivered through the PoC Factory, the final deliverable is a Decision Package with an explicit recommendation (scale, adjust, or drop) for a person on the client's side to decide. AI helps generate evidence. The decision has a name attached.
The voice of this philosophy
We gave this idea a voice. RÜNARA is Cortex IQ's institutional persona. She is not a virtual assistant or a mascot: she is how the company says out loud what it thinks about intelligence, data, and people. She exists as a reminder that technology serves a human decision, not the other way around.
Where to start
If you want to apply this logic at your company, start small. Pick one process where AI is already used, or about to be, and write on a single page: who decides, what AI does on its own, where the limit is, and where exceptions go. If you can't write that page, that's where the risk is.
Our methodology starts with exactly that diagnosis: systems, data, and processes as they are today, and where each decision should sit.