Over the past year, I’ve watched teams roll out increasingly capable AI systems, tooling, and agents, and then struggle to trust, adopt, or scale them. I’d argue that a lot of today’s AI adoption problem starts with how we are framing the shift.
“Human-in-the-loop” (often shortened to HITL) has become one of today’s most overhyped buzzwords. Companies and analysts repeat it earnestly to regulators, auditors, and risk teams as a compliance and assurance signal, shorthand for: “don’t worry, this system is not fully autonomous, there is a responsible person who can intervene and monitor.” HITL is also increasingly becoming a message of reassurance to customers and employees: “If you lean into using AI tools, don’t worry, “humans” like you will remain in the loop!”
This isn’t the first time this phrase has shown up. ‘Human in the Loop’ comes from engineering disciplines (aviation, nuclear systems, industrial control), where systems were increasingly automated. In 1998, the U.S. Department of Defense’s Modeling & Simulation Glossary used “human-in-the-loop” as a phrase to describe “an interactive model that requires human participation.”
The difference between that usage and today’s is subtle but important. In 1998, the DoD was describing tightly scoped, deterministic machine learning systems and automations designed to execute specific processes under controlled conditions. In classic control systems and early automation, the “loop” was some iteration of: sense, decide, act, observe, and then adjust. Machines would collect the signals (radar, gauges, telemetry) and people would then make sense of the data. In the 1980’s systems, people didn’t just intervene, they defined the goals, thresholds, and failure modes. However, today’s usage keeps the same label but describes a framework with less autonomy.
With the rise of LLMs and agentic AI, the loop has become something more along the lines of: the model generates, the person reviews for mistakes, and the agent proceeds.
The Framing Problem
Once you start to turn the phrase over in your mind, the framing is clearly wrong. Why are we calling it “human-in-the-loop” in the first place? The very structure of the phrase paints a picture of AI models doing the work with people invited in somewhere along the way.
This is a fundamental design problem: language that frames AI as the protagonist and relegates people to a supporting role as if they’re an accessory to the system, rather than the catalyst to the system itself. The structure of the phrase implies that AI is the primary actor running the operation, with the ‘human’ positioned as a control mechanism or quality assurance at the end of an automated assembly line.
In product and engineering, responsibility without authority is known as failure mode. And yet that’s exactly what the HITL framework implies: people approving outcomes they didn’t design. In this framework, models generate, systems proceed, and ‘humans’ are brought in to inspect, approve, and ultimately shoulder responsibility if something goes wrong. In any other context, we’d recognize this immediately as a flawed system—one that separates decision-making from accountability.
And then there’s the word ‘human’ itself: cold, sterile, biological, and impersonal. No wonder people tend to distrust these models—this phrase sounds like something a model would generate.
If HITL is the story we’re telling the market, then today’s AI adoption issues shouldn’t surprise us. If we want to fix the adoption problem, first we have to fix our framing.
The point is this: well-designed systems don’t avoid automation, but they do make delegation explicit. People set direction, define intent and constraints, and decide where judgement is required. Automation handles the rest. When that order is clear, AI is a powerful extension of human capacity. When it isn’t — and when systems advance work first and people are pulled in later to review and absorb risk—trust inevitably erodes. Not because automation is too capable, but because authority and responsibility have been misaligned.
The Uncanny Valley of Work
In a culture that prides itself on individual agency, creativity, and innovation, we’ve adopted a strangely passive way of describing how people are meant to interact with AI.
The narrative around “AI-enabled” tools is almost always the same: fewer human touchpoints and automation = more efficiency and speed. The implicit promise is that progress means less human involvement, because you only need the odd person “in the loop” to keep things from going completely off the rails.
I think this framing feeds directly into today’s distrust of these models, not because it always plays out this way, but because of the story it suggests. In this story, people worry about three things:
-
- What if I’m training the very systems that may (at worst) eventually replace me, or (at best) relegate me to a new role that feels less impactful or purposeful?
- What would this new role look like for me? Will I be expected to review, catch mistakes quickly, and approve outputs I didn’t create? Will my job shift away from creation and toward a monotonous cycle of reviewing and rubber-stamping?
- If something goes wrong, will I be held responsible or accountable?
Together, these anxieties produce what I think of as the uncanny valley of work: the feeling that this work looks like my work, the decisions resemble my judgment, everything feels familiar, and yet it still feels hollow because none of it is truly mine.
This framing also subverts the roles we typically play; traditionally, people create and technology supports. In this role reversal, AI generates and advances work, while people curate. In that position, it’s easy to feel indifferent to any outcomes, “I don’t know, the AI decided?”
People derive purpose from effort and achievement, so positioning them as reviewers ‘in the loop’ strips away that sense of meaning and ownership: a perfect recipe for burnout. After all, most people only tolerate administrative work when it supports meaningful or creative work, is time-bound, and has a clear purpose.
This is where the human-in-the-loop term fails; it positions people’s judgment as a step in the process, when our judgement is the entire foundation for success.
On the other hand, when we reverse that framing, suddenly people are the ones setting goals, choosing when to loop AI into the work, and shaping outputs. When thinking about AI implementation and adoption, we should position AI as what it already is: a power tool that can help people distill information, surface patterns, and reduce administrative work, and not something that replaces a person’s authorship or ownership.
Language as Architecture
Well-designed AI systems make delegation explicit. People should set direction, define constraints, and decide where judgement is required while automation handles the rest. In this model, AI expands what experts can do: surfacing patterns, reducing administrative work, and accelerating decisions without eroding authorship or responsibility.
When AI is reframed with a people-first mindset, it becomes empowering. I see this play out every day at Quickbase with our customers and our internal product teams. The organizations that succeed with AI adoption aren’t trying to remove people from the process; they’re giving domain experts better tools to work with their data, adapt in real time, and focus their energy where it has the most impact, especially in environments shaped by labor shortages, shifting supply chains, and tighter project budgets.
The reality of work is messy. Context matters, and our good judgement, experience, and creative problem solving aren’t nice-to-haves, they’re the core of how real work gets done.
If we want AI systems people trust, scale, and stand behind (which is the only way this works out well for everyone), we need to design them around a simple rule: People own the outcomes and AI supports the work. Not the other way around.
