We Built RLHF. Now We’re Living In An Era Of RHLCF — If We Use It Right.

The Discussion Step

One corner of TraversalIQ prepares account executives for discovery — the first real meeting with a prospect, where a deal is quietly won or lost in the quality of the questions asked and the listening and interaction that follows.

Give the account executive (AE) a list of standard discovery questions and an AI tool, and the efficient thing to build is a button: click it, and the AI system researches the company and everyone attending, then hands back a clean, personalized list of discovery questions, tailored to exactly this account and people, at the speed of AI. The AE walks in with a perfect script, but has learned nothing because the system did the learning, and a script read aloud is not the same as understanding the room you are standing in.

Now remember what a discovery meeting actually is. It is not a questionnaire; it is a live conversation, where the value comes from following unexpected answers, reading the person who shifts in their seat, and asking the second question that no list anticipated. An AE working from a generated script can only ask question one.

So we built the prep step differently. Instead of generating a script, the AI explores the account with the AE. It lays out what it found.

• This company just reorganized around the exact problem we solve
• Someone in the room said something telling during last quarter’s earnings call
• The lead executive in the meeting posted this on social media yesterday

Then it turns the question back to the AE: Given all that, what are we really trying to learn in this meeting, and from whom? The AI could easily make the call and hand over a finished script, but at exactly that point we engineered a pause. The questions get shaped in the exchange, not handed down. The AE comes away having genuinely and curiously explored the account and walks in able to manage the conversation rather than read from a list. Smarter, not just faster. It is my hope that reps who prepare this way grow into better sellers over time, not just better-briefed ones.

It is a small piece of friction, placed on purpose, in exactly the spot where a person needs to learn. And it is the whole game because eventually the AE is in the room, and there is no AI at the table.

Turning RLHF Around

That small design choice is an instance of something larger — a loop the industry has named only half of.

The technique everyone talks about is RLHF, reinforcement learning from human feedback: the machine learns from us, a vast amount of human judgment poured in to make it useful. But there is a second human-computer loop, one we have to engineer on purpose — Reinforcement Human Learning from Computer Feedback (RHLCF), the mirror image of one of the industry’s favorite acronyms.

Wherever a person and an AI work together on a task, that loop is running whether carefully designed or not. Build the workflow like a button, and it teaches humans to be less intelligent. And this is what happens when you optimize only for speed. The reverse loop is always on. The only question is how you engineered it. And of course, you always want the loop to run both directions, so we also engineered the AI to learn from each human interaction and get smarter along the way.

Completion Is Not Competence

Learning science has a blunt way of explaining why the button fails: completion is not competence. Finishing the training is not the same as knowing the thing. A person can click through the training, feel confident, and still be confidently wrong — a state more expensive than plain ignorance. The one-click script manufactures exactly that: the sensation of a job done, with none of the understanding.

What the conversation adds is what psychologist Robert Bjork called a desirable difficulty — a small, deliberate effort placed precisely where the learning has to happen. Friction in the wrong place is just inefficiency, and AI should erase it without mercy: the drudgery, the formatting, the blank page. But friction in the right place is pedagogy. The craft of designing AI tools is learning to tell the two apart, to automate the drudgery and leave the desirable difficulty exactly where the human is meant to grow.

The Seductive Button

None of this happens on its own because the market pushes the other way. Every product roadmap wants one more click removed. Friction demos badly; the button always looks better on stage than the conversation does. The version of any tool that makes its users faster will win every quarterly financial report, and quietly lose the things that show up later. So the employers who end up with sharper people will be the ones who deliberately decline the most seductive version of their own tools, who leave the step in when everything commercial says take it out.

That is the choice, and it is being made right now, quietly, inside thousands of products and workflows, mostly by default and mostly toward the button.

The divide AI is opening between workers is not, in the end, a talent gap. It is a design gap between tools that do our thinking and tools that teach us to think better, while still stripping out low-value friction everywhere else, and between employers who are wise enough to know which one they’re handing their people. The good news is that it’s a choice, which means it can be made well.

In my book, Emergence, I explore how our twenty-watt brains unlocked intelligence beyond biology that is plugged into the power grid, and what happens next. There is a line in that book that I keep returning to as I write this article:

“There’s a certain poetry to it — the intelligence we are helping to emerge in silicon is now circling back to help amplify the minds that created it.”

— Emergence

That circling back is the reverse loop, and we can engineer it into the workplace. But it only amplifies us if we deploy it with that intention. We made RLHF to teach the machines; we are living now in the era in which these systems can be deployed to make us more intelligent in ways that were previously out of reach. That loop will run either way. Whether it lifts people or hollows them out is still ours to shape in the workplace.