The AI Ladder of Inference Coach skill for Claude

The problem the skill solves sits in a single word: must. He must be fed up with it. They must be going to a competitor. She must think it is not good enough. Sentences like this feel like observations, but they are conclusions, and that is exactly where the danger lies: the higher you stand on the ladder, the stronger it feels as if you are simply describing the facts, while you have already selected, interpreted and concluded. That belief is then followed by an action, a sharp email, a conversation that opens the wrong way, a deal you write off, and that action is sometimes harder to reverse than the thought was.

It is written for anyone who works with people: the team leader who has to give feedback, the salesperson interpreting a silent client, the colleague who feels a conflict coming, the coach who wants to help someone else reason things through. It is part of the library of 100 free skills we make available through the AI and automation service, with no account and no sales email afterwards.

One thing you should know beforehand: the skill does not tell you who is right, and it will not tell you that you are right just because that reads more pleasantly. Both rules are stated literally in the file. What you get is sharper: the point where your reasoning departs from the facts, and the question you can use to test that in the real conversation.

Add to that the reflexive loop, and that is what really sharpens the model. Your beliefs on rung 6 determine which data you select on rung 2 the next time round. Once you believe the colleague is fed up, you see new proof in every meeting: silent again, no reaction again. That the colleague simply does their work, nods and chats afterwards falls outside the selection. The ladder reinforces itself, and that is why two people can look at the same conversation and both be certain they are right.

In a work context that is costly. A team leader who lives on rung 6 gives feedback on their own interpretation instead of on behaviour. A salesperson who reads silence as rejection lowers a price nobody thought was too high. And a conflict between two colleagues is usually not a clash between two characters but between two ladders that are each correct within their own selection.

You can only interrupt that mechanism at one point, and that is where the skill steps in: it forces the reasoning back down. First name the intended action, then descend rung by rung until only the facts remain, and there ask the question: which data did I leave out, and where is the biggest jump? The file records this as a fixed action: it points out the transition between two rungs that has the least support underneath it.

In diff form, with lines from the second example in the skill: what the same thought looks like before and after descending.

Next comes the method: seven fixed actions the skill carries out in every analysis. It first summarises the situation and presents that summary, so the analysis is about the right incident. It reconstructs the full ladder across all seven rungs, noted from rung 7 back to rung 1. It strictly separates what could actually be observed from what you selected out of it, and names which data was left out. It points out the biggest jump in the reasoning. It builds at least two alternative ladders on the same available data. It formulates an intervention question per rung. And it closes with a recommended action that comes before your planned action: what you check before you act on your current conclusion.

Before anything is analysed, the skill runs through a short mandatory input list of five points: what actually happened, as concretely as possible, the conclusion in your own words, what you intend to do, the relationship with the other person, and earlier incidents that play a role. It only asks about what is missing, in one go, and then carries on working with what is there.

The output always has the same six blocks, in the same order: your situation, the ladder across all seven rungs, the blind spots, the alternative ladders B and C, the intervention questions per rung, and the recommended action. The file also contains two fully worked out examples, the silent colleague from FIG.01 and the client who does not respond, and strict writing rules: businesslike, with no judgement of people, rungs always given with their number so you can refer back to them. If you want to learn yourself how to build such an instruction file, the full explanation is in writing SKILL.md NL.

The model comes from Chris Argyris. The organisational psychologist described in Overcoming Organizational Defenses, from 1990, how professionals reason themselves into a corner: not through stupidity, but by climbing from observation to judgement at lightning speed and invisibly. His key point sits in the first two rungs: before you even interpret, you have already selected, and that selection does not feel like a choice.

Peter Senge made the model well known. In The Fifth Discipline, from 1990, in the chapter on mental models, he uses the ladder to show why teams talk past each other: everyone presents the top of their own ladder as fact. His remedy is the same as the skill's: make your reasoning open for discussion by showing it rung by rung, so someone else can respond to it instead of colliding with it.

The well known diagram comes from Rick Ross. In The Fifth Discipline Fieldbook, from 1994, he worked the ladder out into the drawing you see in almost every training course, with practical conversation questions alongside it. The skill's intervention questions stand in that tradition: not an accusation, but a question that opens up a specific rung.

With that, the ladder touches on a family of thinking tools that we also offer as a skill. Anyone who wants to learn to ask the questions themselves, instead of having them formulated, trains that with the Socratic Method Coach NL. And anyone who wants to know which other systematic thinking errors can creep into a reasoning process alongside the self reinforcing ladder can put the same case to the Cognitive Bias Detector.

Even without installing the skill you can use the model today: write down your conclusion, write underneath it what actually happened, and see how many rungs lie between them. Usually there are more than you thought.

After that, using it is simple: describe what happened and which conclusion you drew, as concretely as possible. Whatever is missing from its five point input list, it asks about itself, in one go, and then carries on working with what is there. If you are new to this kind of file and want to understand more broadly how working with AI works, you will find the wider explanation in the knowledge base.

There are also limits, and the file draws them itself. This is not a substitute for the real conversation with the person it is about: the intervention questions are meant to be asked, not to skip the conversation. It gives no psychological or medical interpretation, not of you and not of the other person; if more than a reasoning question is at play, that belongs with someone qualified for it. And it does not start without a concrete situation: anyone who only supplies a feeling without who, what, where and when first gets asked for the facts.

The most honest limit is this: descending the ladder takes more time than climbing it, and that stays true even with a skill that does the groundwork. The analysis appears quickly, but the conversation in which you actually ask an intervention question is one you have to have yourself. What the skill changes is not that the difficult conversation disappears, but that you open it at rung 2 instead of at rung 6, and that makes exactly the difference between an accusation and a question.

The ladder also applies to your own business. We are just not being found, this is not working for us: that is rung 5, not rung 1. You get the facts with the free SEO scan, which shows in a few seconds where your site stands. And if you want to talk further about what AI can do for your business beyond that, from separate skills to full automation, we simply do that in a conversation.