Reskilling for AI - The Erdos methodology for AI mentoring based on Identity shift - free download paper

Abstract

This paper introduces the Erdos Methodology for AI Reskilling, a professional transformation framework that reframes AI learning as an identity shift rather than a skills-acquisition problem.

It argues that in the age of adaptive AI systems, experience is no longer accumulated through time or exposure but through epistemic authority over how intelligence is specified, evaluated, and interpreted when it fails.

We propose a unifying Experience Spine — Spec → Eval → Failure → Reflection → Redesign — as the substrate through which genuine AI experience is manufactured.

The framework formalises this epistemic loop across multiple professional personas including domain experts, software developers, AI engineers, AI product managers, forward-deployed engineers, and single- and multi-agent system architects.

For each role, it maps a concrete identity transformation from “using intelligence” to “authoring intelligence,” supported by detailed master skill taxonomies covering specification authority, evaluation authority, failure interpretation, reflection capability, and redesign mastery. These taxonomies span AI engineering, AI-assisted development, single-agent systems, multi-agent organisations, AI product management, and field-embedded engineering practice.

Finally, the paper uses the example of HAL from 2001: A Space Odyssey to illustrate the catastrophic consequences of epistemic collapse in AI systems that lack mechanisms for reflection and redesign. It concludes that sustainable AI careers and safe AI systems require reclaiming the learning loop itself: professionals must become stewards of how systems learn, not merely operators of model outputs.

If these ideas resonate with you, we invite you to explore joining Erdos Research

1 - Background

1.1 Introduction

I created this document based on a chance conversation.

At a recent event - I was asked (the now familiar question) - How do I know my job is at risk from AI?

I gave a slightly cynical answer i.e. If you are sitting in front of a computer - your job as you understand it now is at risk.

I used the analogy of HAL from Space Odyssey 2001 - i.e. the AI watches you and can understand and react and learn from your behaviour - ultimately leading to the famous quote “I’m sorry, Dave. I’m afraid I can’t do that.”

I don’t think that’s what my audience wanted to hear!

But that was the reflection behind this work (I expand on the HAL idea at the end of this document)

I see other related advice

Learn soft skills - but not exactly sure how that helps to differentiate yourselves since hard skills pay

Then there is more daft advice - become a hairdresser or (the worst) - don’t learn coding

The flawed concept appears to be - learn something that AI will not replace - Instead of learn how to work with AI to perform more complex tasks that AI cannot easily perform (again the HAL analogy at the end will expand more on this)

Granted that this is a moving goal post - and requires you to reskill - but I believe that AI has gaps - and that there will be plenty of jobs that will arise due to this - where you need higher order thinking.

1.2 Reskilling for domain experts

Over the years of teaching AI, I have spoken with many students - and learned from their AI transformation journeys. We have some very notable examples of success in AI with Magnus and Dr Amit which you can see HERE

The biggest opportunity I see is the ability for non developers (domain experts) to reskill to AI.

Some notes

1. Reskilling involves an Identity shift as we describe below

2. Although this document is described in abstract terms (epistemology) and even science fiction (HAL) - we focus on specific tools to build products ex Gemini, Cursor AntiGravity Lovable and others

3. We also focus on specific job roles ie AI Engineer, forward deployed engineer, AI Product manager, Spec driven development (low code data scientist)

4. The views expressed here are my own (and Erdos Research) but not any other institution I am associated with

5. If these ideas resonate with you, we invite you to explore joining Erdos Research

This is a long paper - best read as a pdf which you can download free HERE