Artificial Idea | AI careers · practical prompts · no hype Monday, December 22, 2025 · Issue #41 · Jobs

The new job map

Every technology transition creates roles that did not previously exist. The interesting question is not whether those roles exist. It is whether you are positioned for them before the job titles are written.

There is a specific moment in any technology transition when the new roles it has created become visible in job postings, salary surveys, and professional association frameworks. That moment is not when the roles begin to exist. It is when enough organisations have created enough of them that the labour market infrastructure catches up and codifies what has been happening informally.

That codification moment for AI-created roles arrived in late 2025. The roles described in this issue appear in increasing numbers in current job postings. Their skill requirements are becoming standardised. Their compensation bands are establishing precedents. The professionals who understand what these roles require and who have been developing the relevant capabilities, in many cases without knowing these specific roles were emerging, are positioned for a labour market moment that the professionals who learn about these roles only from their job titles will arrive at later.

Late is not disqualifying. It is a different starting position with a different amount of ground to cover. Understanding the roles clearly is the first step regardless of where you are starting from.

Role 1: AI output evaluator

What it is: a professional whose primary function is assessing the quality, accuracy, and appropriateness of AI-generated outputs before those outputs are used in consequential professional contexts.

The role exists because AI tools are being deployed at a scale and in contexts where the cost of a wrong output is high enough to require systematic human review, but where the volume of outputs is too large for domain experts to review everything. The AI output evaluator sits at the intersection of domain knowledge and AI literacy, applying the former to evaluate outputs produced by the latter.

The most common current instantiation of this role is in legal, financial, and healthcare contexts where AI tools are generating first-pass analysis, draft documentation, or preliminary recommendations that require expert human review before use. But the role is emerging in every professional function where AI output volume has outpaced the capacity of the domain experts who commissioned it to review it personally.

The skill requirements are the ones that appeared most frequently in the high-premium job postings in the Burning Glass data from Issue #33: domain expertise at a level sufficient to identify errors that are not obvious to non-specialists, familiarity with the specific failure modes of the AI tools being evaluated, and the communication skills to document and explain evaluation decisions in ways that create an auditable record.

What it pays: the Burning Glass data shows a median compensation premium of 34% over equivalent roles without the AI evaluation component, with the premium highest in legal and financial services where the consequence of missed errors is most acute.

What it requires from you now: the deliberate practice of evaluating AI outputs critically rather than accepting them, applied to real outputs in your actual professional domain. Every professional who uses AI tools and reviews their outputs with genuine critical attention is developing this capability. The professionals turning it into a career position are those who have made the evaluation systematic, documented it, and made their methodology visible to the organisations that need it.

Role 2: AI workflow architect

What it is: a professional who designs the processes through which AI tools are integrated into existing organisational workflows, including the decision points at which human judgment is retained, the handoff protocols between AI-generated outputs and human review, and the monitoring systems that flag when the integrated workflow is producing unreliable results.

The role exists because AI tool adoption at the individual level and AI integration at the organisational level are different problems requiring different skills. Most organisations have reached a point where individual employees are using AI tools effectively in their personal workflows while the organisational infrastructure around those tools remains improvised, undocumented, and insufficiently monitored. The AI workflow architect brings systematic design to that infrastructure.

This is not a technical role in the software engineering sense. It is a process design role that requires understanding both how AI tools work at a conceptual level and how organisations work at an operational level. The professionals most naturally positioned for it are those with backgrounds in operations, consulting, project management, or business analysis who have developed genuine AI fluency, because they bring the process design expertise that the role requires alongside the AI understanding that makes the design sound.

What it pays: median premium of 28% in the current job posting data, with the premium growing faster than any other role in this list as organisations recognise the cost of undesigned AI integration.

What it requires from you now: the habit of thinking about AI tool usage at the process level rather than the task level. Not just how to use a specific tool for a specific task, but how the use of that tool integrates with the tasks before and after it, where the human judgment points should be, and how the workflow would need to change if the tool's output quality degraded.

Role 3: AI communications specialist

What it is: a professional whose function is translating between the technical realities of AI systems and the non-technical stakeholders, regulators, clients, and publics who need to understand those systems well enough to make decisions about them.

The role exists because the gap between what AI systems actually do and what the people making decisions about them believe they do is consequentially large in most organisations. The AI communications specialist closes that gap through writing, presentation, and dialogue that is technically accurate without being technically impenetrable.

The skill profile is unusual in that it requires genuine technical literacy without requiring technical expertise. The AI communications specialist does not build models. They understand models well enough to explain them accurately, including their limitations, their failure modes, and the conditions under which their outputs should and should not be trusted. Combined with the communication skills to make that explanation accessible to audiences with no technical background, this profile is rarer than either component alone.

What it pays: median premium of 31%, with the highest premiums in regulated sectors where communication with regulators and auditors about AI systems is both mandatory and consequential.

What it requires from you now: the practice of explaining AI tools and their outputs to non-technical colleagues in ways that are accurate rather than reassuring. Every professional who has explained how a language model works, why a recommendation algorithm produced a specific result, or why an AI output should or should not be trusted to someone without a technical background has been developing this skill. The question is whether they have been developing it with the rigour and documentation that turns it into a professional capability.

Role 4: AI ethics and governance professional

What it is: a professional responsible for ensuring that an organisation's AI deployments comply with applicable regulatory requirements, meet internal ethical standards, and are documented in ways that create an auditable record of how consequential AI-assisted decisions were made.

The role exists because the governance gap described in Issue #37 is real and growing. The EU AI Act, India's proposed AI governance framework, and the emerging US federal AI guidelines are all creating mandatory requirements that organisations must meet, and the professionals with the knowledge to meet them are in short supply.

This is the role this newsletter has identified as the most underappreciated career opportunity in the current transition. It sits at the intersection of legal and regulatory knowledge, domain expertise in the sector where AI is being deployed, and conceptual understanding of AI systems. None of those three components requires deep technical expertise. All three require genuine knowledge and the ability to apply it to specific situations with practical judgment.

The professionals most naturally positioned for this role are those with legal, compliance, or policy backgrounds in sectors where AI is being deployed consequentially, combined with the AI fluency to understand what the systems they are governing actually do. The combination is scarce. The demand is large. The premium reflects both.

What it pays: median premium of 41%, the highest of any role in this analysis, with significant variation by sector. Healthcare and financial services premiums are highest, reflecting the regulatory intensity of AI deployment in those contexts.

What it requires from you now: engagement with the regulatory frameworks governing AI in your sector, at a level of specificity that goes beyond awareness to genuine understanding of what the requirements mean for the specific AI deployments in your professional context.

Role 5: AI training data specialist

What it is: a professional responsible for creating, curating, and validating the data used to train and fine-tune AI models for specific professional applications.

The role exists because the quality of AI model outputs in professional contexts is directly dependent on the quality and appropriateness of the training data those models are built on, and the professionals with the domain knowledge to evaluate that appropriateness are not the same professionals who have the technical skills to manage training data pipelines. The AI training data specialist provides the domain knowledge component of a function that technical teams cannot provide themselves.

This role is more technically adjacent than the others in this list, and it may not be relevant for professionals in functions far removed from AI development. It is relevant for professionals in sectors where organisations are building or fine-tuning AI models for specific applications, including legal, medical, financial, and educational technology contexts, where the domain expertise required to evaluate training data quality is the scarce resource.

What it pays: median premium of 26%, with the premium concentrated in specialised sectors rather than distributed broadly.

What it requires from you now: engagement with how AI models in your domain are trained and what data they are trained on, at a level of specificity that allows you to evaluate whether the training data is appropriate for the professional context the model is being used in.

Role 6: AI-augmented domain expert

What it is: the role that is not a role, in the sense that it does not appear as a job title but describes the professional identity of the largest and most consequential group of professionals navigating the AI transition successfully.

The AI-augmented domain expert is a professional whose primary identity is their domain expertise, who has developed genuine AI fluency as an extension of that expertise rather than as a separate capability, and who uses AI tools to produce domain-expert level outputs faster, with better information, and with higher analytical rigour than was possible without them.

This is the profile described in Issue #5 as the most valuable position in the AI-augmented labour market. It is the profile whose compensation premium the Burning Glass data identifies as highest and most growing. It is the profile that every issue of this newsletter has been arguing is within reach of any professional willing to invest in the combination of deep domain development and genuine AI fluency.

It is also the profile that requires the most time to develop and the most deliberate investment to maintain. There is no shortcut to deep domain expertise. There is a significant accelerant to the AI fluency component of it, and that accelerant is the consistent, structured, application-grounded practice this newsletter has been describing since Issue #2.

What it pays: variable by domain and sector, but consistently at the upper end of the premium distribution in every sector where AI fluency enhances rather than replaces domain expertise. The upper end of that distribution is the 35 to 52% premium range described in Issue #33.

What it requires from you now: honest assessment of how deep your domain expertise currently is and what it would take to deepen it, combined with the AI fluency investment that makes that depth more productive. Both components require active investment. Neither compounds without it.

The action

Map each of the six roles above against your current professional profile. For each one, identify whether the role is accessible to you with current capabilities, accessible with a defined development investment, or not currently relevant to your professional context.

The roles that are accessible with a defined development investment are the ones worth building a specific plan around. Not all six. The one or two where the combination of your existing domain expertise and the AI fluency you are developing creates a genuine positioning advantage in a growing market.

Thursday we give you the prompt toolkit for building that specific plan, grounded in the role mapping you have done here and calibrated to the first ninety days of 2026 rather than to a generic annual development objective.

Ninety days is the right planning horizon for the AI transition. It is long enough to produce compound returns on a focused investment. It is short enough that the assumptions underlying the plan remain valid for its duration. Both of those properties matter in an environment changing at the rate this one is.

— The Artificial Idea team