Website of Peter Westerman

There is a genuine question underneath the anxiety about AI and professional work: if AI does half my work, is there half as much of me? The answer the evidence suggests is: no — there is more of you, applied to a different part of the work. The part AI handles is the production work. The part that remains is the direction, evaluation, synthesis, and domain judgment that make the production work valuable. That remaining work is not the lesser half. It is the more valuable half, applied to more output than you could previously produce alone.

This framing is not reassurance. It is what the evidence shows in the case study documented in this series and in what practitioners in other knowledge work domains are reporting as AI tools become more capable and more integrated into professional workflows. The shift is real, the direction is consistent, and understanding it clearly is more useful than either dismissing it or catastrophizing it.

The Role Shift Across Three Domains

In software development, the shift is the clearest because vibe coding is a concrete methodology with documented outcomes. Before AI collaboration, the primary professional skill was code-writing: translating requirements into working implementations, debugging failures, managing the codebase. The time concentration was in production — the writing and debugging of code. After AI collaboration at the level described in this series, the time concentration is in direction: defining requirements precisely, evaluating architectural proposals, testing implementations against intended behavior, managing the quality and coherence of the codebase over time. The primary professional skill is judgment about what to build and whether what was built is right — not the mechanics of building it.

In marketing, the shift follows the same pattern. The old role concentrated time in producing: writing copy, creating content, building campaigns from briefs. The new role concentrates time in directing: defining campaign goals with precision, identifying the audience insights that make campaigns effective, evaluating which of ten AI-generated creative directions aligns with the brand and the audience, making the judgment calls that determine creative direction. The ability to test ten creative variations in the time it previously took to produce one means the bottleneck is no longer creative production — it is creative judgment. The professionals who have deep creative judgment built on domain expertise become more valuable relative to those whose primary value was in creative production.

In financial analysis, the shift parallels the others. The old role concentrated time in building: constructing models, compiling data, running scenarios, producing reports. The new role concentrates time in the analytical work that precedes and follows building: defining the analytical question with precision, specifying the model structure and key assumptions, evaluating model outputs against business logic and reasonableness, producing the synthesis narrative that makes the analysis actionable for decision-makers. The model-building mechanics that consumed significant time are compressed. The judgment about what question to ask, how to structure the model to answer it, and what the output reveals about the underlying business remains entirely human.

The New Scarce Skills

The skills that become scarce and valuable as AI absorbs production work are consistent across domains:

Precise problem definition — the ability to specify exactly what is needed with enough clarity that an AI can produce it, or that a human collaborator can act on it without repeated clarification. This requires deep domain knowledge, analytical clarity, and communication precision. It is harder than it sounds. The gap between “I know what I need” and “I can specify it clearly enough for an AI to produce it” is where most early AI collaboration failures occur.

Expert-level output evaluation — the ability to assess AI output not for whether it seems generally good but for whether it is correct, complete, and appropriate for purpose. This requires the same domain expertise that the production work previously required. A financial analyst who can verify that a model’s output is plausible given the underlying business dynamics needs deep financial judgment, whether they built the model or directed an AI to build it. The evaluation skill requires the production skill as a foundation.

Synthesis across AI-produced components — the ability to integrate multiple AI-produced outputs into a coherent whole that addresses the actual problem. AI produces components: the market analysis section, the competitive positioning section, the financial model. The human synthesizes them into a strategic recommendation that accounts for how the sections interact, what the combined picture implies for the decision at hand, and what the audience needs to understand to act on the recommendation. This synthesis is distinctly human work that current AI systems do less well than experienced domain experts.

Accountability and judgment — the professional who directed the AI, evaluated the outputs, and approved the final product owns the outcome. This accountability is the foundation of professional value in a world where production is increasingly automated. The judgment behind the approval — and the responsibility for that judgment — is not something AI systems can hold. It remains with the professional.

What Atrophies and What Must Be Maintained

Some production skills will exercise less frequently in an AI-augmented practice. Writing code by hand. Drafting documents from a blank page. Building financial models from scratch. These skills are exercised differently — in evaluation and correction rather than in original production. If the production practice is entirely abandoned, the evaluation skill that depends on it will degrade over time.

The mitigation: deliberately maintain production skills at a sufficient level to support evaluation. Not at the level required for production to be the primary activity, but at the level required to recognize when AI-produced work is wrong in ways that are not obvious from the output surface alone. Think of it as professional fitness: you do not need to write every line of code to remain capable of evaluating whether AI-generated code is architecturally sound. You need to write code often enough that the patterns remain fluent and the subtler forms of error remain recognizable.

What grows, actively and consistently, in an AI-augmented practice: direction skill, evaluation skill, synthesis capability, and the domain expertise that makes all three possible. These get more exercise per hour of professional work in an AI-augmented practice than in a traditional one. Each session that involves directing AI, evaluating its output, and iterating to a better result exercises these skills. Each good evaluation that catches a subtle problem and directs a specific correction deepens them. The compound effect of this exercise, over months and years of consistent AI collaboration, produces professionals whose judgment has been refined through many more evaluations than traditional production workflows would have provided.

The Career Development Reorientation

The professional development priorities that produce the most value are changing. Investing in production speed — getting faster at writing, modeling, drafting — is less valuable than it used to be. Investing in judgment depth — the domain expertise that makes evaluation and direction excellent — is more valuable than it has ever been, because judgment depth now multiplies AI production rather than substituting for it.

This reorientation is available to any professional who makes it deliberately. The first step is the audit described in the adoption roadmap post: identify which of your current activities are production work and which are judgment work. The second step is investing in the judgment work — developing deeper domain expertise in the areas where your judgment is most scarce and most valuable, because that is what amplifies AI production most effectively. The third step is building the context infrastructure that makes your domain expertise available to AI as a systematic resource rather than only through your direct involvement in each session.

The professionals who make this reorientation now, while it is still optional, will have built deep judgment-based capabilities by the time the production-work market has contracted substantially. The ones who defer the reorientation will be making it under pressure, from a weaker position, in competition with professionals who have had years to build the capability they are just beginning to develop.

The Professional Value Proposition in an AI-Augmented World

The traditional professional value proposition was built on two foundations: domain expertise and production capability. Expertise in the domain gave you the judgment to make good decisions. Production capability gave you the ability to implement those decisions. The combination — expert judgment applied to skilled production — was the value that commanded professional compensation.

AI fundamentally changes the second foundation. Production capability — the ability to implement decisions — is increasingly available from AI systems at a fraction of the cost of skilled human production. The organizations that used to pay professionals primarily for their production capability are discovering that AI can provide that capability at lower cost. The organizations that pay professionals primarily for their expert judgment are discovering that AI makes their expert professionals dramatically more productive.

The professional value proposition that is durable in an AI-augmented world: expert judgment applied to directing and evaluating AI production. Not just judgment. Not just AI. The combination of expert judgment with the production leverage that AI provides. This combination produces output that neither can achieve independently and at a cost structure that is compelling relative to either pure human teams or pure AI systems.

Building that combination deliberately — developing the judgment depth that makes AI direction excellent while building the AI direction skills that make judgment productive — is the professional investment that produces the most durable and compounding return in the current moment.

Practical Guidance for the Transition

The transition from traditional production-focused professional work to AI-augmented direction-focused professional work is not instantaneous. It requires a deliberate development path. The most practical path:

Start by systematically using AI for your lowest-stakes production tasks — the ones where you would be willing to review and correct AI output even if the first pass is mediocre. Build the evaluation skill on low-stakes material where a mediocre evaluation has minimal cost. As the evaluation skill develops, apply it to higher-stakes material. As the evaluation skill becomes fluent on higher-stakes material, reduce the time you spend on direct production in those areas and shift it to the direction and evaluation activities that AI cannot replace.

This gradual transition preserves the production skill through continued maintenance exercise while building the direction and evaluation skills that will be more valuable over time. It also provides the feedback of experience: as you observe AI producing outputs that your domain expertise allows you to evaluate accurately and improve efficiently, you are building evidence that the methodology works in your specific domain — evidence that informs how broadly and confidently you can apply it going forward.

How Recent AI Innovations Change This Picture

The picture of AI doing “half the work” in a knowledge worker’s day has shifted in the time since this post was written. The fraction that AI can reliably handle has expanded, and the nature of what remains for the human has become clearer.

Computer use represents a category expansion. In the methodology described here, AI contribution was primarily through text generation, code generation, and analysis. AI could write the report but could not navigate the systems to gather the data for it. Computer use changes that: Claude can now navigate web interfaces, fill forms, extract data from applications, and perform multi-step tasks in GUI environments. Work that previously required human hands on a keyboard — because it involved navigating software rather than producing text — is now within scope for AI execution.

Extended thinking changes the quality ceiling on analytical work. The “AI does the first draft, human refines” model works well for production tasks. For analytical tasks that require reasoning through complex, ambiguous situations, the first draft is often the hard part — not because writing is hard, but because figuring out what the analysis should say requires substantial deliberation. Extended thinking applies that deliberation before generating the analysis. The human’s role shifts from extensively revising shallow analysis to reviewing and validating deeper analysis. The work is still there; it is just easier work.

The 1-million-token context window changes what “doing half the work” means for professionals managing large information environments. A lawyer who previously could only ask Claude to review one document at a time can now load an entire case file. A financial analyst can load an entire reporting period’s data. The AI’s contribution expands from targeted, small-context tasks to large-context synthesis tasks that were previously out of scope. “Half the work” becomes a larger half as scope constraints relax.