How to Conduct a Human-AI Partnership Audit

Do you remember your first AI lightning moment? An AI lightning moment is a deeply personal moment of reckoning, a jarring inflection point that rattles your sense of self. When AI completes a task that is core to what you do and who you are, and does it faster, better, and at scale; when it penetrates your domain expertise and walks right into the inner sanctum of your professional identity, that’s a lightning moment. You’re witnessing a kind of cognitive compression you didn’t think was possible.

Perhaps you watched AI structure a legal brief, read a CT scan, write code, build a valuation model, translate requirements into a conceptual design, synthesize a large data set, detect hidden correlations in molecular patterns, or build a detailed project plan.

A lightning moment delivers the profound realization that what you have been doing is no longer scarce. And you know another encounter with these same forces of commoditization is on its way. AI will turn something else you do into intellectual manual labor. As a result, your emotional response may range from excitement to existential dread.

In the world of human labor, expertise is territory. Professional identity is spatial as much as technical. So when AI produces the output you produce, you may see it more as a trespass than a support.

The experience resembles what psychologist Pauline Boss called ambiguous loss: a loss that remains unclear and unresolved. In the workplace, the loss is usually not the job itself. It’s the certainty that the skill will continue to be a proxy for status, security, and a clear professional identity. No one sends a note when the task you spent years mastering becomes a prompt. You keep your title and credentials. You may even keep your job. But the basis of your confidence has been upended.

Regardless of your role, you must look into the foggy future and predict AI’s ongoing incursion into your professional life. Rather than indulge in a sudden burst of tool enthusiasm, how can you rebuild your personal competitive moat—and then do it again as the next wave arrives?

Step 1: Classify your work into own, augment, automate

Start by conducting a personal Human-AI Partnership Audit. The purpose is to define the current state of your partnership with AI and figure out where to go next. Classifying your responsibilities into the three buckets of own, augment, and automate is step one. It gives you a rubric to examine your current ways of working and the maturity of your AI integration.

We’ve helped a variety of professionals conduct current-state audits of this kind. An architect who completed her audit found a three-part own/augment/automate ratio of 80:20

. An HR recruiter landed on 40:30, and an accountant on 60:30. Ratios vary widely from job to job and person to person. But as AI models improve, ratios are shifting inexorably to the right. The question is how to forecast and prepare for the shift, and then the one after that.

The own/augment/automate frame may help you bucket your responsibilities now, but it doesn’t tell you which of your skills are truly durable. The challenge is finding more predictive power in the face of an opaque future.

Step 2: Apply the algorithms of computational and evaluative work

After classifying your responsibilities, identify what aspects of your job are irreducibly human. This means distinguishing between two kinds of work that have long traveled together inside human labor—computational work and evaluative work. Most jobs combine both, but the individual whose role is composed primarily of computational work is at greater risk of being marginalized, or perhaps even displaced, by AI.

Consider the basic definition of an algorithm: a well-defined sequence of steps. The computational work algorithm includes three sequential steps: (1) process information, (2) recognize patterns, and (3) generate outputs. While humans use this algorithm all the time, compared to AI systems we have limited capacity in processing speed, pattern recognition, synthesis, simulation, and prediction.

The evaluative work algorithm consists of three different steps: (1) assign value, (2) exercise judgment, and (3) bear responsibility. Not only do humans have an advantage over AI systems here—they have a monopoly.

First, can an AI system assign value? No. It can only do it artificially after we tell it what to value and how to weight those values. We clarify principles and purpose. We cast the vision. We set priorities and boundaries, and we establish criteria for choice. Moral and ethical reasoning, as well as goal and objective setting, are the province of human beings.

Second, can an AI system exercise judgment? Not ultimate judgment. It can only exercise artificial judgment based on criteria we give it. The human being must integrate values, context, and human relationships, evaluate options and tradeoffs, and finally commit to a course of action under conditions of uncertainty.

Third, can an AI system bear responsibility? Clearly not. A non-sentient entity has no ability to answer for results and bear responsibility for outcomes. The human carries the weight of consequence. It’s the human that must build credibility, trust, and safety. It’s the human that must live with both intended and unintended consequences.

While the machine can process, recognize, and generate, the human must value, judge, and own.

Applying the audit: a UX designer’s partnership audit

To identify what is irreducibly human about your role, isolate the tasks and workflows that fall under the evaluative work algorithm. This will help you see the core-vs-crust distinction in the ways you create value. The software engineer leans into deciding what should be built and why. The radiologist focuses on communicating risk and uncertainty. The lawyer emphasizes strategic trade-offs—when to settle versus litigate. The recruiter invests more time defining what “fit” actually means in their cultural and organizational context.

Consider one UX designer we worked with who designs onboarding workflows. Here is her completed audit, with responsibilities broken out for each step in both algorithms.

Computational work

• Process information: analyze workflow analytics, review session recordings, evaluate support tickets.
• Recognize patterns: identify abandon or skip steps, drop-off patterns, churn correlates.
• Generate outputs: produce annotated wireframes, generate screen variations, write user stories.

Evaluative work

• Assign value: define successful onboarding, set friction thresholds, set ethical limits for data collection.
• Exercise judgment: decide when to act on research, resolve tension between user and business needs, make changes based on trust erosion.
• Bear responsibility: own results when a workflow goes live, answer to stakeholders when retention falls, admit errors when a feature fails.

Before the audit, this designer was in full flight mode as she came face to face with the reality that many of her responsibilities were being augmented or fully automated by AI. After completing the audit, she gradually gained new hope and confidence because she could see opportunities to deepen her expertise under the evaluative algorithm. She thought she was retreating out of survival. Now she feels she is moving to higher cognitive ground where she can compound her contribution.

What this means for leaders

The audit is a personal exercise, but it has organizational implications. AI adoption is often treated as a tool rollout. Leaders should manage the mechanical rollout, but they also need to manage the identity transition AI creates. That requires four moves.

First, make it safe. Your team needs to feel comfortable calling out their AI lightning moments. The employee who says, “That used to be my job,” is not necessarily resisting the technology. They may be telling you where they attach identity to their work.

Second, separate tasks from value. Help teams identify which parts of their work are becoming reproducible by AI and which parts remain uniquely human. This prevents two errors: defending every old task as if it were sacred, or automating work without understanding the judgment embedded in it.

Third, redesign roles in public. Ambiguity breeds fear. Be explicit about what the organization intends to automate, what it intends to augment, and what it expects humans to own.

Fourth, reward accountable work. If performance systems continue to reward only speed, volume, and visible output, employees will chase whatever makes them look productive. As AI increases output, leaders need to reward the quality of problem framing, judgment, risk management, and accountable decision-making.

---

AI doesn’t ask permission before it commoditizes work. As it continues to devour computational work, ask yourself how you can leverage your evaluative work. The work that remains scarce is the work people must still be trusted to own: setting direction, making tradeoffs, protecting values, interpreting consequences, and standing behind the result.