Deconstructing Work and Tasks

Beyond the Job Title

Simply giving employees AI access rarely changes business outcomes. Meaningful gains require us to stop looking at 'Job Titles' and start looking at the discrete tasks that make up a workday. Jobs aren't automated; individual tasks are.

The Three Dimensions of Work

To unbundle a role effectively, we evaluate every task across three dimensions. First, cognitive complexity: does it require deep context or lateral thinking? Second, the cost of error: what happens if there is a hallucination? Finally, information structure: are the rules clear and verifiable, or is it based on human sentiment?

The Three Allocation Buckets

Every task in your organization belongs in one of three buckets. AI-alone tasks are low-risk and highly structured, like categorizing tickets. Human-plus-AI tasks are collaborative; the AI drafts, but a human validates. Human-alone tasks are protected, reserved for high-stakes ethics, empathy, and relationship building.

The Allocation Decision Tree

We use a decision tree to classify tasks consistently. We start by asking if the cost of an undetected error is unacceptable. If it is, we check if a human can reliably inspect the output. We then look for tasks requiring physical presence or live trust. Finally, we determine if the output can be automatically verified.

Defining the Review Gate

For any collaborative task, you must define where the human sits. Human-in-the-loop means the AI cannot act until a person explicitly approves the draft. Human-on-the-loop uses exception-based monitoring, where the AI executes standard work but flags anomalies for human triage.

Recap: Deconstruction & Buckets

Let's pause to consolidate. High-performing AI adoption doesn't replace roles; it reorganizes task allocations. Remember, these allocations are dynamic—as models improve, tasks may graduate from collaborative to autonomous.

Scenario: The Procurement Specialist

Consider a Senior Procurement Specialist. Instead of asking if AI can 'do procurement,' we look at tasks like invoice normalization. Or high-stakes rate negotiations. Click each task to see how we apply our framework to transform this role. Invoicing is low-risk and structured. We allocate this to AI-alone, using an automated parser to flag exceptions. Rate negotiation requires live trust and subtext. This is strictly Human-alone, freeing up time for the specialist to lead these strategic discussions. Drafting RFP summaries is moderate risk. We use Human-plus-AI, where the LLM extracts specs and the lead reviews qualitative responses.

Apply the Tree

It's your turn. A company needs to draft a PR response for a high-stakes product recall. Run this task through the decision tree. How would you allocate it, and what review gate would you use? Type your reasoning below.

Your 4-Step Implementation Path

To audit a department, start by inventorying atomic tasks as specific verbs and objects. Next, run them through the decision tree with the practitioners in the room. Establish clear verification checkpoints for collaborative work. Finally, document the matrix and re-evaluate every six months as models evolve.

Common Pitfalls to Avoid

Beware of common failure modes. Don't evaluate job titles—it's too vague. Avoid the 'Blanket Ban' trap; even high-stakes tasks benefit from AI synthesis. And never assume Human-plus-AI is 'zero-effort' automation. Reviewing takes time, and your workflows must budget for it.