The AI-proof career audit: score your exposure, build your moat, reposition in 90 days

June's payroll report added 57,000 jobs and the market shrugged. Underneath that number, entry-level knowledge work — admin, content, support, junior analysis — keeps getting absorbed by agents that cost less than a coffee per workday. You've read those stories. The useful response isn't anxiety. It's an audit.

This is the audit. It takes one weekend, produces a number, and converts that number into a 90-day plan. No hand-waving about "learning AI," no lists of "safe careers." Your job title isn't exposed or safe — your task mix is, and you can change your task mix without changing employers.

Step 1: Inventory what you actually do

Open your calendar and sent mail from the last two weeks. Write down every distinct task you performed — not your job description, what you actually did. Aim for 15 to 25 items. "Meetings" is not a task; "negotiated scope with a client," "gave project status to leadership," and "sat in a meeting I could have skipped" are three different tasks with wildly different futures. Next to each, estimate the percentage of your working time it consumes. Force the percentages to total 100. This is the uncomfortable part: most people discover that 30 to 40 percent of their week is stuff they'd forgotten they do.

Step 2: Score each task on three axes

Score every task from 0 to 2 on each of these three questions:

• Routine. Does the task have roughly the same shape every time? (0 = every instance is novel, 1 = a familiar pattern with real variation, 2 = same template, different inputs)
• Digital. Do the inputs and outputs live entirely on a screen — text, data, images, code? (0 = mostly physical or in-person, 1 = mixed, 2 = fully digital)
• Checkable. Could someone verify the output is good in a fraction of the time it takes to produce it? (0 = quality is a judgment call that takes expertise, 1 = partially checkable, 2 = quickly verifiable against clear criteria)

Add them up. A task scoring 5–6 is red: routine, digital, checkable work is exactly what current agents do well, because it can be delegated and verified cheaply. 3–4 is amber: partially exposed, likely to become "human reviews the agent's draft." 0–2 is green: for now, this work needs you specifically.

Then compute one number: multiply each task's score by its time percentage, sum, and divide by 6. That's your exposure index, from 0 to 100. Below 30, your week is mostly moat. Above 55, a meaningful majority of your paid hours are automatable with today's tools — not someday, today.

A worked example

A marketing coordinator inventories her fortnight: weekly status decks (12% of her time, scores 6), first-draft campaign copy (18%, scores 5), performance reporting (10%, scores 6), scheduling and inbox wrangling (15%, scores 5), stakeholder alignment calls (20%, scores 2), vendor and event management (15%, scores 2), and campaign strategy sessions (10%, scores 1). Her exposure index: about 58. More than half her paid time is red-zone. But look closer — the audit just handed her the plan. Her green tasks (alignment, vendors, strategy) are the ones her manager actually values at review time. The move isn't defending the decks. It's handing them over first.

Step 3: Know which moat you're building

Green-zone work clusters into four durable moats. Pick the one or two that fit how you're already wired, and steer your task mix toward them:

• Accountability. Someone must own the decision and absorb the blame — budget authority, sign-offs, licensed judgment. AI can draft the recommendation; it can't be fired for it.
• Presence. Work where being physically in the room is the product: site visits, hands-on operations, high-stakes in-person persuasion.
• Relationships. Trust networks, institutional memory, the ability to get a yes in one phone call that would take a stranger six meetings.
• Taste. Deciding what "good" looks like under ambiguity — the editorial call, the product call, the hire. This is the moat that pairs best with agents, because taste scales when production is cheap.

And there's a meta-moat that multiplies all four: being the person who operates the machines. Every red task you automate yourself becomes evidence of a skill, not evidence of your replaceability.

Step 4: The 90-day repositioning

Days 1–30: automate your own red zone. Take your three highest-scoring tasks and build an agent workflow for each — the $2 Test system is the full manual. Keep quality notes: what the agent nails, what it fumbles, how long review takes. You're building two things at once: freed hours and a documented capability.

Days 31–60: shift the mix. Spend the freed hours visibly in the green zone. Volunteer for the client-facing piece, the vendor negotiation, the ambiguous project nobody has scoped. Decline or delegate one red-zone recurring commitment. Your calendar two months from now should look measurably different from the audit — that's the metric.

Days 61–90: make it legible. Quiet competence gets automated; visible judgment gets promoted. Book the conversation with your manager and run this script: "Over the last quarter I've automated X and Y — here's the quality record. That freed about N hours a month, which I've been putting into A and B. I'd like to make that official." You've just repositioned from a person who does red-zone tasks to a person who converts red-zone tasks into capacity — which is a role no org has enough of. If the conversation goes well, the follow-up is a comp conversation, and now you have the evidence file for it.

Failure modes

• Auditing your title instead of your tasks. "Marketing manager" tells you nothing. The 22 things you did last fortnight tell you everything.
• Hiding the automation. Using agents secretly feels safe and reads as dispensable when discovered. The leverage is in being loudly good at it.
• Moat-shopping into work you hate. If you're miserable in sales conversations, "relationships" is not your moat. Build from strengths; the audit shows where they already earn green scores.
• Treating green as permanent. Scores drift as capabilities improve. Re-run the audit quarterly — it takes an hour after the first pass.
• Waiting for the org to decide for you. By the time your company runs this analysis on your role, you want to be the person who did it first — on yourself, on your own terms.