T-C-L-D Audit: Spot AI's Erosion of Your Role

Hollowing Out: AI Erodes Roles Before Replacing Them

Knowledge jobs don't vanish overnight like in hype videos; they get hollowed out gradually. AI targets routine pieces—info gathering, writing, summarizing—leaving a shell that looks productive until economic shocks (recessions, reorgs) force cuts. Travel agents illustrate: Online booking commoditized routine reservations first, without immediate job losses. Downturns later exposed the change, shifting survivors to complex planning, emergencies, and human judgment.

Data backs this: OpenAI/UPenn estimate 80% of US workers have 10%+ tasks AI-affected; 20% see half impacted. Anthropic's index shows 49% of jobs with 25%+ tasks using LLMs. Microsoft Bing Copilot analysis of 200k sessions reveals top uses: writing, info provision—core to 'visible throughput' rewarded by old performance systems.

"AI doesn't have to replace your whole job to put you on thin ice. It only has to pick away at enough of the pieces inside the job that when the next shock comes, the rest of the story stops holding together."

Performance reviews lag because they measure output volume ('Did the deck get made?'), not necessity ('Did it need a human?'). This creates a 'dangerous window' where calendars fill with low-value work, masking erosion. Theater (performative rituals) collapses first since it was already low-attention; commodity follows as AI scales without human limits.

Run the T-C-L-D Audit on Your Work

This 30-60 minute exercise dissects your last 10 business days into four buckets, forcing honesty about value. Prerequisites: Access to calendar, sent emails, Slack/DMs, docs/tickets. Assumes knowledge worker role (emails/meetings heavy); do it manually first for calibration, then AI-assist.

Steps:

1. Open all sources side-by-side.
2. Tag each item (meeting, email, doc, message)—not projects/roles—with T, C, L, or D. Use first instinct; agonize = L.
3. Count totals by time (hours) or items for proportions.
4. AI acceleration (optional, via Claude/computer use): Chunk by tool (e.g., one agent per email/calendar). Provide clear definitions/prompts: "Tag as T if performative with no examined value." Expect iteration; full automation needs your judgment input.

Bucket Definitions & Tests:

• T (Theater): Organizational performance, not value. Disappears without admitting waste. Examples: Unblocking status meetings, unread decks for flipping, ritual check-ins/feedback post-decision, legacy reviews. Test: Main fallout is exposing fiction? >"Tagging T means admitting you spent professional time on something that did not need to happen."
• C (Commodity): Real value, but not you-specific. Examples: Summarizing known inputs, routing decisions, status reports anyone competent writes, first-draft docs in fixed formats. Test: Spec it out—could junior/vendor match output? Valuable but scarce no more; AI compresses throughput.
• L (On the Line): Gray zone, vulnerable soon. Examples: Structured pattern recognition, history-based relationships, repeatable synthesis, junior-doable + your 'judgment' (hard to articulate). Feels expert but commoditizing.
• D (Durable): You irreplaceably alter outcomes. Examples: Reading rooms to reframe problems, presence shifting decisions via taste/context/courage. Test: Output relies on indescribable judgment; you changed the question, not just answered it. Rare, power-law distributed (few high-impact hours define careers).

Common pitfalls: Undercount T (confuse 'expected' with 'valuable'); overclaim D (self-image vs. hours logged); ignore L's migration to C.

Redirect to Durable Work: Results, Pitfalls, and Six Moves

Expect: High T/C (invisible erosion), low D (under-allocated), L signaling shifts. Reveals mismatch: Identity clings to imagined uniqueness, but weeks prioritize defensible routines.

Core Principle: Question-Holding vs. Answering. Durable = holding ambiguity (diagnose real issues, evolve questions via context/judgment). Commodity/theater = answering knowns. AI excels at latter; humans at former. "Durable work is question-holding instead of question-answering."

Legibility Paradox: Visible busyness (T/C) props up reviews; durable often invisible (e.g., quiet reframing). Cutting T/C exposes you short-term but frees capacity.

Post-Audit Moves (Prioritize by Impact):

1. Stop defending T: Delegate/async/AI (e.g., bot summaries).
2. Automate C: Prompt LLMs for drafts/routings; spec for juniors.
3. Probe L: Articulate judgment— if specifiable, shift to C; else build toward D.
4. Amplify D: Propose projects centering it; track/quantify impact.
5. Update identity: Self-image as 'question-holder' before reorgs force it. Pour saved time into durable, not more C (trap: 2x productive at collapsing value).
6. Re-audit biweekly; share anonymized with peers for calibration.

"The first sign that your job is on thin ice is often a full calendar and no clue what's happening."

"Your week is not organized around durable work."

Practice: After tagging, journal one D item—why durable? Prototype AI for top C. Fits early/mid-career pivot in AI era; scales to teams (aggregate audits for reorg prep).

Key Takeaways

• Tag last 10 days' items as T/C/L/D to quantify vulnerability—aim <20% T, minimize C/L.
• Eliminate theater first: If no one examines output, AI it now.
• Test commodity: 'Could I spec this for anyone?' → Automate/offload.
• Seek durable: Did you reframe the question? Double down there.
• Avoid identity trap: Audit hours, not self-image; redirect saved time to D.
• Use AI for audit (chunked prompts) but supply your definitions.
• Re-run biweekly; downturns accelerate shifts—act pre-shock.
• Power-law careers: Few D moments define you; organize week around them.
• Question-holding wins: AI answers; you evolve problems.
• Leaders doubling C productivity lose—shift before systems update.