AI Job Risk 2026: Michael Meeks’ Task-Exposure Index — 35% vs Reskill Now

This guide verifies Michael Meeks’ Task-Exposure Index against the live, complete option.

It focuses on when crossing the 35% threshold should trigger a reskilling decision.

How It Works

Michael Meeks' AI Task-Exposure Index works at the task level, not the job-title level, and that distinction is the whole mechanism. Instead of asking "will AI replace this occupation?", the index breaks your job into its individual tasks, scores each one for how exposed it is to current AI capabilities, and rolls those scores up into a single figure for your role. TaskExposed, which publishes a comparable task-level methodology, gives each task a capability-weighted exposure score on a 0–100 scale and then classifies it into one of three buckets at canonical thresholds. Your overall exposure is therefore an average of your actual work, weighted by how much time each task consumes — which is why two people with the same job title can land in very different places.

The scoring logic follows a defined sequence. First, build a task inventory: list what you actually do in a typical week. Second, source the taxonomy — O*NET task lists are the standard input, and BLS Time-Use data supplies the minutes-per-task weighting so that a task eating 20 hours a week counts far more than one eating 20 minutes. Third, apply capability-weighted scoring, where a task's exposure reflects how well today's AI systems can perform it, not how futuristic the task sounds. Fourth, aggregate bottom-up into your index value. The Iceberg Index at MIT models workers as agents with skills, tasks, and location across thousands of counties and more than 32,000 distinct skills, which means the same task can score differently depending on the skill mix and market around it.

Before you act on a score, verify the terms behind it — this is the same verify-before-you-commit discipline you'd apply to any major decision. A single number is only as good as the inputs, so check which taxonomy was used, whether time weights reflect your real week, and which capability set the scoring assumed.

Two subtleties matter when reading your result. Exposure is not replacement: a high score can mean augmentation, where AI amplifies your output, or substitution, where it displaces it. Research published in the tourism-management literature distinguishes amplified and rebalanced roles from substituted and divergent ones, and notes that the latter need significant upskilling. Second, exposure alone is not urgency. The Reskilling Urgency Index (RUI) study found that occupations with the highest reskilling pressure — including paralegals, purchasing agents, and computer systems engineers — combine high exposure with limited skill transferability and intermediate specialization. A high exposure score with highly transferable skills is a far softer signal than the same score without them.

Your verification checklist before committing to any reskilling decision on the strength of an index: confirm the task list matches your real work, confirm the time weights are current, confirm the capability assumptions are recent, and confirm the score is task-level rather than an occupation-level proxy. No specific trigger percentage appears in the published methodology sources cited here, so before treating any score as a reskilling trigger, check the live tool for its stated threshold — such as the canonical bucket cutoffs TaskExposed uses on its 0–100 scale — and only act once that threshold is verified at the source.

Key Factors to Consider

This section alone lists the top decision criteria and numbers. The first criterion to verify is task coverage. According to TaskExposed's methodology, the index is built bottom-up from tasks, so before committing to any readout, you must confirm the live list includes every duty you actually perform, not a shortened version pulled from a job title alone. If the option you are looking at cannot show its full task set, treat that as an incomplete total and do not compare it to another option until both show like-for-like coverage.

The second criterion is the scoring basis. TaskExposed states that each task is given a capability-weighted exposure score on a 0–100 scale before any classification takes place. Because of that weighting, you should only compare scores that were produced by the same capability-weighting approach. A simple count of tasks will not be equivalent to a capability-weighted total, so the number to confirm here is not a count, but whether the output you are viewing is explicitly labeled as capability-weighted on that 0–100 scale.

The third criterion is labor-market friction. As shown by the ScienceDirect study introducing the Reskilling Urgency Index, exposure alone does not fully capture reskilling pressure, since the index also accounts for skill non-transferability and job non-specialization across 879 O*NET-defined U.S. occupations. For that reason, when reviewing any option, you should check for a stated treatment of those friction factors in the live output, rather than assuming exposure is the only input being used.

As for the numbers that matter, the concrete items to verify are the 0–100 capability-weighted scale and the bottom-up aggregation itself. Both are stated by TaskExposed as the measurable basis, while the 879-occupation scope is stated by the ScienceDirect Reskilling Urgency Index as the scale of that broader framework. Any other numeric labels should be treated as items to confirm directly from the live, complete option, with no assumption that they are fixed across every view.

Common Mistakes

One pitfall is comparing unlike outputs. For example, someone might pull an occupation-level summary from one report and stack it directly against a task-level readout from Michael Meeks’ AI Task-Exposure Index, treating both totals as interchangeable. According to TaskExposed's methodology, the build is bottom-up by tasks, so the unit of measure has to match before any comparison. The check is not the name of the role, but the denominator: confirm exactly what was counted on each side before committing to any side-by-side judgment.

A second pitfall is cherry-picking a single task. Picture a role where one very visible task is shown in a preview, and that single task is used to judge the entire job. The preview may leave out the rest of the task list entirely. The Reskilling Urgency Index research published by ScienceDirect points to uneven pressure across occupations, which makes it clear that the mix of tasks matters. Pulling only one item breaks the rule of comparing like-for-like totals, because you are no longer comparing the same set of work.

To catch the first pitfall, do not accept a "job" total unless the live option states what it is built from. If one option says "occupation" and the other says "tasks", do not force them to equal each other. To catch the second, force a count check: get the complete task list for that exact role from the live option you intend to use, and only compare it to another option that shows the same task definitions and the same count. If either option only gives a summary with no way to see the full list, do not commit until you can verify the complete option.

Both mistakes share the same fix: verify the live, complete option before you commit, and refuse to compare totals or terms that do not match. Skipping either check is what turns a preview into a false comparison, and that is the exact kind of error this section alone covers with these concrete examples.

Insider Tactics

This section alone gives non-obvious strategies and timing tips. One non-obvious strategy is to trim down to only the tasks you personally perform, not every task attached to an occupation. Since TaskExposed states that tasks are aggregated bottom-up, a full occupation list can easily hide which of your own duties are the ones that would actually need to shift. By forcing yourself to work off of your real workload first, you create a fixed set to verify against any option before you commit to it.

With that trimmed set in hand, don't look for a role that is "less exposed" as a whole. Instead, look for transfer points between your kept tasks and tasks that appear in a nearby role. The goal is not to find a perfect match, but to see if any of your existing duties can connect to the live option you're checking. If none of those transfer points exist in the complete, live version of that option, then shrinking the scope of your search is a more honest move than expanding it.

For timing, treat every public snapshot as a hypothesis. As shown by the Reskilling Urgency Index described in ScienceDirect, reskilling urgency is shaped by more than exposure alone, so a readout can change in relevance depending on context. Because of that, the moment to act is not the moment you first see a concerning readout, but the moment you can verify the live, complete option for the exact trimmed tasks you just isolated.

The strict timing tip is this: do not move from "verify" to "commit" until you can confirm that path is still open and complete for those specific tasks at the time of action. If you cannot confirm it live, do not pull in more tasks from the full list to justify moving forward. Keep your check locked to the set you trimmed, and only expand that set after you have verified a new option for it on its own terms.

Comparison

To compare like-for-like, only two options need to be weighed for this decision: an occupation-level readout, or a task-level readout in the manner of Michael Meeks’ AI Task-Exposure Index. According to TaskExposed, the task-level approach aggregates from the bottom up, giving each task a capability-weighted exposure score from 0–100 and placing tasks into three buckets at canonical thresholds. The occupation-level option cannot show those splits by design. For that reason, the task-level option is the winner for a commit decision, but only if the live, complete task inventory for that exact role is verifiably present. If that full inventory cannot be confirmed, the task-level option cannot be fairly compared at all.

That split matters when looking at urgency. The ScienceDirect study introducing the Reskilling Urgency Index (RUI) found that occupations such as computer systems engineers and architects, administrative services managers, purchasing agents, web developers, financial managers, paralegals, telemarketers, and clerical or customer service workers face the highest reskilling urgency under its combined measure. However, an RUI total and a task-level exposure total are not the same thing. The check here is strict: do not stack the RUI ranking against an occupation-level exposure figure. Only stack them against a task-level exposure figure that was built from the same task set, or the comparison is invalid.

On the action side, Rework states that a reskilling vs. rehiring cost comparison is the most direct budget justification metric for employers choosing between the two in 2026. No universal dollar figure is provided by that source for this section, so the comparison must be done by scope, not by guessing an amount. If rehiring is the alternative being weighed, the task-level option must be measured against rehiring for the exact tasks in question, not for the entire occupation. The occupation-level option cannot satisfy that scope check in most split cases.

Side by side, the occupation-level option can only win when there is no verifiable task split and the full scope is uniform. The task-level option wins whenever there is any split and its totals and terms can be verified against a complete list. The rule for this comparison is simple: verify the live, complete option before committing, and only pick a winner by comparing like-for-like totals and terms.

What to do next

Frequently Asked Questions

At what level does Michael Meeks’ index evaluate AI exposure?

It works at the task level, not the job-title level.

How does TaskExposed score an individual task?

It gives each task a capability-weighted exposure score on a 0–100 scale.

How are tasks combined and categorized in the methodology?

Tasks are aggregated bottom-up and classified into one of three buckets at canonical thresholds.

How many distinct skills does the IcebergIndex cover?

The IcebergIndex covers more than 32,000 distinct skills.

What attributes does IcebergIndex assign to each worker agent?

Each worker is modeled as an agent with skills, tasks, and location.

What decision is tied to crossing the index threshold?

Crossing the threshold should trigger a reskilling decision.

Quick answers

Also worth reading: SB 293 Audit Ratio: Legal Trigger, Not Forensic Finding: SB 293 Audit Ratio: Legal · Hiring bias audits: Threshold in 12 days vs retrain in 9 weeks: Hiring bias audits: Threshold in · The Psychological Cost of Being Lied to at Work A Neural Analysis of Trust and Decision-Making: Psychological Cost of Being Lied