Judgment Gets a $100 Million Bonus Pool at EY

On August 31, EY US announced a $100 million reward pool for work that combines technology with business acumen, judgment, adaptability, and other human skills. The firm did not present those abilities as consolation prizes for people who cannot use AI. It put them beside technology adoption and measurable business results.

The money changes the conversation. Employers have spent years saying that judgment, communication, collaboration, and curiosity matter. Employees have heard the same abilities described as culture, leadership potential, or the extra work expected at the next level. Once cash is attached, the questions get sharper. Which action counts? Who saw it? How is a shared result divided? Does an award supplement base pay and promotion, or replace them?

EY’s announcement names three categories: everyday leadership, transformation, and enterprise impact. It says colleagues at all ranks can recognize one another. A WorldatWork summary adds two useful limits: spot recognition can reach $500, while awards for individual or team work that materially affects the business can reach $25,000.

Neither source supplies the denominator. They do not say how many people are eligible, how many awards EY expects to make, how the $100 million will be divided, how nominations will be calibrated, or how outcomes will be reported across roles and ranks. The announcement is evidence of a program and a budget. It is not yet evidence of how the rewards will land.

The missing denominator is the management problem. Human skills carry more weight when AI handles more routine production, but they also become easier to praise vaguely. A fair program has to connect an observable decision or behavior to an outcome, preserve credit for collaborators, distinguish development from compensation, and leave room for a person to challenge the account.

EY has put a price on the question. Managers and employees now need a work record that can answer it.

August 31 puts $100 million behind judgment

EY describes the program as part of a broader effort to reward people who lead the firm into its next stage. The wording matters. The pool recognizes technology adoption, but adoption is not the only object. Business acumen, judgment, adaptability, innovation, skills, behaviors, and outcomes all appear in the same short announcement.

Those terms do different jobs.

Technology adoption is an activity. A person can use a tool, introduce it to a team, or redesign a process around it. Judgment is a decision under uncertainty. Adaptability describes a response when conditions change. Business acumen connects work to a customer, risk, cost, or revenue consequence. An outcome is the result that appears after those actions.

Combining the units can reward the full chain from tool to result. It can also hide weak attribution. A high volume of AI use does not prove good judgment. A strong business result may depend on market timing, a sales relationship, or work performed by people who never appear in the nomination. A difficult intervention may prevent a loss without producing a visible gain.

The three award categories help, but they do not settle the unit.

Everyday leadership appears designed for frequent recognition across colleagues. Transformation points to work that changes how a team or process operates. Enterprise impact implies a result with importance beyond one immediate group. Those are sensible distances from local behavior to firm-level outcome. They still need separate evidence standards. A colleague can directly observe patient coaching during a difficult handoff. The same colleague may not be able to verify the financial effect claimed for a global process change.

Peer recognition expands the field of view. People often see work that a direct manager misses. An analyst catches a false assumption. A specialist explains a model limitation to a client. A coordinator prevents two teams from duplicating a migration. A junior colleague documents an exception before it spreads. Letting every rank nominate can surface those contributions.

Peer nomination also creates an attention market. Work performed in large meetings attracts more witnesses than quiet review. Client-facing teams can narrate impact more easily than internal control functions. Senior employees have broader networks. A polished account of an ordinary contribution may travel farther than the technical objection that saved an engagement.

Money raises the cost of that visibility gap. A thank-you note can be imperfect and still useful. A $25,000 award becomes part of a person’s compensation history, even if the company treats it as discretionary. Repeated access to the award can affect retention, confidence, and perceived career opportunity.

EY made a larger compensation and technology announcement in June 2024. The firm said it would invest $1 billion over three years in early-career compensation, AI-enabled platforms, its 360 Careers experience, education, and wellbeing. That figure should remain separate from the new $100 million pool. The older commitment covered several programs and a longer period. It does show that the latest rewards sit inside an existing effort to connect pay, technology, and career development.

An employee still needs to know which lane applies. Base salary pays for the continuing scope of a job. A promotion recognizes sustained work at a higher level. A bonus rewards performance against a period or objective. A spot award recognizes a particular contribution. Learning funds help a person build capability. Mixing those lanes can make a generous pool less useful than it appears.

If sound judgment has become a recurring requirement of the role, a one-time award does not settle whether the salary band and level are right. If a person made one exceptional intervention, a permanent pay change may be the wrong instrument. The first decision is not who deserves praise. It is what kind of compensation or career decision the evidence supports.

The difference persists after payment day. Base salary becomes the starting point for later percentage increases and is paid for as long as the role continues. A one-time award does neither. An employee can reasonably celebrate $25,000 and still ask whether a repeated expansion of the job belongs in base pay. The award notice should answer that question directly instead of leaving the employee to infer that recognition closed the matter.

EY rewards the skills that make AI useful

“Human skills” sounds soft when nobody names the work. AI makes the missing unit easier to see.

A system can draft a client memo, classify a transaction, summarize a meeting, produce code, or suggest a forecast. Someone still has to decide whether the task was suitable, whether the inputs were permitted, whether the output fits the context, which exception matters, and when to stop using the system. The decisive contribution may occur before the prompt, during review, or after the output reaches another person.

Consider professional judgment in an audit or tax engagement. The useful action is not a favorable personality rating. It may be the decision to reject an apparently complete answer because one assumption conflicts with the underlying record. Evidence can include the flagged assumption, the alternative analysis, the reviewer who confirmed it, and the client or risk outcome. The same structure works outside professional services. A product manager may narrow an AI feature after a failed evaluation. A recruiter may decline an automated recommendation after finding that the source data does not support it. A support lead may recognize when a generated response will make an angry customer angrier.

Business acumen also needs a work object. A person might choose a lower-cost model because latency matters more than marginal benchmark quality. Another might redesign a workflow because review consumes the supposed time saving, or decline an automation because its exceptions would move work to a more expensive team. The evidence is a tradeoff and its result, not a manager’s impression that someone thinks commercially.

External labor data points in the same direction, with firm limits. PwC’s 2026 Global AI Jobs Barometer analyzes more than one billion job advertisements across 27 countries. PwC reports an average 62% wage premium in jobs requiring AI skills. It also finds that entry-level roles exposed to AI are seven times more likely than other entry roles to request senior human skills. Job advertisements show employer demand and advertised requirements. They do not prove that one skill caused an individual’s pay, that every posting became a hire, or that a worker received the advertised wage.

Technical and human ability are not opposing teams. A person who understands a model but cannot explain a risk to a client leaves value stranded. A skilled communicator who cannot inspect the work may confidently transmit an error. The business value often sits in the handoff between technical possibility and a decision.

Microsoft’s 2026 Work Trend Index exposes a gap in that bridge. Its research combines product telemetry with a survey of about 20,000 AI users in ten countries. Only 13% of respondents said they were rewarded for reinvention even when the immediate result fell short. In Microsoft’s model, organizational factors accounted for 67% of the variation associated with self-reported AI impact, compared with 32% for individual factors. Those are associations in a vendor study, not proof that a particular reward raises productivity. Asking people to experiment without changing recognition, goals, or leadership support remains an organizational choice.

A September 1 BambooHR survey offers another view. The sample included 1,608 US salaried desk workers and 520 HR managers and senior HR leaders. Workers reported using AI for 87 minutes a day. HR respondents said 43% of job descriptions had been updated and 77% planned updates. Self-reported minutes do not measure value, and a revised description does not prove that work changed. AI activity can spread faster than job architecture, leaving managers to reward new work through old categories.

A human-skills pool can make review, exception handling, explanation, coordination, and responsible adaptation visible before an annual role-architecture project catches up. The pool cannot replace that project. When the same contribution repeats, the organization has learned something about the job.

A useful nomination contains a before-and-after account. What decision was at stake? What did the person observe? What action did they take? Who else contributed? What changed for the customer, team, control, or cost? Which part remains uncertain? That record asks more than “showed great judgment,” and gives the employee something real to own or correct.

A $500 thank-you and a $25,000 award solve different problems

The two reported award limits imply two different operating systems.

A spot award of up to $500 can recognize a specific behavior soon after it happens. Speed is part of its value. A colleague documents the contribution while the evidence is fresh, the recipient learns what the organization noticed, and a manager can reinforce useful conduct without waiting for an annual cycle.

A reward of up to $25,000 requires more than a larger version of the same note. At that size, verification, attribution, calibration, and compensation context matter. The nomination may involve a team result, a material client outcome, or a change adopted across the enterprise. People outside the immediate team may need to test the claim.

Treating the tiers identically creates two opposite errors. A committee that demands a business case for every $200 recognition payment will make everyday appreciation slow and scarce. A program that sends a $25,000 claim through the same lightweight approval as a thank-you will turn narrative confidence into money.

The dividing line should be the consequence of the award, not a judgment about the employee’s prestige. A practical design can use three lanes:

Reward lane	Evidence standard	Review	Compensation boundary
Immediate recognition	One observable action, named witness, short description of who benefited	Manager or delegated local approver; quick duplicate check	One-time recognition; no promise of level or salary change
Material contribution	Work record, collaborators, verified outcome or avoided risk, role context	Cross-functional verifier and calibration group	One-time award; review whether recurring scope belongs in base pay
Sustained higher scope	Repeated evidence across a meaningful period, decision rights, complexity, peer comparison	Normal promotion and pay process	Level, base salary, and role design; not replaced by a spot award

The third lane does not need to come from EY’s program to be essential to its fairness. A bonus committee should be able to redirect evidence into the normal promotion or salary process when the nomination describes continuing work at a higher level. Otherwise, the organization can keep paying occasional awards while avoiding the recurring cost of recognizing the job that already exists.

The reverse boundary matters too. One dramatic result does not automatically prove sustained higher scope. A team may land an unusual client, recover from an incident, or deliver a difficult launch through effort that should be rewarded but is unlikely to recur. A meaningful award can recognize that work without inflating a title.

Individual and team rewards need different arithmetic. When a group changes a process, equal division may ignore different contributions. Managerial discretion may favor visible leaders. Naming only the person who presented the result can erase the people who found the problem, built the fix, tested it, trained users, and handled exceptions.

The nomination should list contributors before it proposes a split. Each person can be attached to a work object: problem definition, customer context, model evaluation, implementation, control design, adoption, or outcome verification. The group can then decide whether the award recognizes a shared result, several distinct contributions, or one exceptional intervention inside a team effort.

Managers need a specific attribution rule. A leader should not receive the same credit as the team merely because the work happened in that reporting line. Managerial contribution can be real: securing budget, removing a constraint, protecting review time, or changing a decision after challenge. Naming that contribution prevents both automatic credit for hierarchy and automatic exclusion of useful management work.

No formula will eliminate judgment from a judgment award. The aim is to make the judgment inspectable. A calibration group can compare evidence, ask whether similar work was treated similarly, and record why an amount changed. It should not pretend that $12,500 represents twice as much adaptability as $6,250.

Frequency also changes meaning. Five small awards across a year may show a pattern of useful contribution. They may also reflect a team with an active nominator. A single large award may recognize exceptional impact or access to a high-profile project. Managers should use award history as a prompt to inspect work and opportunity, not as a points table for promotion.

Keeping the distinction protects employees on both sides. A person with several awards should not have to argue that praise counts as pay progression. A person with no award should not be treated as a low performer when the program may have offered fewer visible nomination moments in their role.

Shared work makes human contribution harder to price

AI-enabled work with real consequences usually has more than one author.

A consultant may frame the problem while a model drafts an analysis. A specialist finds a regulatory exception. An engineer repairs the data path, a manager chooses the client recommendation, and an operations team absorbs the change. The result belongs to a chain. A reward program fails if it gives all credit to the person nearest the final slide.

AI introduces a second attribution problem. Organizations can observe tool usage more easily than judgment. Login counts, prompt volume, generated tokens, accepted suggestions, or time inside an assistant can be collected automatically. That makes them tempting inputs for a recognition program.

They are poor proxies for value. A high prompt count may reflect a difficult task, an inefficient workflow, repeated correction, or enthusiastic experimentation. A low count may belong to someone who used a system once, found a serious flaw, and stopped a risky deployment. An expert may solve the problem with a short exchange because the expert knows what to ask. A colleague may contribute the decisive context without touching the tool.

The metric soon changes behavior. If AI activity helps win an award, employees have a reason to generate visible activity. If successful outputs are rewarded while rejected outputs disappear, people have a reason to hide the review work that made the result safe. The adoption chart rises while the organization learns less about where the system fails.

Record the work event instead. Suppose a team uses AI to prepare a client proposal. The evidence could include the original objective, the approved data sources, the draft, a reviewer finding that changes the recommendation, the revised proposal, the colleagues who supplied subject expertise, and the eventual customer outcome. The AI contribution is part of the record. It is not the score.

Some of that evidence will be confidential. A client name, tax position, security flaw, personnel matter, or model output may not belong in a broad nomination system. The program needs a restricted evidence path that lets an authorized verifier confirm the claim while the general record keeps a short reason code. Otherwise, employees face a bad choice between exposing sensitive work and submitting an unverifiable story.

Some human contributions prevent outcomes rather than produce them. An employee may stop confidential data from entering the wrong system, challenge an implausible forecast, refuse to send an unsupported claim, or slow a launch until an accessibility problem is fixed. The immediate result may be delay, extra cost, or a smaller headline. A reward system focused only on positive financial outcomes can punish the judgment it claims to value.

Avoided loss requires care because it is easy to exaggerate. A nomination can describe the risk, the available evidence, the decision taken, and the person who independently confirmed the issue. It should not invent a dollar value for an event that never happened. Risk owners can verify whether the intervention was reasonable without claiming that one review saved the entire account.

Adaptability has a similar measurement problem. A person who accepts every new tool is not necessarily adaptable. Someone who changes course after evidence improves may be. The work record should show the changed condition, the previous plan, the new action, and the result. That is different from rating whether the employee seemed positive during disruption.

Collaboration becomes distorted when it is treated as likability. A useful collaboration record names a dependency that crossed a boundary. Did the person translate a customer requirement for an engineering team? Did they share evidence that changed another group’s decision? Did they transfer a method so colleagues could work without them? Did they credit the person who supplied the missing fact?

Visibility is uneven before a reviewer opens the first nomination. Control, security, legal, quality, support, data operations, and enablement teams often prevent problems or improve other people’s output. Junior employees may have fewer enterprise-facing projects. Remote colleagues and people in smaller offices may have fewer informal witnesses. Part-time workers have fewer total hours in which a nomination event can occur. Unequal opportunity can enter the program without one reviewer making an overtly biased decision.

Calibration should therefore compare opportunity as well as awards. Which groups received work that qualified for the largest category? Who was asked to present it? Which nominations arrived through peer recognition and which through leaders? Who appeared as a contributor but never as the named recipient? Those questions do not prove discrimination. They reveal where the program needs a closer look.

The toughest cases will remain disputed. A transparent record gives the employee and reviewer something specific to dispute: an omitted collaborator, an unsupported outcome, an incorrect role expectation, or an inconsistent amount. “Leadership did not see enough judgment” offers no such path.

Career Residency moves the test earlier

EY is also moving its human-skills thesis into the entry path.

On August 17, the firm announced Career Residency, an eight- to twelve-month paid learning experience positioned against the familiar eight-week internship. The curriculum names critical thinking, professional judgment, curiosity, AI and technology fluency, collaboration, and communication. Participants would use a Skills Arcade and receive a personalized Skills Card. Applications are due to open in fall 2026, with the first program scheduled for January 2028.

The dates impose a clear evidence boundary. Career Residency has an announced design, not an operating record. No cohort has completed it. There are no published conversion, performance, retention, or participant-experience outcomes. The program can still show how EY expects people to build the abilities it plans to reward.

Longer paid work can expose judgment in context. A short simulation can test whether a candidate recognizes one planted error. Months of supervised work can show whether the person asks for evidence, learns from correction, explains uncertainty, transfers a method, and changes course when a client or system behaves differently. The relevant evidence is a sequence, not a personality snapshot.

Entry-level work is changing at the same time. Routine drafting, research, formatting, and analysis have often served as both production and training. When AI performs more of the first pass, a junior employee can lose the repetitions through which context and judgment develop. Digidai examined that risk in Junior Roles Lost the Work That Taught Judgment. A longer residency can help only if residents perform real work, receive review, see exceptions, and understand why a senior colleague accepts or rejects an output.

The Skills Card could become useful evidence or an attractive label. The difference lies in what it contains. A record of completed modules says the person encountered material. A work-based record can name the task, the decision, the feedback, the revision, and the capability demonstrated. Neither should become a universal score that follows the participant without context.

Current European Commission AI literacy guidance offers a useful contrast. It explains that measures should reflect people’s knowledge, experience, the context in which AI is used, and the risks involved. It does not require a particular certificate or individual knowledge score. The guidance concerns AI literacy obligations, not US compensation or EY’s program. Its context-based approach is still a better model than declaring someone “AI ready” from one generic test.

A residency also makes opportunity design visible. Who can afford to join an eight- to twelve-month path? What is the pay? Where must participants live? Which benefits and employment protections apply? What happens to someone who completes the work but does not receive a full-time offer? EY’s announcement calls it paid and says participants may be eligible for an elevated analyst role. It does not publish those operating terms yet.

If the program supplies evidence for future rewards or promotions, employees need to know how it compares with other entry routes. An existing employee who learned the same skills on a project should not require the branded residency to receive credit. A resident should not be treated as fully proven because a Skills Card is easy to read. Comparable work needs comparable review.

Here the $100 million pool meets the career ladder. An early-career employee demonstrates judgment in bounded tasks. A more experienced employee applies it across uncertain cases. A manager creates conditions in which others can exercise and challenge judgment. An enterprise leader changes incentives and accepts evidence that invalidates a favored plan. The skill name may stay the same while the scope, consequence, and decision rights change.

The reward should follow that change in scope. Otherwise, “human skills” becomes a single badge attached to several different jobs.

Build a bonus rubric without scoring personality

A workable rubric begins with evidence fields, not adjectives.

Nobody needs a form for every decent decision. Small recognition can stay light. Material awards need enough structure that another reviewer can understand what happened, compare it with similar work, and correct a missing attribution.

Field	Question to answer	Evidence that fits	Weak substitute to avoid
Work event	What decision, exception, handoff, or change occurred?	Dated project record, client issue, review note, incident, process change	”Always shows leadership”
Human contribution	What did the person notice, decide, explain, or change?	Specific intervention linked to the event	Personality label or popularity
AI contribution	Where did a model or tool assist, and where did it fail or stop?	Approved workflow, output, evaluation, correction	Login, prompt, or token count as value
Business consequence	What changed for a customer, cost, risk, quality, speed, or colleague?	Verified result, bounded avoided risk, adoption record	Unsupported savings estimate
Shared credit	Who supplied context, built, reviewed, trained, or operated the change?	Named collaborators and work objects	Credit only for presenter or manager
Role context	Was the action expected, exceptional, or recurring at this level?	Job scope, goal, decision right, comparable work	Title used as proof of impact
Verification	Who can confirm the event and consequence?	Independent owner, customer evidence, control record	Nominator repeats the claim
Award decision	Which tier and amount fit, and why?	Written comparison with calibrated cases	Unexplained manager discretion
Career boundary	Does the evidence imply development, recognition, base pay review, or promotion?	Routed follow-up with owner and date	Bonus used to close the issue
Challenge	How can the employee correct the record?	Feedback channel, correction log, named reviewer	Secret score with no response path

Three examples show how the same table can handle different skills.

For judgment, an analyst finds that an AI-generated summary omits a condition that reverses the recommendation. The record includes the source, the omitted condition, the corrected recommendation, the reviewing specialist, and the outcome. The contribution is the decision to investigate and stop the wrong recommendation, not a general claim that the analyst is thoughtful.

For adaptability, a project lead learns that an automated workflow creates more review work than it removes. The lead changes the process, reduces the scope of automation, and reallocates reviewers. The record includes the original assumption, observed exception rate, revised design, and later operating result. Adaptability means changing a plan in response to evidence, not displaying enthusiasm for every rollout.

For collaboration, an operations employee translates recurring support failures into a test set that engineering can use. Product defects fall after the team adopts it. The record identifies the support cases, translation work, engineering collaborators, test set, and measured change. Collaboration is a dependency crossed and made reusable, not an assessment of warmth.

The approval path can remain proportional:

1. A nominator records the event, contribution, collaborators, and claimed consequence.
2. The recipient sees the nomination and can add missing contributors or correct the account.
3. A person who owns the affected work verifies the event and outcome.
4. A calibration group compares material awards with cases from other teams and roles.
5. Compensation or talent owners route recurring higher-scope evidence into salary, level, or job-design review.
6. The program reports distribution and challenge data without exposing confidential client or employee information.

Letting the recipient review the record before a large award may feel unusual. It solves several problems. The person can name collaborators, correct an inflated result, disclose that the action was part of ordinary assigned work, or explain why public recognition would create a client or privacy concern. An award should not require accepting a false heroic story.

The review step needs protected response options. Correcting a senior nominator or declining public recognition should not reduce the award. A confidential route matters when the dispute concerns credit, a manager’s account, or sensitive client work. The record can note that a correction occurred without broadcasting its contents.

Calibration needs a stable comparison set. Reviewers can compare the consequence, reach, evidence quality, role expectations, and contribution type. They should not force unlike work into a single score. A security intervention and a sales-process redesign may both matter without sharing a common dollar value.

Reviewer time belongs in the program budget. Large awards require people who can inspect technical work, client context, control evidence, and compensation history. If that work is added to already full calendars, decisions will drift toward the easiest stories to understand. Sampling small awards and giving material awards a staffed review path is cheaper than pretending every nomination receives equal scrutiny.

The program also needs a negative rule: AI adoption alone is not award evidence. Adoption can be part of the chain when it changes useful work. This avoids paying people merely for using the approved product and helps colleagues who improve the system through refusal, constraint, or careful review.

Existing compensation records provide another check. Digidai’s analysis of AI research equity grants showed how a large variable component can sit beside a much narrower salary band. The instruments differ, but the lesson carries over. A headline amount does not tell an employee what is recurring, liquid, likely, or available to comparable colleagues. Every award notice should state the amount, timing, conditions, and whether it changes any other compensation or career decision.

A rubric will not remove discretion. It can stop discretion from hiding inside flattering language.

Award day still owes employees a denominator

Winner stories can wait. The first useful program report is a denominator.

Employees need to know how the announced pool became decisions. That does not require publishing individual compensation. EY or any employer using a similar program can report aggregate counts and rates:

• eligible employees, nominations, unique nominees, recipients, and total awards;
• amount allocated, median and range by award category, and the unallocated balance;
• individual versus team awards, including how team awards were divided;
• nomination and receipt rates by rank, function, office or work arrangement, and other legally appropriate groups;
• peer, manager, and leadership nomination sources;
• time from work event to nomination, decision, and payment;
• number of corrected nominations, added collaborators, challenges, and changed decisions;
• cases routed to development, base-pay review, job redesign, or promotion;
• repeated recipients and groups with eligible employees but no nominations.

The report should also show the eligible period and population for every rate. An office that joined halfway through the cycle should not be compared with one that had a full year. Contractors, residents, people on leave, recent hires, and employees who depart before payment need explicit inclusion rules. Silence in those fields makes an apparently precise award rate hard to interpret.

No one should optimize every rate until it is equal. Roles have different opportunities and business consequences. A tax specialist, security reviewer, recruiter, and account leader should not produce identical nomination patterns. Large differences are questions. Reporting allows the organization to ask whether the difference comes from work, access, evidence, program design, or reviewer practice.

The award pool itself needs a time period. A $100 million commitment across one cycle means something different from the same amount across several years. Employees need an eligibility date, decision cadence, payment timing, treatment of departures, and rule for unused funds. Public materials reviewed for this article do not provide those details.

Managers need a denominator too. A leader with ten reports who submits ten nominations is behaving differently from a leader with five hundred reports who submits ten. A team with no awards may have little qualifying work, poor evidence, an absent nominator, or a manager who saves recognition for the annual review. Counts without eligible populations and opportunity context reward the most active narrators.

Program outcomes should remain separate from company outcomes. Revenue, retention, client satisfaction, cycle time, quality, and risk may change after the rewards launch. The pool will be one of many influences. A company can track whether rewarded practices spread and whether recipients stay without claiming that the bonus caused every later result.

Employees can ask five immediate questions before the first cycle closes:

1. Which actions and outcomes qualify for each category?
2. Who verifies a material claim and adds shared credit?
3. How are award amounts calibrated across roles and ranks?
4. When does repeated evidence trigger a base-pay, role, or promotion review?
5. How can a person correct a nomination or challenge an omission?

Asking those questions makes discretionary recognition usable. A program can move quickly at the $500 level and still become more deliberate as money, consequence, and career effect rise.

EY has moved human skills out of a closing paragraph and into a funded decision. The move also exposes how little the phrase explains on its own. Judgment is not a trait that a manager senses. It is a decision made in context, with evidence and consequences. Adaptability is not compliance with change. It is a defensible change in response to new information. Collaboration is not likability. It is work that crosses a dependency and leaves other people able to act.

The $100 million pool can reward those contributions. It can also teach EY which jobs, levels, and development paths have changed as AI enters more work. That learning will depend on the record beneath each award and the denominator across all of them.

Until those details appear, the precise conclusion is limited. EY has committed real money and named a set of capabilities. It has not yet shown who receives the money, how judgments are calibrated, or whether recurring human skill changes base pay and career progression.

The budget turns the claim into a test. The distribution will supply the answer.