How AI Is Rewriting Workplace Culture: A Leadership Operating Guide for 2026

Culture changes when the default behavior changes

A company does not acquire an AI culture when it buys licenses, publishes a policy, or adds a chatbot to the intranet. Culture changes when ordinary choices change: where an employee looks for an answer, what evidence a manager expects, how a team prepares for a meeting, who can challenge a recommendation, and whether people are rewarded for improving a process rather than merely keeping it moving.

That distinction matters because the same model can create radically different workplaces. In one company, AI becomes a private shortcut used by a confident minority. In another, it becomes a governed shared capability that reduces search, translates specialist knowledge, and helps people test ideas. In a third, it becomes an opaque monitoring layer that makes employees defensive. The technology does not select the culture. Product design, incentives, access, leadership behavior, and decision rights do.

The most important cultural effects also arrive before an org chart changes. A junior analyst can produce a credible first draft sooner. A field supervisor can question a plan with evidence that was previously buried in three systems. A manager can review more work, but may understand less of how it was produced. These small changes alter status, learning, trust, and control long before a formal restructuring appears.

Bizz AI development services focus on that operating context as well as the model. A useful workplace system needs trusted knowledge, workflow integration, explicit authority, feedback, and a user experience that helps people understand when to rely on AI and when to slow down.

• Culture is repeated behavior under real incentives, not a launch message.
• Private productivity and shared organizational capability are different outcomes.
• Access, workflow design, and management behavior shape adoption more than slogans.
• AI can distribute expertise or concentrate control depending on how it is implemented.
• The first signs of change appear in meetings, decisions, learning, and handoffs.

The first divide is not believers versus skeptics; it is supported versus unsupported use

Leaders often describe adoption as an attitude problem. Employees are labeled enthusiastic, resistant, or slow. That framing hides a more useful question: does a person have a safe tool, relevant examples, permission to experiment, access to usable data, and a manager who can discuss the work? A cautious employee in a regulated function may be behaving rationally. An enthusiastic employee pasting confidential data into an unapproved service is not necessarily ahead.

Uneven support produces a new form of workplace inequality. People with more technical confidence, more discretionary time, or stronger professional networks learn faster. They turn that advantage into better drafts and greater visibility. Colleagues doing frontline, sensitive, multilingual, or tightly controlled work may receive fewer useful tools even when their workflows carry the greatest friction.

A fair adoption program therefore starts with work, not personality. Map recurring tasks by role, sensitivity, frequency, frustration, and decision impact. Provide approved patterns that solve a real problem in each environment. Give employees protected practice time and a clear route for reporting a bad answer or requesting a new integration. Measure who benefits, not only how many prompts are sent.

This approach also prevents shadow AI from becoming the de facto innovation program. When official systems are irrelevant or difficult to use, employees will route around them. A credible internal product should make the responsible path the easiest path.

• Compare access and workflow relevance across roles, locations, shifts, and languages.
• Distinguish informed caution from blanket resistance.
• Offer approved examples for actual jobs rather than generic prompt classes.
• Create protected practice time instead of expecting learning after hours.
• Track benefit distribution alongside overall adoption.

Expertise becomes easier to reach and harder to recognize

Workplace AI can loosen the link between knowing where information lives and being able to use it. That is often healthy. A salesperson should not need a personal friendship with a product manager to find the current policy. A new engineer should not have to know which veteran remembers an old incident. Search, retrieval, and contextual assistance can make institutional knowledge less dependent on proximity and tenure.

But fluent output can make expertise harder to see. A polished response may combine current policy, stale guidance, an unsupported inference, and an invented detail in one confident paragraph. The experienced employee may notice the seam because they understand the domain, while a novice sees only professional language. If the organization mistakes presentation quality for judgment, it can weaken rather than democratize expertise.

The cultural answer is evidence-visible work. Important answers should expose the source, date, owner, scope, and confidence or limitation. Review should focus on the reasoning and evidence appropriate to the consequence. Employees need permission to say that an answer is plausible but not proven. Experts should be recognized for maintaining decision assets, defining exceptions, and teaching evaluation, not only for answering the same questions repeatedly.

A strong enterprise data management foundation helps turn scattered files into governed, attributable knowledge. That technical foundation supports a cultural shift from knowing a person who knows to being able to inspect what the organization knows and why it trusts it.

• Make citations, ownership, effective dates, and access boundaries visible.
• Reward experts for curating knowledge and defining exceptions.
• Teach employees to evaluate evidence, not just edit prose.
• Use higher review thresholds when consequences are greater.
• Preserve a clear distinction between generated synthesis and approved policy.

The apprenticeship ladder needs deliberate repair

Many careers are built through apparently low-value tasks. A junior lawyer summarizes records and gradually learns which details matter. An analyst cleans data and discovers how business definitions fail. A developer fixes small defects and learns the architecture's hidden constraints. If AI absorbs the first draft, lookup, or routine repair, the organization may save time while removing the experiences through which judgment develops.

Keeping every repetitive task would be an expensive form of nostalgia. The better response is to separate effort from learning value. Some tasks are pure friction and should disappear. Others are useful because they expose a novice to variation, consequences, and feedback. Those learning loops need to be redesigned rather than accidentally deleted.

A modern apprenticeship can ask a junior employee to form a view before seeing the AI answer, compare alternatives, inspect sources, explain a rejected recommendation, simulate an exception, or review a deliberately flawed output. Managers can use saved production time for case discussion and feedback. Progression criteria should include problem framing, verification, escalation, and domain reasoning instead of rewarding document volume.

The strongest signal is not whether junior staff use AI. It is whether they become independently capable of spotting when it is wrong. If output rises while error detection, customer understanding, or system knowledge falls, the company is borrowing productivity from its future talent pipeline.

• Inventory which entry-level tasks create judgment and which create only delay.
• Require prediction or framing before revealing assistance in selected training flows.
• Use exception cases to teach boundaries and trade-offs.
• Convert saved time into review, customer exposure, and coached practice.
• Assess error detection and reasoning, not prompt fluency alone.

Managers move from traffic controllers to designers of judgment

A large part of management consists of collecting updates, translating context, routing approvals, checking standard work, and reminding people about commitments. AI can reduce those coordination costs. It can also tempt an organization to automate them without asking why so many handoffs and status rituals exist.

When coordination becomes cheaper, a manager's value should move toward clearer goals, stronger decisions, coaching, conflict resolution, capacity design, and accountability for system quality. Managers need to define which evidence matters, where teams can act, what requires review, and how an exception is handled. They become stewards of a human-and-software operating system, not merely supervisors of task completion.

That transition can feel threatening. A manager whose status came from possessing information or controlling access may quietly resist tools that distribute both. Another may overuse dashboards and generated summaries to avoid difficult conversations. Leadership should make the new role explicit and train managers in workflow design, evaluation, data interpretation, and responsible delegation.

Do not use AI to produce a denser layer of managerial reporting. If every employee can generate more updates and every manager can generate more requests, coordination expands to consume the capacity the tool created. Remove rituals when the underlying information is already observable.

• Retire status meetings that duplicate trustworthy system state.
• Define manager accountability for AI-supported decisions and exceptions.
• Train managers to coach verification and reasoning.
• Measure decision latency and rework, not report volume.
• Watch for automated micromanagement disguised as visibility.

Decision speed changes the politics of work

AI can compress research, synthesis, and coordination. A team may prepare an options memo in hours rather than days. That sounds purely beneficial, but faster information changes who gets time to participate. A rapid recommendation can move through the organization before a frontline expert, affected community, security reviewer, or regional leader has a meaningful chance to challenge its assumptions.

Speed also exposes unclear authority. When an agent can prepare a refund, change a campaign, schedule a technician, or update a record, the old approval path may no longer make sense. Yet removing a click is not the same as transferring accountability. The organization must decide which actions can be automatic, bounded, sampled after execution, reviewed before execution, or prohibited.

Good systems introduce productive friction where consequences demand it. They show the evidence behind a decision, disclose uncertainty, capture dissent, require a reason for high-impact overrides, and route novel cases to a named owner. Low-risk work should flow quickly. High-impact work should become more legible, not merely faster.

Bizz enterprise software development can encode these decision rights into workflow, identity, audit, and review controls. The cultural benefit is clarity: people know where they can exercise judgment, where the system acts, and who owns the outcome when reality does not match the expected path.

• Classify actions by consequence, reversibility, novelty, and evidence quality.
• Protect participation time for people affected by high-impact decisions.
• Record rationale and dissent when a decision has material consequences.
• Use productive friction rather than universal approval queues.
• Name an accountable owner for every automated action class.

Psychological safety determines whether the system learns

Every production AI system will produce weak, surprising, or inappropriate output. The difference between a learning organization and a brittle one is whether employees can report that behavior without being blamed for slowing adoption or admitting their own uncertainty. If the launch narrative says the tool is revolutionary and the executive sponsor treats criticism as resistance, defects go underground.

Employees also need safety to reveal how work actually happens. Process documentation often describes the approved path, while experienced staff use informal checks, personal spreadsheets, and judgment built from exceptions. Treating those workarounds as embarrassment can erase the very context an AI product needs. Treating them as evidence can reveal broken systems and control gaps.

Build feedback into the point of work. Let a person mark an answer unsupported, a source stale, an action inappropriate, or an escalation missing. Ask for structured detail only when it changes diagnosis. Close the loop by showing what was fixed, what is under review, and what remains a known limitation. Anonymous reporting should be available for sensitive workforce impacts.

A high feedback count early in rollout is not automatically failure. It may show that people are paying attention and trust the organization enough to speak. The more dangerous dashboard is smooth adoption with no recorded exceptions and a growing private workaround culture.

• Make critical feedback a contribution to quality, not disloyalty to the program.
• Capture feedback in the workflow with enough context to reproduce the issue.
• Publish known limitations and remediation status.
• Give employees a route to challenge harmful monitoring or evaluation uses.
• Measure response quality and closure time, not just feedback volume.

AI can improve inclusion, but only if the baseline is designed for it

Translation, summarization, speech interfaces, adjustable explanations, and writing support can make work more accessible across language, disability, communication style, and experience. Employees who were excluded by dense documents or meeting-heavy coordination may gain a more usable way to participate. Distributed teams can carry context across time zones with less dependence on being in the right room.

The same systems can reproduce exclusion. Speech models may perform unevenly across accents. Summaries may erase minority views. Generated performance language can normalize one communication style. A recommendation trained on historical decisions may preserve unequal opportunity behind a polished interface. Accessibility features added after the core workflow often fail where the real action occurs.

Inclusion therefore belongs in the product requirements. Test across the actual languages, devices, roles, network conditions, and assistive technologies in the workforce. Preserve attribution and dissent in summaries. Let employees correct names, terminology, and identity-related errors. Do not infer sensitive traits to personalize work unless there is a legitimate, transparent, and consent-aware purpose.

The cultural question is not whether everyone receives the same interface. It is whether different employees can reach the same meaningful capability, understand the system's role, and challenge a decision that affects them.

• Include affected employee groups in discovery and acceptance testing.
• Test task outcomes, not only model benchmarks.
• Preserve minority views and attribution in meeting intelligence.
• Support accessible correction and appeal paths.
• Measure benefit and error rates across relevant workforce segments.

Employee monitoring is a design choice, not an inevitable feature

Workplace AI can observe documents, messages, calls, schedules, locations, keystrokes, customer interactions, and workflow events. Combining those signals can help route work or identify a service problem. It can also create a surveillance system that employees cannot understand or contest. Once people believe every draft and hesitation may become a performance signal, experimentation and candor decline.

The OECD's research on algorithmic management in the workplace is a useful reminder that software increasingly performs functions once associated with managers. Organizations should not treat the collection of data as neutral simply because it is technically available. Purpose, proportionality, access, retention, and employee rights all matter.

Start with a prohibition mindset for individual scoring based on weak proxies. Time in an application, response speed, sentiment, and generated-message volume rarely represent contribution on their own. For any workforce analytics use, document the legitimate purpose, minimum data, inference limitations, affected decisions, human review, retention, access, and appeal route. Consult legal, security, HR, and worker representatives appropriate to the jurisdiction.

Aggregate process signals can often solve the business problem without constructing a continuous portrait of an individual. If a queue is overloaded, improve capacity and routing before ranking employees by inferred effort. Trust is difficult to rebuild after a system quietly crosses the line from assistance to observation.

• Never reuse assistance telemetry for performance evaluation by default.
• Collect the minimum data required for a stated operational purpose.
• Disclose what is observed, inferred, retained, and shared.
• Provide human review and a meaningful appeal for consequential decisions.
• Prefer process-level diagnosis over individual surveillance.

Meeting culture reveals whether AI removes work or manufactures more of it

Meeting assistants can record, transcribe, summarize, assign actions, and retrieve prior context. Used well, they reduce note-taking and make absence less costly. Used carelessly, they create permanent records of tentative discussion, distribute inaccurate summaries, and generate action lists that no one agreed to own.

A meeting should declare whether recording is active, who can access the record, how long it is retained, and whether participants can correct it. Sensitive conversations may require no recording at all. An AI summary should distinguish decisions, proposals, unresolved questions, and attributed objections rather than flattening everything into a confident narrative.

More importantly, teams should ask whether the meeting still needs to exist. If the purpose is status transfer, trusted workflow state may replace it. If the purpose is interpretation, conflict, creation, or commitment, automation should prepare the room and preserve outcomes without pretending to replace the human event.

The cultural metric is not hours of meetings summarized. It is fewer unnecessary meetings, better participation in necessary ones, clearer commitments, and less time spent reconstructing what happened.

• Separate decisions, proposals, actions, objections, and unresolved questions.
• Require explicit owners and confirmation for generated action items.
• Set recording and retention rules by meeting sensitivity.
• Let participants correct summaries and attribution.
• Retire meetings whose only purpose was moving observable status.

Writing changes from a product of labor to evidence of thought

AI makes acceptable business prose abundant. Emails, proposals, policies, reviews, and project updates can become longer and more polished with little additional thought. That abundance lowers the value of surface fluency and raises the value of a clear position, specific evidence, accountable recommendation, and knowledge of the reader's decision.

Organizations need new norms. Generated text should not be judged by whether it sounds professional. Reviewers should ask what claim is being made, which facts support it, what uncertainty remains, who owns the recommendation, and what changed because the document exists. A short note with a decision can be better work than a comprehensive report no one can act on.

Authorship rules should match consequence. Routine rewriting may need no disclosure. A regulated filing, expert opinion, performance review, customer commitment, or public claim may require documented human review and source retention. Employees should not be forced to disclose disability-related or language-support use when the important requirement is accuracy and accountability.

Leaders can model the norm by refusing generated verbosity. Ask teams to state the decision first, expose evidence, label assumptions, and delete text that does not help the reader think or act.

• Judge documents by decision value rather than polish or length.
• Require sources and accountable review where consequences are material.
• Use disclosure rules that are proportional and privacy-aware.
• Train people to edit for truth, specificity, and reader need.
• Do not reward a larger volume of machine-assisted communication.

Performance systems must stop confusing output volume with contribution

When drafting, analysis, and coding accelerate, old productivity proxies become unstable. Lines of code, tickets closed, campaigns produced, documents written, and messages answered can rise without improving customer outcomes. In some cases the extra output creates more review, maintenance, and coordination for everyone else.

Performance should move toward outcomes, quality, learning, collaboration, and responsible use. That does not mean every role can be reduced to one business metric. It means the organization should trace how a role contributes, identify leading and balancing measures, and avoid rewarding local throughput that harms the system. A support agent should not be punished for escalating a risky case correctly. An engineer should not be rewarded for generating changes that increase incidents.

AI proficiency should also be defined thoughtfully. Prompt tricks are temporary. Durable capability includes framing a task, selecting trustworthy context, protecting data, evaluating output, handling exceptions, documenting a decision, and improving the workflow. Employees who identify where AI should not be used may demonstrate more maturity than those who maximize usage.

Never introduce an AI-adoption target into individual appraisal without a demonstrated relationship to the job's outcome. Forced usage creates performative prompting, hidden errors, and resentment.

• Replace raw output metrics with outcome, quality, and system-health measures.
• Credit responsible escalation and defect discovery.
• Assess durable evaluation skills rather than tool-specific tricks.
• Do not rank employees by prompt count or assistant time.
• Use balancing metrics to reveal downstream rework and risk.

Trust grows from inspectable boundaries, not human-like personality

A warm tone can make an assistant easier to use, but personality is not a substitute for trustworthiness. In the workplace, trust comes from knowing what the system can access, where its answer came from, what it is allowed to do, how recent the information is, when a person will review the action, and how a mistake can be corrected.

Anthropomorphic design can blur those boundaries. Employees may assume a conversational system understands intent, keeps a confidence, or possesses authority that it does not. Naming an agent like a colleague and saying it knows the company can encourage overreliance. The interface should make the system's role legible without forcing users to understand every technical detail.

For answers, show evidence and scope. For actions, show the target, important parameters, consequence, and approval state. For persistent memory, let users inspect and correct relevant stored context. For uncertainty, provide an appropriate route to a person or trusted source. Avoid false precision and theatrical confidence.

Bizz custom software solutions can shape these trust cues around the actual workflow instead of wrapping a generic chat interface around every job. The goal is calm, predictable assistance that earns reliance through behavior.

• Show what the system knows, can do, and cannot do at the moment of need.
• Distinguish a suggestion from an authorized action.
• Expose sources, scope, effective date, and limitations.
• Make memory and personalization inspectable where appropriate.
• Use human-like tone carefully and never to disguise uncertainty.

The strongest AI policy is a set of usable paths

A long acceptable-use document can establish boundaries, but it rarely tells an employee how to complete Tuesday's task. People need operational guidance: which approved tool fits the job, what data can enter it, what review is required, how output should be stored, and where to report a problem. Policy becomes culture only when the path is usable under time pressure.

Organize guidance by task and consequence. Drafting public material, summarizing internal notes, interpreting customer data, making an employment recommendation, generating code, and triggering a financial action require different controls. A universal ban is usually ignored; universal permission is reckless. A risk-tiered catalog gives people a concrete starting point.

Support the catalog with technical enforcement. Identity, access, data classification, retrieval permissions, action limits, logging, and retention should carry the policy into the system. Training should use realistic edge cases and explain the reason for the boundary. Employees are more likely to follow a rule they can connect to a person, obligation, or failure mode.

Review the policy as products and regulations change. Publish revisions in plain language. A stale policy teaches employees that governance is ceremonial.

• Publish approved patterns by task, data type, and consequence.
• Pair written rules with identity, access, logging, and action controls.
• Teach realistic edge cases for each function.
• Provide a fast route for requesting an exception or new approved pattern.
• Version the guidance and communicate meaningful changes.

Change the rituals that produce culture

Culture programs often stop at training and communications. More durable change comes from recurring management rituals. Add an evidence-and-limitations check to decision reviews. Ask for one eliminated step in process retrospectives. Include AI incidents and near misses in operational reviews. Invite employees to demonstrate a useful pattern and a failed one. Review benefit distribution by role each quarter.

Teams can hold a short workflow clinic where the person doing the job walks through a case, including unofficial tools and exception paths. Product, operations, data, security, and affected employees then decide whether to remove, redesign, or support a step. This is more valuable than collecting a list of broad automation ideas with no owner.

Leadership behavior is the strongest ritual. Executives should disclose uncertainty, correct AI-assisted work publicly when necessary, and avoid claiming savings that have not changed a budget or outcome. They should ask who bears new review work, whose expertise is missing, and what happens when the system is wrong. Those questions give everyone permission to build responsibly.

A practical AI workflow product team can then translate those rituals into a backlog, instrumentation plan, and governed release process. Cultural change becomes visible in the product and operating cadence rather than remaining an aspiration.

• Decision review: evidence, uncertainty, affected parties, and accountable owner.
• Workflow clinic: real case, exceptions, hidden tools, and removable steps.
• Operational review: failures, near misses, overrides, and unresolved feedback.
• Talent review: changed skills, learning loops, and equitable access.
• Portfolio review: realized outcomes, transferred work, and stopped experiments.

Measure culture through behavior and consequence

Culture is often measured with a broad sentiment question or adoption percentage. Both can help, but neither explains whether work improved. Build a small measurement system that connects employee experience, behavior, workflow, quality, risk, and outcomes. Use qualitative research to understand why the numbers move.

Leading signals include access to relevant approved tools, manager support, confidence in reporting a problem, ability to verify an answer, participation in learning, and time spent on avoidable coordination. Behavioral signals include source inspection, appropriate escalation, reuse of approved patterns, correction rate, and cross-team contribution to shared knowledge. Outcome signals depend on the workflow: resolution, rework, cycle time, quality, customer effort, safety, or released capacity.

Segment carefully. An average can hide that senior office workers gain time while frontline employees acquire more review work. Compare roles, locations, shifts, tenure, language, and other relevant groups while preserving privacy and avoiding simplistic individual scoring. Look for benefit transfer: whose time was saved, whose burden increased, and whether the saved capacity produced an observable change.

Do not set a target that everyone must trust AI more. Healthy trust is calibrated. In some workflows the desired cultural outcome is faster confident use; in others it is more frequent verification or escalation.

• Access: can each role use a relevant, approved capability?
• Agency: can employees inspect, correct, decline, and escalate?
• Learning: are verification and domain judgment improving?
• Workflow: did queues, handoffs, rework, or customer effort change?
• Equity: who receives benefit and who absorbs new labor or risk?

A ninety-day culture experiment should be small enough to learn from

Choose one workflow with recurring friction, a willing manager, reachable data, and an outcome that can be observed. Avoid starting with a company-wide assistant whose value dissolves into self-reported anecdotes. Include employees who perform, receive, review, and are affected by the work.

During the first three weeks, observe cases and establish a baseline. Document the official path, exceptions, hidden coordination, current evidence, decision rights, and employee concerns. Define prohibited uses and a clear stop condition. Select a narrow behavior to change, such as preparing an evidence-backed case brief or routing a request with the right context.

In weeks four through eight, release to a small cohort. Hold weekly workflow clinics, inspect errors, and adapt the interface, knowledge, and controls. Give managers a script for discussing assistance without pressuring usage. Track who uses the capability, who cannot, and where review labor moves. Preserve a comparison where feasible.

In weeks nine through thirteen, test durability. Remove intensive support, examine novel cases, and verify whether the behavior survives. Decide to scale, revise, contain, or stop based on workflow and cultural evidence together. A valid result may be that the technology works but the operating conditions are not ready.

• Start with one unit of work and one observable cultural behavior.
• Baseline the formal and informal workflow before building.
• Include affected employees in design and weekly review.
• Track transferred labor and uneven access from the beginning.
• Make scale conditional on outcome, trust, learning, and control evidence.

Twelve questions leaders should be able to answer without a slide deck

A mature workplace AI program can explain itself in operational language. Leaders do not need to know every model detail, but they should know which work is changing, why it matters, who owns the outcome, and how employees can influence the system. If the answers exist only in a steering committee presentation, they are not yet part of the culture.

Use the questions below in an executive review, manager forum, or employee listening session. Differences between answers are useful. They reveal where policy, product behavior, incentives, and lived experience have drifted apart.

The objective is not consensus on whether AI is good. It is shared clarity about the conditions under which a particular system helps people and the organization perform responsibly.

• Which recurring job or decision is changing, and whose outcome should improve?
• What work disappears, what new work appears, and who receives each effect?
• What knowledge can the system use, and how is its authority kept current?
• Which actions can it take, and who remains accountable for them?
• How can an employee inspect, correct, decline, or escalate its output?
• Which groups have less access or experience worse performance?
• What entry-level learning loop might be removed, and how will it be rebuilt?
• Can assistance telemetry affect performance evaluation, and under what safeguards?
• What does the manager do differently after this capability launches?
• Which meeting, report, approval, or queue will the organization remove?
• What evidence would cause the team to pause or stop the system?
• What outcome has changed beyond usage, satisfaction, or theoretical hours?

The cultural goal is better agency, not maximum automation

A humane AI culture is not one in which people avoid effort or delegate every judgment. It is one in which employees can reach the context they need, understand the systems shaping their work, exercise meaningful discretion, learn faster, and spend less energy navigating avoidable bureaucracy. Automation is useful when it supports those conditions and dangerous when it quietly removes voice or accountability.

That goal requires choices. Some friction should be removed. Some should remain because it protects deliberation, consent, safety, or learning. Some roles will change substantially, and leaders owe employees more than optimistic language: they owe transparent plans for skills, transitions, evaluation, and participation.

Companies that handle this well will not be those with the highest prompt count. They will be the ones that make reliable knowledge easier to use, redesign work around outcomes, preserve human challenge where it matters, and learn openly from failure. Their culture will show up in calmer operations and clearer responsibility, not in louder claims about the future.

• Use automation to expand meaningful capability, not merely activity.
• Preserve human agency where values, rights, safety, and ambiguity are involved.
• Make transitions and learning commitments specific.
• Let employees participate in the design of systems that shape their work.
• Treat trust as an outcome earned through inspectable behavior.