Leading with AI

Yes. They can help monitor adoption, cost, training needs and unusual use, and they can flag a workflow worth examining more closely. They are diagnostic telemetry, not a productivity outcome, and they never earn a green status on their own. From Stop Counting Prompts. Measure the Work That Improved

It can, in two ways. People may perform for the visible AI and the audience around it rather than think out loud, and the AI's fluent, confident answers can anchor a discussion so that quieter dissent never surfaces. Neither is inevitable. A leader who treats the shared AI as a first draft to argue with, names disagreeing with it as expected, and models correcting it in the open turns the visible surface into a place that strengthens candour rather than suppressing it. From A Shared Team AI Identity Changes How You Lead

AI can structure approved records into draft claim cards, expose missing denominators and flag estimates presented as results. A human who understands the work must select the grade, accept the uncertainty and own the recommendation. The drafting model never adjudicates a grading dispute, provides assurance or approves the initiative. From Your AI Update Needs an Evidence Grade

No. The skill's guardrails label the recommendation as a draft for the decision-maker to test and refuse to state a final decision. The memo organises the options, evidence and risks so a named human can decide. Accountability for the decision stays with the leader, never with the tool. From Your First Skill File: A Decision Memo On Demand

It can draft them; it cannot certify them. A fluent recap is not proof of an accurate record, and the OAIC notes generative AI can produce inaccurate or false results. The person with authority for the decision, or the chair against the committee's governance rules, confirms the final wording. The AI operator, meeting organiser and model are not default decision owners. From The AI Joined the Meeting. The Decision Record Still Belongs to You.

Often, yes, and perception is what matters. In the Coman and Cardon study of 1,100 professionals, employees who sensed heavy AI assistance rated the same supervisor messages as far less sincere. Messages that arrive suspiciously fast, unusually polished and slightly off the leader's normal register read as low effort. Even when nobody is certain, the doubt itself does the damage on relationship-bearing messages like praise and feedback. From Don't Let AI Write the Messages That Build Trust

Not as the sole judge. AI can organise options against locked fields, quote the supplied evidence and identify blanks. A qualified person verifies every classification and records the survival decision. The prompt should also forbid the model repairing an option while grading it, because a quietly repaired option hides the very defect the gate exists to catch. From AI Made Drafting Cheap. Build a Rejection Rubric.

For organisations covered by the Privacy Act, the OAIC says personal information captured for AI-generated minutes must be reasonably necessary for the organisation's functions or activities, and collecting sensitive information requires consent unless an exception applies. Because meeting content is uncertain, the OAIC says seeking consent is best practice, and clear notice is relevant to whether the collection is lawful and fair. From The AI Joined the Meeting. The Decision Record Still Belongs to You.

No. Projects are available on every ChatGPT plan including Free, and each project takes its own instructions, so the file runs as project instructions plus knowledge files. Claude projects exist on every plan including Free. Native Skills features add convenience on the plans that include them, but they are not required. From Your First Skill File: A Decision Memo On Demand

No. Survival is not approval. A surviving option advances to source verification, accessibility checking and whatever legal, compliance and authorised communication reviews the work requires. The rubric narrows the field; an authorised person still owns the decision and its consequences. From AI Made Drafting Cheap. Build a Rejection Rubric.

No, if the friction is selective. Adding a checkpoint everywhere rebuilds the old bureaucracy and wastes the gain. The skill is to add friction only where reversibility, compounding or judgement warrant it, and to let the reversible, low-stakes work run fast. Targeted friction protects the few steps that matter without taxing the many that do not. From The Leader's Case for Slowing an Agent Down

No. AI surfaces the case against. The leader makes the call and carries it, and accountability does not transfer to a tool. A model told to argue will manufacture some thin objections, and a fluent objection is not a correct one, so you weigh what it raises rather than obey it. From Make AI Disagree With You Before You Decide

It can. Safe Work Australia identifies lack of role clarity, poor support and insufficient training as risks when AI and digital technologies are introduced, and warns that automating routine tasks can leave workers with more complex, cognitively demanding work or with more of their work focused on reviewing AI output. An apprenticeship should prepare people for that residual work, not merely teach faster generation of routine output. From AI Made the New Starter Faster. You Still Owe Them an Apprenticeship.

The evidence is promising but limited. A 2025 ACM study with 18 employees found an AI reflection prototype helped clarify meeting purposes and change preparation, but it was a small technology probe. A preregistered field experiment across 361 employees and 7,196 meetings found a brief pre-meeting goal prompt produced no statistically significant improvement in reported meeting effectiveness. Preparation helps; a prompt alone is not a meeting system. From The AI Joined the Meeting. The Decision Record Still Belongs to You.

Not in this form. APRA's 30 April 2026 letter to industry expects ownership and accountability across the whole AI lifecycle, from design through deployment and monitoring to decommissioning, and human involvement in high-risk decisions. The one-page stop authority is a practical leadership design that gives lifecycle ownership a named person at the moment it is actually tested. From Who in Your Organisation Can Actually Stop the Model?

No. The material service provider test turns on reliance for a critical operation or exposure to material operational risk, subject to the standard's specified categories and APRA's powers. It does not capture every contractor, labour-hire firm, outsourced team or AI supplier. Where an arrangement is material, the handshake supplies operational evidence for the entity's broader service-provider controls, but it does not determine materiality or transfer accountability to the provider. From Your AI Rules Stop at Payroll. Your Risk Does Not.

No. A provider may use its own identity, records and incident systems. What both sides must demonstrate is the agreed outcome for the specific use case, with named people and evidence. An external quality check is not equivalent unless your accountable work owner understands what it checks and can reject the result. From Your AI Rules Stop at Payroll. Your Risk Does Not.

No. Risk, customer impact and regulatory obligations can justify different controls, and a team may reasonably use AI less because its tasks are unsuitable. Fairness requires each team to have a credible, properly controlled route from access to useful practice, not the same permissions or the same usage rate. From Your AI Advantage Cannot Belong to One Team

Three studies give it shape. A 2026 Organization Science comparative field study found experts at one planning organisation amplified an AI tool's apparent precision while another moderated it, keeping uncertainty visible. A 2025 PNAS paper reported a social evaluation penalty for disclosed AI use across four preregistered experiments. A CHI 2025 survey of 155 knowledge workers found credit was assigned differently by contribution type, amount and initiative. None of them measured Australian financial services directly, and the article treats each within its limits. From The Polished Output Hid the Expert Who Made It Safe

No. The immediate goal is to remove low-value activity and redirect capacity to work that changes an outcome. Workforce changes are a separate decision that carries its own consultation, legal review and accountable sign-off. Using a work-kill review to reverse-engineer a predetermined headcount answer corrupts the method and the trust it depends on. From Use AI to Kill Work Before You Accelerate It

Not as a general rule. That result came from one customer-support setting, using one real-time suggestion tool, with 89 per cent of agents located outside the United States, mainly in the Philippines. It does show why an average productivity lift can hide different contributions. The study notes that top performers supplied many examples from which the system learned. From Do Not Grade Your Experts on Prompt Speed

No. It is a conversion mechanism for a specific failure pattern, not a general operating-rhythm meeting, and it does not replace formal incident response, notification, remediation, consultation or root-cause processes where they apply. One severe event may require immediate escalation instead. From Run the AI Retro Before the Error Becomes the Process

No. Section 7B(3) of the Privacy Act 1988 limits the private-sector employee records exemption to an act or practice of an organisation that is or was the individual's employer, directly related to the current or former employment relationship and an employee record held by that organisation and relating to the individual. The Office of the Australian Information Commissioner states the exemption does not cover contractors and subcontractors handling another organisation's employee information, and that a contractor collecting employee records must comply with the Australian Privacy Principles, including the APP 5 notice requirements. From Your AI Rules Stop at Payroll. Your Risk Does Not.

No, and the article does not claim it. In the Boston Consulting Group field experiment, AI assistance improved speed, completion and quality across 18 tasks inside the model's capability frontier. The 19 percentage point drop came from a single task deliberately designed to sit outside it. That is a demonstration of why fluency cannot substitute for evidence, not an error-rate forecast for your work. From AI Made Drafting Cheap. Build a Rejection Rubric.

Jobs and Skills Australia's national Gen AI Capacity Study concludes that generative AI is more likely to augment human work than replace it, and its labour-market summary says displacement is limited so far, with most impacts involving upskilling, redeployment and evolution of roles. That is a labour-market finding, not a guarantee for any employee, employer or occupation, and "so far" is the boundary. From Stop Predicting Their Jobs. Map Their Next Skills.

Partly, and it is worth being honest about that. The figures come from an Accenture survey of 1,928 executives across 28 countries conducted in December 2025 and reported in MIT Sloan Management Review. It is a global sample, not an Australian one. The direction is credible and consistent with Australian policy settings, but any specific percentage should be read as international evidence rather than a local measurement. From Sovereign AI Is a Strategy Choice, Not a Compliance Box

Not generally. Reading everything is the problem compression was introduced to solve, and reverting relocates the cost rather than removing it. The practical position is selective: accept compression for the routine, demand the source and a second reading for the consequential, and keep at least one recurring channel that nothing summarises. From The Summary Kept the Number and Lost the Caveat

No, and the research does not say that. It says where AI enters matters. Writing in MIT Sloan Management Review in July 2026, Boussioux, Doshi, Hauser and Hosanagar report that AI in idea generation consistently reduced diversity, while AI in idea selection preserved variety at levels comparable to human-only work. Wharton's Human-AI Research group, where Hosanagar sits, makes the same point. So the move is sequencing, not restriction. Let people diverge first, then bring AI in to pressure-test, cost and choose. Nobody gives up the tool. From AI Is Narrowing Your Team's Idea Pool. Only You Can See It.

It can, if the leader consumes AI output instead of evaluating it. The risk is rubber-stamping a confident answer. It sharpens judgement when the leader uses AI for breadth and speed, then does the harder human work of questioning the output, weighing it against values and context the model lacks, and owning the decision. The discipline is to stay the evaluator, not become the audience. From You Are No Longer the Smartest Person in the Room

Ask what risk the step addresses, who owns that risk, what evidence the step creates and whether the control is required by law, policy, contract or professional standard. A step that annoys the team may still be protecting a legal, safety, financial or customer outcome. Involve the qualified risk owner before removing it, and record the decision. From Use AI to Kill Work Before You Accelerate It

Never paste real personal, claim, health or incident data into a model that is not an approved enterprise instance. Use placeholder tokens such as [TEAM], [ROLE], [SITE] and [DATE], or fill them with non-sensitive descriptors only. In financial services, use your organisation's approved enterprise tool, not a personal account, for anything touching the business. From Set the AI Norm: Your Team Copies How You Use It

Use a rubric defined before the trial, source verification, human sampling, defect or correction rates and severity levels. Keep the standard independent of the model producing the work, and track severity as well as frequency, because one material error can outweigh fifty clean drafts. From Stop Counting Prompts. Measure the Work That Improved

Ask two questions per workflow: how bad is it if this goes out wrong, and can you take it back. Low harm and easily reversed is spot-check. Real harm or hard to reverse is full human review. Severe, regulated or irreversible is two-person or named sign-off. Write the answer down. From The Review Tax: AI Adoption Is Done, Now Design the Checking

Run a five-step method this week. List your team's recurring decisions, usually ten to thirty. Name a specific role that decides each. Place AI on the ladder at inform, recommend or act within bounds. Engineer override and escalation paths for anything AI recommends or acts on. Write it on one page and revisit it. From Decision Rights Are the Leadership Job AI Just Made Urgent

Use AI to prepare, never to deliver. Draft your own three lists (what is changing, what you can commit to, what you cannot promise), have ChatGPT, Claude or equivalent turn them into plain talking points, then run a red-team pass that attacks every overpromise and buzzword. You say the words in person. The tool never appears in front of your team. From Talk to Your Team About AI Before the Rumours Do

Four steps in order. Brief the role and stance, never a real person. Tell the model not to be agreeable or concede quickly. Run your real opening line three ways, against a defensive, a withdrawn and an agrees-too-fast reaction. Then type STOP to switch the model from counterpart to coach for a short debrief. From Rehearse the Hard Conversation Before You Have It

Make four behaviours visible, starting at your next meeting. Model your own AI use in the open, including the errors you fixed. Set the norm explicitly: what is encouraged, what is off-limits, what good looks like. Make space to experiment without penalty. Hold the quality bar so the standard does not drop. From Set the AI Norm: Your Team Copies How You Use It

Keep the markdown file as the source of truth in your shared document store. On Claude Team and Enterprise plans an organisation owner can provision skills for everyone, and ChatGPT's native Skills feature is generally available on Business, Enterprise, Healthcare and Edu. On individual plans, each member installs the file as project instructions plus project knowledge. From Who Owns Your Team's Skill Library?

Drill it with fictional material and measure behaviour. Can a person find the route, stop the output without the sponsor's permission, and get a named owner and time. Then review monthly for control health: unacknowledged receipts, expired holds, repeated source failures, challenges closed without retest, and high-AI-volume teams with no recorded exceptions. From Silence Is Not Human Oversight

Do not collapse the pairing card into a single score. Record tool contribution and judgement contribution separately, alongside the recorded test run, the trap log and the human-reviewed rubric. When you report, separate the workflow improvement, the tool method and the expert judgement, then decide what changes, what needs another test and what remains human-led. From Do Not Grade Your Experts on Prompt Speed

Ask one question of each recurring task: does doing this build a capability we need to protect? Most tasks fail that test and should be accelerated with AI. A few pass it, the analysis a junior needs to reason through to become senior, the judgement calls that keep an expert sharp, and those you deliberately keep human, or structure so the person does the thinking and AI does the support. Name which tasks are development reps, protect them, and make the trade-off explicit rather than letting efficiency quietly decide it. From Protect the Reps: Lead So AI Does Not Deskill Your Team

A 2026 Harvard Business Review study found that when organisations roll out AI, middle managers absorb three new responsibilities on top of unchanged delivery pressure: validating and checking AI outputs, coaching their teams to use the tools, and managing the organisational change. Because the rollout is framed as a technology project, none of that work is planned, resourced or acknowledged, so it lands as invisible overload on the manager layer. From Your AI Rollout Landed on Your Managers. Resource It.

It removes the knowledge advantage. When anyone can get a competent synthesis of a market, a competitor or an option in seconds, being the most informed person in the room stops being the source of a leader's value. A 2026 Journal of Business Research study argues the shift is human rather than technological: leadership moves from holding information to framing the real question, judging what is right, and taking responsibility for the outcome, which AI cannot do. From You Are No Longer the Smartest Person in the Room

APRA's 30 April 2026 industry letter said regulated entities should establish ownership and accountability across the AI lifecycle, retain human involvement and accountability for high-risk decisions, and train staff on AI limitations and secure use. A verified ledger of judgement-changing interventions is one way to evidence that human involvement, although the letter is not a recognition framework and does not replace an entity's own obligations. From The Polished Output Hid the Expert Who Made It Safe

Carefully. It is three teams in one firm, in software engineering, measured through developer tooling. That is a real longitudinal study rather than a vendor survey, which is why it is worth reading, but it is not evidence about legal, claims, HR or finance teams. Treat the design lesson as transferable and the numbers as specific to the setting. From Set the Evaluation Window Before the Pilot Starts

A decision-rights map answers who decides, placing AI on a ladder of inform, recommend or act. A delegation charter answers which recurring tasks the team hands to AI and which stay human. One is about authority over decisions, the other about the work itself. They complement each other; a task on the charter may still have its decisions governed by the map. From The AI Delegation Charter Your Team Actually Needs

A private chat is a tool one person uses out of sight, so its errors and its influence are invisible to the team. A shared AI identity, such as Anthropic's Claude Tag, is one Claude that a whole channel uses, with its work visible to everyone and pickable up mid-task. That visibility is the change a leader manages. The AI now shapes how the team talks in front of each other, not just how one person works alone, so credit, blame and candour all move onto the shared surface. From A Shared Team AI Identity Changes How You Lead

A decision-rights map answers who is allowed to decide. A review tier answers how much checking a finished output gets. This answers a third question: where inside a fast, unattended run a human touchpoint has to exist at all, and which workflows deserve one. The three are complements, not substitutes. From The Leader's Case for Slowing an Agent Down

Long enough to capture normal variation and meaningful exceptions. Four to eight weeks may suit a frequent workflow, while lower-volume or seasonal work may need longer. Set the period in the measurement contract before the trial starts, not after the results arrive. From Stop Counting Prompts. Measure the Work That Improved

Fewer than you fear and more than one. Most organisations can group their AI uses into three or four data classes and attach a sovereignty position to each class rather than to each system. The work is defining the classes once and applying them consistently, which is the same discipline that makes model routing governable. From Sovereign AI Is a Strategy Choice, Not a Compliance Box

Quarterly suits most teams. Each owner runs the maintain prompt against their skill, checks the file still matches how the team actually works, then keeps, updates or retires it. A stale skill is worse than none because it standardises yesterday's process with today's confidence. From Who Owns Your Team's Skill Library?

Set a quarterly review at the latest, and give each line an event trigger that forces an earlier look: a new tool, a new regulation, or a near-miss. A charter that never changes is a poster. A charter that gets revisited on a cadence and on real events is a practice, and it is the only version worth building. From The AI Delegation Charter Your Team Actually Needs

Mixed, and worth knowing precisely. Doshi and Hauser is peer-reviewed, published in Science Advances in July 2024. Boussioux and colleagues is peer-reviewed, published in Organization Science in 2024. Hosanagar and Ahn is an arXiv preprint from December 2024 and has not been peer-reviewed. All of it comes from lab and crowdsourcing experiments on short stories, cartoon captions and business-idea challenges, run with crowdworkers and solvers, not from leadership teams doing real planning work. Applying it to your planning cycle is a reasoned extrapolation, not a measured finding. From AI Is Narrowing Your Team's Idea Pool. Only You Can See It.

That assumption is where the trouble starts. Framing AI as a headcount saving ignores that the technology creates significant new supervisory and coaching work in the near term, and that work concentrates on managers. A leader who cuts the layer that is absorbing the transition can lose the very capacity that makes the adoption succeed. In the short run AI expands the manager's job before it shrinks anything. From Your AI Rollout Landed on Your Managers. Resource It.

Yes, for most of it. Informational messages such as status updates, logistics and policy reminders exist to be clear and correct, and AI drafts them well. Messages carrying decisions benefit from AI pressure-testing, provided the leader stays the author of the judgement. The line sits at relationship-bearing messages, meaning recognition, personal feedback and apologies. The research points at those as the messages where perceived AI involvement collapses trust. From Don't Let AI Write the Messages That Build Trust

No. You rehearse against a described archetype only, never a real person's name, performance notes or medical information. In Australia, restructure and performance conversations engage a duty to manage psychosocial hazards, so keeping real personal data out of the prompt is both a privacy reflex and a safety practice. From Rehearse the Hard Conversation Before You Have It

No. It is a scheduled, disclosed development exercise built on prepared, fictional work samples. It does not inspect private chats, count prompts, infer attitude from tool logs or let AI recommend employment action. If an observed gap may become a performance matter, leave the calibration frame and use the organisation's ordinary human-led process with appropriate advice. From The AI Enthusiast and the AI Refuser Need the Same Test

The opposite. It is an argument against measuring one thing. The authors conclude that multi-dimensional assessment frameworks such as SPACE are essential to capture the effect, precisely because a single-metric evaluation would have shown nothing. From Set the Evaluation Window Before the Pilot Starts

No, and the Australian Government's own framework makes the point well. The Hosting Certification Framework describes sovereignty in terms of ownership structure, liability, supply chain and transparency arrangements, not location alone, and its highest certification level is available only to service providers that allow the government to specify ownership and control conditions. Location is one input. Control is the question. From Sovereign AI Is a Strategy Choice, Not a Compliance Box

No. The point is not to slow AI down, it is to keep building the capability the organisation runs on. Harvard Business Review has warned that leaning on AI to the generic standard can strip an organisation's distinctive judgement, leaving it more efficient yet less legitimate, and California Management Review argues the tacit knowledge in your people's judgement is the real competitive moat. Protecting the reps is a capability investment, not a brake. Most work can and should be accelerated. The leader's job is to protect the small share that builds the expertise you will need later. From Protect the Reps: Lead So AI Does Not Deskill Your Team

No, and that is why it is harder to catch. A hallucination introduces something that is not in the source, so a check against the source finds it. Decontextualisation keeps only what is in the source. Every sentence verifies. The distortion sits in what is absent, and absence does not trigger a fact check. From The Summary Kept the Number and Lost the Caveat

It bears on one. Safe Work Australia's model Code of Practice on managing psychosocial hazards at work, published in July 2022, names lack of role clarity as a psychosocial hazard, defined as uncertainty, frequent changes, conflicting roles or ambiguous responsibilities and expectations. It separately names poor support, which includes inadequate training, tools and resources for a task. Making a person answerable for checking work they were never given the time or standard to check sits inside both descriptions. This is general information, not legal advice, and duties vary by jurisdiction. From When the AI Gets It Wrong, Who Carries It?

No. A promotional deadline is information about the vendor's pricing calendar, not evidence that your team is ready. Deciding to beat a deadline front-loads the cost of getting governance or fit wrong. Decide on real need, switching cost and governance readiness, and let the window close if those are not settled. From A Deadline Is Not a Decision: Greenlighting AI Before the Free Window Closes

Publish only formally authorised protection. The workflow owner should obtain approval from the relevant people, WHS and employee-relations functions for whether a good-faith stop on an identified trigger is excluded from avoidable-delay measures, who owns the investigation and how workload targets treat an authorised hold. Any exception for deliberately false reporting needs situation-specific advice. From Silence Is Not Human Oversight

Yes. Declaring the boundary is itself a trust signal. Telling your team you use AI for updates and logistics but never for recognition or feedback shows the relationship is taken seriously, and it removes the guessing game the research shows employees are already playing. The boundary becomes part of how you lead rather than a secret you are keeping. From Don't Let AI Write the Messages That Build Trust

Generally avoid usage quotas or rankings. They reward low-value activity, punish people who have no suitable use case or who are protecting sensitive data, and create privacy and employment risks. Compare workflow outcomes instead, and discuss individual support in context. From Stop Counting Prompts. Measure the Work That Improved

Yes. The answer is not to delay access to approved AI. It is to stop confusing assisted output with independent capability. Give new starters the tool, then design the work so they must expose their reasoning, encounter edge cases and receive expert feedback before any authority expands. That is operating design, not an optional learning activity squeezed around production. From AI Made the New Starter Faster. You Still Owe Them an Apprenticeship.

Three patterns undo progress. The absent sponsor announces AI matters then never engages, leaving a vacuum. The secret user uses AI privately, so all the modelling value is lost. The enforcer mandates use and sets targets, producing compliance and resentment rather than absorption. The lever is permission and example, not a quota. From Set the AI Norm: Your Team Copies How You Use It

Decision rights define who decides, who is consulted and who is merely informed, and on what basis. AI has made them urgent because it now moves from informing to recommending to acting inside the decision flow. Where no explicit rule exists, the system fills the gap and quietly assumes the right itself. From Decision Rights Are the Leadership Job AI Just Made Urgent

Machine gate: output volume and failure behaviour stable enough to plan. Verification gate: the approved human review keeps pace without deferred checks, falling quality or unplanned overtime. Exception gate: exception volume, resolution time and oldest-case age stable within the team's limits. Capability gate: protected learning and calibration time actually happened. Only when all four hold over a representative period has usable capacity moved, and even then the workforce decision stays human. From An Agent Queue Is Not a Workforce Plan

Human-led work, where a person still frames the issue, weighs evidence or carries accountability. AI-assisted work, described as a bounded activity with permitted inputs, required source grounding and a human acceptance test. New oversight work created or expanded by AI use, such as testing outputs, resolving exceptions and monitoring quality. And adjacent option evidence, which records a plausible direction rather than a destination. From Stop Predicting Their Jobs. Map Their Next Skills.

Machine throughput (inputs accepted, outputs produced and abandoned, processing time), human verification (outputs reviewed under the existing control, review minutes, rejections and rework), exceptions (cases the system cannot process safely, escalations, reopened work and the age of that queue), and capability-building (protected hours for training, supervised practice, calibration and test-case maintenance). Measure all four over the same workflow and period. From An Agent Queue Is Not a Workforce Plan

Notice, explain, challenge and own. Notice the facts, assumptions and missing information that could change the outcome. Explain the process, evidence and reasoning in their own words. Challenge the AI output where it is unsupported, incomplete or too certain. Own the decision by making or escalating it within their actual authority and recording why. These acts turn an output into evidence of capability. From AI Made the New Starter Faster. You Still Owe Them an Apprenticeship.

Approved tool and identity, data boundary, human review, authorised record, escalation, and incident and exit. For each one, record the internal requirement, the external commitment, the proof supplied, any gap, two named owners and an expiry or review date. A contract clause is an input to that record. Evidence that the control actually operates is the handshake. From Your AI Rules Stop at Payroll. Your Risk Does Not.

Three comparable scenarios from one task family. AI-required tests whether the person can brief the approved tool, verify its output and own the result. AI-optional tests selection judgement: they choose the method and record one sentence explaining why. AI-prohibited tests independent domain capability and the ability to respect a boundary. One rubric covers all three: quality, evidence, risk, judgement and explanation. From The AI Enthusiast and the AI Refuser Need the Same Test

Spot-check, for low-stakes internal work the reviewer samples rather than reads in full. Full human review before it leaves the team, for client-facing work, numbers, or anything where the raised bar bites. Two-person or named sign-off, for regulated, legal, financial, safety-critical or irreversible output. From The Review Tax: AI Adoption Is Done, Now Design the Checking

Authority without information, where the named person cannot see the signals that would justify a stop. Information without authority, where the people closest to the system can only raise concerns. And authority and information with no fallback, where nobody stops the system because the manual process was retired and there is nothing to stop it to. Each failure has a different fix, and misdiagnosing which one you have wastes months. From Who in Your Organisation Can Actually Stop the Model?

Over-trust, where a busy leader lets AI climb the ladder unchecked until recommendations become decisions and bounds disappear. Under-use, where a wary leader keeps everything at inform, redoes every suggestion by hand, and mistakes the drag for prudence. A decision-rights map steers between both by forcing a deliberate rung for each decision. From Decision Rights Are the Leadership Job AI Just Made Urgent

Workspace agents are OpenAI's evolution of custom GPTs. Powered by Codex, they run in the cloud, can be shared across a team, connect to apps such as Slack, Google Drive, Salesforce and Notion, and can run on a schedule or via an API. OpenAI has said it will deprecate the organisation custom GPT standard and require teams to move to workspace agents. From A Deadline Is Not a Decision: Greenlighting AI Before the Free Window Closes

AI cannot own the why: the choice of goal, the weighing of values, and the judgement about what the organisation is for. It cannot own accountability when a decision affects livelihoods, customers or integrity, since a named human must answer for it. It cannot own the decisions that define a team's culture. From Decision Rights Are the Leadership Job AI Just Made Urgent

Three things. It cannot decide which problem is the real one to solve, because that requires context and priorities it does not hold. It cannot decide what is right, because that requires values and accountability. And it cannot own the outcome, because responsibility cannot sit with a model. AI can illuminate what is possible. Deciding what to do with that, and answering for it, stays with the leader. From You Are No Longer the Smartest Person in the Room

OpenAI's free access period for workspace agents ends on 6 July 2026. From that date, agent runs invoked inside ChatGPT move to credit-based pricing on the Business, Enterprise, Edu and Teachers plans. The free period was originally set to end on 6 May and was extended to 6 July when the feature reached general availability. From A Deadline Is Not a Decision: Greenlighting AI Before the Free Window Closes

An evidence line with six fields: the exact task boundary, the observed baseline, the assisted end-to-end result, the full human cost including preparation, verification and corrections, the quality and control result, and a confidence label of observed, sampled or self-reported. Reject the dividend when periods do not match, the definition of done changed, review time is missing or quality fell outside the existing limit. From AI Saved the Time. Your Calendar Will Take It Back

APRA's 30 April 2026 letter, based on targeted engagement in late 2025 with selected large banks, insurers and superannuation trustees, observed that assurance practices were not keeping pace with the scale, speed and complexity of AI and noted an overreliance on vendor presentations and summaries. ASIC's REP 798 found the potential for a governance gap after reviewing 624 consumer-impacting AI use cases reported by 23 licensees as at December 2023, while stating its sample was not representative. From Your AI Update Needs an Evidence Grade

The Digital Transformation Agency's evaluation of the non-randomised whole-of-government Microsoft 365 Copilot trial reported that 41 per cent of 807 post-use respondents believed Copilot enabled more time on higher-value or complex tasks. The evaluation says the methodology relied on self-assessment, which may under or overestimate benefits, particularly time savings. Record it as perceived reallocation, not a measured causal capacity estimate. From AI Saved the Time. Your Calendar Will Take It Back

Across 5,172 customer-support agents and about three million chats at one software company and its subcontractors, AI assistance increased successful resolutions per hour by 15 per cent on average, with larger gains among less experienced and lower-skilled workers. It had little productivity effect for higher-skilled or more experienced workers, and the authors found small but statistically significant declines in chat quality among the most skilled agents. From Do Not Grade Your Experts on Prompt Speed

Three agile teams at a large technology consulting firm, over approximately thirteen months, using quantitative telemetry from Jira, SonarQube and Git alongside qualitative surveys, comparing pre-adoption and post-adoption sprints. The tools studied were internal GPT tools and GitHub Copilot. From Set the Evaluation Window Before the Pilot Starts

Jobs and Skills Australia finds generative AI is more likely to augment human work than replace it. Safe Work Australia's AI guidance says automation can leave workers with more complex or cognitively demanding tasks and more output-review work, and names insufficient training, unclear roles and work intensification as potential hazards. APRA's April 2026 letter notes AI moving into claims triage and loan processing and expects human involvement and accountability for high-risk decisions. From An Agent Queue Is Not a Workforce Plan

Measured means a defined method applied to a stated population or sample during a stated period, with an inspectable source and a reproducible calculation. Observed means a named human directly saw a result without a complete denominator. Estimated means the result is calculated from assumptions, sampling or extrapolation with those inputs exposed. Asserted means the support is presently opinion, expectation or a vendor statement. Asserted does not mean false. From Your AI Update Needs an Evidence Grade

The duty of care to real people, accountability for decisions about roles, and trust. AI can change what work gets done, but it cannot decide who is affected, cannot take responsibility for the human consequences, and cannot extend or repair the trust a team places in its leader. Those stay with the person in charge. From Talk to Your Team About AI Before the Rumours Do

A one-page delegation brief with a fixed shape, Context, Goal, Constraints, Decision rights, Check-in points and Done-looks-like. Decision rights are split explicitly into what the team decides alone, what escalates, and what is still unassigned, so delegation stops relying on everyone remembering the conversation the same way. From Who Owns Your Team's Skill Library?

It means losing the skills you no longer practise because AI now does the task. People build expertise by doing the reps, the drafting, the analysis, the wrestling with a hard problem and the checking of the answer. When AI does those steps, the output still appears, but the learning that used to come with it does not. A Microsoft and Carnegie Mellon study found higher confidence in AI is associated with less critical thinking, and an MIT study found sustained AI use left people under-engaged and less able to recall their own output. The skill atrophies quietly, and you notice only when you need it. From Protect the Reps: Lead So AI Does Not Deskill Your Team

It means the ideas get better one by one and more alike as a set. In a Science Advances study published in July 2024, Anil Doshi and Oliver Hauser had 293 writers produce short stories with no AI, with one GPT-4 idea, or with five. Novelty rose 5.4% with one idea and 8.1% with five. At the same time the stories converged, by 8.9% of the total range in the five-idea condition. The authors concluded that individual creativity rises while collective novelty is at risk. Nothing goes wrong for any individual. The loss only exists in the set. From AI Is Narrowing Your Team's Idea Pool. Only You Can See It.

It means adding friction back into a workflow on purpose: a human approval at a named step, a mandatory pause before an irreversible action, a named sign-off on the result, or a staging step where the agent writes to a draft rather than to production. The point is not to be slow, it is to put a human touchpoint back where speed removed one that mattered. From The Leader's Case for Slowing an Agent Down

Relevance, Evidence, Audience and Consequence. Relevance asks whether the option solves the assigned problem. Evidence asks what carries each material statement. Audience asks whether the intended recipient can use the option correctly. Consequence asks what happens if the option is used or believed. Each criterion carries hard reject triggers that no score elsewhere can offset. From AI Made Drafting Cheap. Build a Rejection Rubric.

That it is often a reaction, not a trait. Classic forecasting experiments found people became more reluctant to use an algorithm after seeing it err, even when it outperformed a human, and later studies found people were far more willing to use an imperfect algorithm when allowed to adjust its output even slightly. A refuser may be responding to a visible error, unclear permission or poor task fit. Test the work before choosing the explanation. From The AI Enthusiast and the AI Refuser Need the Same Test

In the MIT Sloan Management Review article of 16 July 2026, the authors describe sovereign AI as governing where data is stored and processed, whose infrastructure is used for training and operating AI models, and how algorithmic decisions are reviewed. That is three separate questions, and an organisation can hold a different position on each. Treating them as one question is what produces the binary in or out framing that leaders find unhelpful. From Sovereign AI Is a Strategy Choice, Not a Compliance Box

That historically about 15 per cent of Australian workers changed occupation from one year to the next, based on linked administrative data. JSA's detailed paper notes those historical pathways are not yet clearly showing consequential generative-AI changes, and some underlying sources predate mainstream large language model use. It is context about normal movement, not a target, a probability for any individual, or proof that a destination exists. From Stop Predicting Their Jobs. Map Their Next Skills.

According to the Australian Bureau of Statistics, 12 per cent of businesses reported using AI in the 2024-25 financial year, including 24 per cent in Financial and Insurance Services. The ABS also says the question is not designed to measure intensity or extent of use within the business, which is exactly the blind spot licence counts share. From Your AI Advantage Cannot Belong to One Team

Structured reflection, within limits. A meta-analysis of 61 studies covering 915 teams and 3,499 individuals found after-action reviews improved training-evaluation criteria, and a 2024 meta-analysis found a positive reflexivity-performance relationship moderated by team size and team tenure. Neither studied generative AI governance in Australian financial services, so the retro must prove itself through retests, not citations. From Run the AI Retro Before the Error Becomes the Process

In a randomised experiment at a large UK firm, 78 employees from three occupational groups completed article conceptualisation tasks and 76 completed the article execution task, with access to a bespoke generative AI tool randomised. AI was more effective at closing gaps for adjacent occupations than distant ones, and more effective for conceptualisation than detailed execution. It is a draft working paper in a specific writing setting. From Do Not Grade Your Experts on Prompt Speed

The decision behind the conversation, the duty of care in the room, the formal written record and the relationship the morning after. The model can play the counterpart and critique your framing. It does not get a vote on the message itself. From Rehearse the Hard Conversation Before You Have It

Strong work in all three suggests calibrated method choice. Strong AI-assisted work with weak evidence checks points to verification coaching. Strong unaided work with a failed AI-required condition points to access, instruction, tool fluency or a genuine refusal to follow a clear requirement. Similar failures across several employees point at the task, training or workflow, not several simultaneous attitude defects. From The AI Enthusiast and the AI Refuser Need the Same Test

A 2025 study in The Quarterly Journal of Economics examined a generative AI assistant used by 5,172 customer-support agents. Access increased issues resolved per hour by 15 per cent on average and by approximately 30 per cent for less-skilled and less-experienced agents, with treated two-month-tenure agents matching untreated agents of more than six months on the productivity measure. The most skilled agents saw little benefit and slight quality declines. The authors caution it was one firm and one type of work. From AI Made the New Starter Faster. You Still Owe Them an Apprenticeship.

A 2026 peer-reviewed field experiment with 791 Procter & Gamble professionals found individuals using AI matched the performance of two-person teams without AI on product-innovation challenges, with human judgement retaining value in evaluating ideas. The Jagged Frontier experiment with 758 consultants found large gains inside the model's capability frontier and a 19-percentage-point accuracy drop on the task outside it. Neither study sets the exception rate for your workflow; measure it locally. From An Agent Queue Is Not a Workforce Plan

Bounded associations, not proof. Edmondson's 1999 field study of 51 work teams in a single manufacturing company found team psychological safety was associated with learning behaviour, and a 2007 study across 23 neonatal intensive-care units linked psychological safety to learn-how activity and perceived implementation success. Neither examined AI, Australian workplaces or financial services, which is why the route must be tested through drills rather than assumed. From Silence Is Not Human Oversight

Five fields per entry. Before records the claim, assumption, omission or proposed action as it arrived for review. Intervention records what the human added, removed, reframed, verified or stopped. Basis records the source, policy, evidence, expertise or escalation that warranted the change. Effect records what changed for the reader, customer, decision owner or control. Credit records who exercised the judgement and who independently confirmed it. From The Polished Output Hid the Expert Who Made It Safe

Name an admin owner, decide who may build, run, publish and connect agents, scope which connectors can touch sensitive systems, and confirm you can monitor and suspend agents through the compliance controls. If a connector would reach real customer or employee data before those rules exist, you are not ready to greenlight, deadline or not. From A Deadline Is Not a Decision: Greenlighting AI Before the Free Window Closes

Then do not promise it. Credibility is the whole asset here, and a promise you cannot keep destroys it the first time reality contradicts you. Be honest about the uncertainty, be specific about what you can commit to (genuine consultation, notice, reskilling support, how decisions will be made), and be clear about what is not on the table. Honesty about uncertainty beats false comfort. From Talk to Your Team About AI Before the Rumours Do

Measure total effort and rework across the whole workflow, not generation speed. AI often moves effort from creation to checking. If the checking burden exceeds the benefit, narrow the use case, improve the source material, change the review tier or stop the trial. A trial that gets stopped on evidence is a success of the method, not a failure. From Use AI to Kill Work Before You Accelerate It

A booked, protected slot for the allocated destination, a named service trigger that is the only condition allowed to displace it, and a record of every displacement. If displacement repeats, the record shows either that the claimed dividend was not available at team level or that leaders chose another destination without saying so. From AI Saved the Time. Your Calendar Will Take It Back

A team-level comparison of five conditions: safe access, protected practice time, suitable tasks, review burden and recognition. Each is rated clear, constrained or absent, with evidence, the reason for any constraint, a named owner and a repair date. The ratings are not totalled into a maturity score because the pattern matters more than the arithmetic. From Your AI Advantage Cannot Belong to One Team

A small record created when someone stops or questions AI-assisted work. It has four fields: the stop state, so the challenger knows what happens now; the challenge statement describing the observable issue; the response contract naming the human owner, acknowledgement time and decision time; and the retest evidence showing what changed, which test was rerun and who accepted it. From Silence Is Not Human Oversight

A six-field record that keeps the meeting aimed at changed work: the trigger, the approved de-identified evidence, the specific control change, a named human owner with authority and capacity, a retest defined before implementation, and closure linked to test evidence and an authorised human acceptance. From Run the AI Retro Before the Error Becomes the Process

A rule that every external claim is labelled by what it actually measures: technical exposure, observed adoption, historical mobility, a modelled scenario, or an actual change in your team's work. JSA's own analysis says exposure scores do not reflect the practical context of the task, assume current technologies are implemented to their peak potential, and cannot with any certainty be taken as definitive measures of expected employment or wage effects. From Stop Predicting Their Jobs. Map Their Next Skills.

A verified human act that materially changes what a piece of work claims, how certain it appears, who may act on it or whether it should proceed. Examples include rejecting an unsupported comparison, restoring a limitation the draft dropped, bringing in an operational exception the draft missed, and keeping a recommendation or approval with the accountable person. It is the unit of credit that survives when AI produces most of the visible text. From The Polished Output Hid the Expert Who Made It Safe

It is a concept introduced by Madeleine Clare Elish in Engaging Science, Technology, and Society in 2019, describing how responsibility for an action may be misattributed to a human actor who had limited control over the behaviour of an automated system. Her comparison is to the crumple zone in a car, which absorbs the force of an impact. The difference is that a car's crumple zone protects the driver, while the moral crumple zone protects the integrity of the technological system at the expense of the nearest human operator. Elish was writing about aviation and nuclear accidents in an American context, but the shape travels to any workplace where a person is the last step before an automated output goes out. From When the AI Gets It Wrong, Who Carries It?

A compact record with six fields: the decision written as an action or approved position; the evidence relied on with version or date; the conditions, limits and stop triggers attached; the authority that approved it and, separately, the delivery owner; material dissent and unresolved questions; and execution details including due date, review date and the authoritative system where the record lives. From The AI Joined the Meeting. The Decision Record Still Belongs to You.

A skill file is a short markdown instruction file that captures how your team does one repeatable task, in six sections covering purpose, when to use, inputs required, method, output format and guardrails. Written once, it runs in ChatGPT, Claude or Microsoft Copilot and holds every draft to the same standard. From Your First Skill File: A Decision Memo On Demand

A one-page document, written per AI system before any incident, with five parts: a named person and a named deputy; three to six specific trigger conditions decided while everyone is calm; the standing that makes the authority real, including the sentence that no delivery commitment overrides it; a rehearsed rollback with a staffed fallback; and a decision record capturing what was seen, decided, by whom and what happened next. From Who in Your Organisation Can Actually Stop the Model?

It is one of four AI-driven leadership skills identified in the Bevilacqua and colleagues study: using AI as a genuine partner in thinking rather than an answer machine. A strategic co-thinker frames the real problem so the model solves that and not the obvious one, then treats the output as a draft to interrogate, asking what is flawed or missing, before applying their own judgement to decide. From You Are No Longer the Smartest Person in the Room

A shared, indexed set of skill files, the reusable markdown instruction files a team relies on for recurring work such as briefs, memos and updates. It becomes a library, rather than a pile, when every file has a name that follows one convention, a single owner, a last-reviewed date and an entry on a one-page index. From Who Owns Your Team's Skill Library?

Testing the changed control against the reproduction case that exposed the issue, an adjacent case with a meaningful variation and a boundary case where the control should stop the work or route it to a person. Acceptance criteria are set before the control is changed so an ambiguous result cannot quietly become a pass. From Run the AI Retro Before the Error Becomes the Process

It is a one-page, living record of your team's recurring tasks that names, for each one, the current owner, the role AI plays (none, draft, assist or do-and-check), the part that stays human, and when the line gets reviewed. The team writes it together, so AI becomes something they decide on rather than something done to them. From The AI Delegation Charter Your Team Actually Needs

A pre-mortem imagines the plan has already failed, then lists the reasons it did. The method comes from Gary Klein's 2007 Harvard Business Review work. AI makes it available to a single leader in fifteen minutes on any decision. You instruct the model to act as a sceptic, run the failure scenario, develop the strongest objection, and audit the assumptions, then you decide. From Make AI Disagree With You Before You Decide

Make it explicit that the shared AI produces drafts, not decisions, and that questioning its output is the expected behaviour rather than a challenge to whoever tagged it in. Say it out loud, model it once yourself by visibly correcting the AI on a low-stakes task, and name who owns the output when it goes to anyone outside the channel. One clear norm, modelled early, does more than a written policy nobody reads. From A Shared Team AI Identity Changes How You Lead

A two-way exchange between a tool-fluent colleague and a context expert on one bounded workflow. The expert supplies boundary cases, evidence traps, escalation triggers and acceptance criteria. The tool-fluent colleague contributes an approved technique and makes the tests repeatable. It is not remedial training for the expert, reverse mentoring dressed up with a fashionable label, or a ranking of generations. From Do Not Grade Your Experts on Prompt Speed

A June 2026 preprint on compressing financial filings and earnings-call transcripts frames it as information fidelity, where compression loses fidelity when it changes the decision induced by the source. The authors name two diagnostic patterns: decontextualisation, where salient evidence is retained but separated from the caveats and contextual qualifiers needed for correct interpretation, and model dependency, where different compressors expose different views of the same source. From The Summary Kept the Number and Lost the Caveat

It is when the time AI saves at the drafting stage leaks straight back out at the review stage. AI produces a fast first draft, but someone must check it, the bar for what passes has risen, and nobody has said what to do with the recovered time, so it dissipates into more checking. From The Review Tax: AI Adoption Is Done, Now Design the Checking

It has three rungs. At inform, AI gathers and summarises while a human decides. At recommend, AI proposes a specific option and a human approves or rejects it. At act within bounds, AI executes inside a defined envelope, like sending a routine response, with a human notified who can reverse it. From Decision Rights Are the Leadership Job AI Just Made Urgent

There is no universal measure. Choose the closest credible outcome for the specific workflow, such as resolution rate, decision cycle time or usable first-pass quality, then pair it with quality, rework and risk guardrails so a gain in one place cannot hide a loss in another. From Stop Counting Prompts. Measure the Work That Improved

Adoption is people using the tools, a licence-utilisation number you can put on a dashboard. Absorption is the organisation redesigning how it works to capture value, an actual result. You can buy adoption, but absorption only comes when people change how they work, which happens when their manager makes it safe, expected and normal. From Set the AI Norm: Your Team Copies How You Use It

A one-time sort of every recurring message a leader sends. Inform covers facts and logistics, where AI can draft freely. Coordinate covers decisions and plans, where AI can structure and pressure-test while the leader authors the reasoning. Connect covers recognition, feedback and apologies, which the leader writes personally, with AI limited to flagging typos and unclear sentences. Each lane gets a one-line AI rule, written down once and applied on autopilot. From Don't Let AI Write the Messages That Build Trust

Fix three things in writing before the pilot starts: how long you will run it before judging, which dimensions you will judge it on, and what result would cause you to stop. Making those calls afterwards means the available data decides, and the most available number is almost always activity. From Set the Evaluation Window Before the Pilot Starts

Require a what-was-left-out line on any paper that supports a consequential decision: the caveats, the dissent and the range that did not make the summary. It costs the author two minutes, it is impossible to write without re-reading the source, and it converts an invisible omission into a visible choice someone has signed. From The Summary Kept the Number and Lost the Caveat

Manager behaviour and the surrounding environment matter far more. Microsoft's 2026 Work Trend Index found organisational factors like culture, manager support and talent practices account for more than twice the reported AI impact of individual factors like mindset, at 67 per cent versus 32 per cent. From Set the AI Norm: Your Team Copies How You Use It

Eight fields: the claim itself, the grade, the denominator and period including exclusions, the source and method, the unresolved control gap, the accountable human owner, the decision requested, and the next evidence date. The denominator defends against decorative percentages and the control gap stops a benefit claim from outrunning safe operating conditions. From Your AI Update Needs an Evidence Grade

Take the public accountability yourself, by name, before anyone asks who did it. Then separate the process question from the person question and run them on different clocks. Ask what the standard said the check was, whether the check was actually done, and whether the time and access to do it properly existed. Do not open with who approved this, because that question ends the inquiry at the nearest human and leaves the workflow that produced the failure completely untouched. From When the AI Gets It Wrong, Who Carries It?

Change the pattern rather than confessing. The next time someone genuinely earns recognition, write it yourself, name the specific thing they did, and send it the same day. Specificity is the tell that a human was paying attention, and it is the one thing a model cannot fake, because it was not in the room. One concrete hand-written acknowledgement rebuilds more signal than a month of polished generic praise. From Don't Let AI Write the Messages That Build Trust

Name where the saved time goes and protect it. If AI takes a two-hour task to thirty minutes, say out loud what the recovered ninety minutes is for, deeper client work, a stalled priority, or thinking time. Saved time that is not claimed gets reabsorbed into busywork and double-checking. From The Review Tax: AI Adoption Is Done, Now Design the Checking

The largest opportunity gap that can be addressed without weakening the control environment: a de-identified practice pack, a small amount of protected time, a reviewer clinic or recognition for people who make corrections reusable. Run the repair as a 30-day experiment with named approvals, stop conditions and a human decision at day 30. From Your AI Advantage Cannot Belong to One Team

Anything where a wrong result could reach a customer, harm a person or corrupt the record before a human sees it, and any task that is really how someone builds a skill. The judgement, the relationship, the sign-off and the accountability stay human. The charter names those explicitly so the boundary is a decision, not an accident. From The AI Delegation Charter Your Team Actually Needs

Treat AI adoption as an operating-model change, not just a software rollout. Name the new work the rollout creates for managers, subtract something from their load to make room for it, resource the coaching and validation with time and training, and close the gap between how executives and managers experience AI by getting into the real workflow. The sustainability of the manager layer is a leadership responsibility, not something to delegate to the tool. From Your AI Rollout Landed on Your Managers. Resource It.

Three lines beat an activity dashboard. How many consequential AI systems have a named stop authority with a deputy. When was each last tested with a drill, and what did the drill find. And has anyone exercised the authority this year, and what happened to them afterwards. The third line reveals whether stopping is survivable in your culture. From Who in Your Organisation Can Actually Stop the Model?

That both sides can produce what they promised under one realistic boundary failure, such as a prohibited input, a rejected output, a changed feature, access failing mid-case or a suspected disclosure in a provider log. The exercise records what each side actually produces and repairs the first control that fails. It does not declare a legal data breach; authorised privacy, security and legal functions assess that. From Your AI Rules Stop at Payroll. Your Risk Does Not.

Keep real personal, claimant or commercially sensitive information out of the prompt unless the tool is sanctioned for it. The model is a thinking aid, not a system of record, and it should not become a back door for sensitive data. From Make AI Disagree With You Before You Decide

A sharp increase in performance and perceived efficiency concurrent with flat developer activity. The authors read that as generative AI improving the quality of work rather than the raw volume of it, and teams also reported improved well-being alongside the performance gains. From Set the Evaluation Window Before the Pilot Starts

When the action is hard to reverse, when small steps compound into something no one sees until it is large, or when the step carries a judgement, a relationship or an accountability a person must answer for. Those are the workflows that warrant deliberate friction. Reversible, low-consequence, high-frequency work is where speed is right and friction is just cost. From The Leader's Case for Slowing an Agent Down

Before the rumours, not after. The moment AI is visibly changing work in your team, or a rollout is coming, is the moment to talk, while you can still shape the narrative. Waiting until people are already anxious or a change is already decided means you are managing damage instead of building trust. From Talk to Your Team About AI Before the Rumours Do

Reserve it for decisions that are hard to reverse, expensive to get wrong, or the ones you feel most certain about, because certainty is usually the signal that you have stopped looking for problems. You do not need to run it on every call. A thinking partner earns its place on exactly those high-stakes decisions. From Make AI Disagree With You Before You Decide

Not after a fixed number of completed cases. Move a person from observe to construct to defend when their evidence is consistent across an ordinary case, an ambiguous case and an exception, and when they can repeat the performance while the assistant is unavailable. Retain sampling by a qualified human even after an authority gate is passed. From AI Made the New Starter Faster. You Still Owe Them an Apprenticeship.

On evidence, not the calendar. Use three entry conditions: an incident where AI-assisted work contributed to an actual adverse outcome, a near miss where human review caught a credible consequence before use, and a repeated correction where the same defect class appears in two or more reviewed outputs. A suitably authorised person decides the response and whether formal processes must also run. From Run the AI Retro Before the Error Becomes the Process

When any of five tests fails: the option is specific enough to investigate, the demand evidence is current and scoped, the employee wants to explore it, a real capability gap is named, and an authorised owner can provide a next test. Close it openly. A shorter honest register is more useful than a page of imaginary careers. From Stop Predicting Their Jobs. Map Their Next Skills.

Run it only for conversations that carry weight, like a restructure, serious feedback or a negotiation you cannot afford to fumble. Run it for everything and it becomes avoidance with extra steps. The signal you are using it well is walking in slightly bored by your own opening, because you have already heard it land badly once and fixed it. From Rehearse the Hard Conversation Before You Have It

Automate where inputs, rules, quality thresholds and exceptions are well understood and the consequences of an error are contained, with monitoring and an exception path in place. Augment where context, judgement or accountability remains material, keeping a human as the author of the decision and AI in a preparation or checking role. From Use AI to Kill Work Before You Accelerate It

Start with high-volume recurring work that has a visible recipient and low safety or legal consequence. Monthly reporting, project approvals and coordination meetings are good candidates. Avoid beginning with a politically sensitive workflow or a poorly understood critical control, because the first review should build the habit, not test the organisation's appetite for conflict. From Use AI to Kill Work Before You Accelerate It

In bounded settings, less experienced workers. Noy and Zhang's online writing experiment found time fell 40 per cent, quality rose 18 per cent and inequality between workers decreased. Brynjolfsson, Li and Raymond found a 15 per cent average productivity gain across 5,172 support agents, largest for less experienced agents. Neither study is a rollout guarantee for Australian financial services. From Your AI Advantage Cannot Belong to One Team

The same people as before AI arrived. Australia's National AI Centre advises organisations to assign, document and communicate accountability and maintain human oversight, and APRA's 30 April 2026 letter expects ownership and accountability across the AI lifecycle with human involvement for high-risk decisions. In a meeting, that means a named recorder, a named confirmer, and no AI-produced artefact creating doubt about who exercised the authority. From The AI Joined the Meeting. The Decision Record Still Belongs to You.

Whoever the team has decided owns it in advance, and if no one has decided, it lands on whoever tagged the AI in, which is rarely fair. A shared identity makes the nearest-human question harder, because several people touched the thread. The workable rule is that the AI never holds accountability and the named human who releases a piece of work owns it, exactly as they would for any draft. Decide that before the first visible error, not during it. From A Shared Team AI Identity Changes How You Lead

One named person per skill, not necessarily its author. The owner fields improvement suggestions, applies changes, announces new versions and keeps the index row current. Shared ownership is no ownership, because when everyone can edit and nobody must answer for the file, versions fork and drift goes unnoticed. From Who Owns Your Team's Skill Library?

The leader's, and it cannot be delegated to the technology team. McKinsey's research points to leaders, not employees, as the main brake on getting value from AI, and the deskilling risk is the same shape: it is a workforce-capability decision, not a tooling one. Deciding which work stays human for development reasons, and holding that line against short-term output pressure, is a leadership judgement about the future capability of the team. From Protect the Reps: Lead So AI Does Not Deskill Your Team

Because most of the anxiety about AI is about things being decided about people without them. A charter the team co-writes turns AI from a threat into a shared choice, and it draws on the people who actually know which parts of a task carry judgement and which are safe to hand over. A charter imposed from above rebuilds the very fear the conversation was meant to settle. From The AI Delegation Charter Your Team Actually Needs

Ten people each reporting 30 minutes saved does not establish five available team hours. The reports may cover different tasks, periods and baselines, review or rework may sit elsewhere, and the minutes may already be spent on other work. Never convert individual estimates into headcount, staffing reductions or service commitments without separate workflow and workforce evidence. From AI Saved the Time. Your Calendar Will Take It Back

AI cannot decide its own review tier, because the stakes are a judgement about the business, not something the model reads off the text. It cannot certify that its output clears a human standard, and it cannot carry accountability for what ships. Responsibility sits with a person. From The Review Tax: AI Adoption Is Done, Now Design the Checking

The training method behind most assistants, reinforcement learning from human feedback, can reward responses that match user beliefs over truthful ones. OpenAI rolled back a 2025 update for being overly flattering, and Anthropic's research found five leading assistants all exhibit sycophancy. People often rate the agreeable answer above the correct one, which is how the behaviour got trained in. From Make AI Disagree With You Before You Decide

Because they experience AI differently. Research shows executives tend to see AI as a strategic advantage viewed from above, while managers meet its flaws inside real workflows, under real constraints, without enough time or support. The C-suite sees the promise and the potential headcount saving; the manager sees the messy reality of making it work. That perception gap is why the burden goes unseen and unfunded. From Your AI Rollout Landed on Your Managers. Resource It.

Because many checkpoints existed only because the work was slow. When a person had to carry a task to its next step, that hand-off was a natural moment for a second look. An agent that runs the whole sequence in one unbroken pass closes that gap, and nobody decided to remove the review. The speed removed it, silently, which is why it is easy to miss. From The Leader's Case for Slowing an Agent Down

Re-prompting rebuilds the standard from memory every time, so quality depends on who is typing and how patiently they correct. A skill file states the standard once, in full, and every use starts from it. It also survives staff changes, because the standard lives in a file rather than in one person's chat history. From Your First Skill File: A Decision Memo On Demand

Individual tools change actions one person can alter more easily than work requiring coordination. An NBER field experiment across 66 firms and 7,137 knowledge workers found the 80 per cent of treated workers who used the tool spent two fewer hours on email each week, without a detected change in the quantity or composition of tasks. A second NBER paper using Danish administrative records found no detectable effect on earnings or recorded hours. Faster components get absorbed by coordination, oversight work, additional output or existing demand. From AI Saved the Time. Your Calendar Will Take It Back

Because proximity is the easiest answer available, and because that person is the only party the organisation can actually sanction. The vendor is not in the room, the model cannot be performance managed, and the decision to deploy the tool was made somewhere above the failure. Elish's account of Three Mile Island is precise on the mechanism: the operators knew the system was malfunctioning, but they did not have sufficient information or authority to take corrective actions. Someone can be answerable for an output without ever having had the control, the information or the time to prevent it going wrong. From When the AI Gets It Wrong, Who Carries It?

Ninety per cent passed review is uninterpretable without the number of outputs, the selection rule, the test period, the definition of pass and the excluded cases. If nine of ten hand-picked low-risk examples passed, that may be useful early evidence, but it is not evidence about every live case, and the executive needs to see the difference. From Your AI Update Needs an Evidence Grade

Because it proves presence, not capability. A worker may know an output is wrong yet stay silent when the deadline is visible, the challenger bears the delay and the tool's sponsor receives the challenge. Oversight requires both capability and permission: staff who can stop the work without seeking the sponsor's approval, and a leader who owns the response. From Silence Is Not Human Oversight

Because faster generation can move the constraint rather than remove it. Drafting speeds up while verification, exceptions and specialist judgement absorb the time, and the hours that should protect coaching and capability-building quietly disappear. Output counts describe the machine lane. Completed, accepted work within quality, risk and queue limits is the number a workforce plan needs. From An Agent Queue Is Not a Workforce Plan

Approval is diffuse and low cost, because approving agrees with the direction of travel. Stopping runs against it and costs someone their date, their number and occasionally their bonus. An authority that exists only on a slide will not be exercised by someone who has to spend their own credibility to use it, which is why the standing and the reporting line matter more than the framework. From Who in Your Organisation Can Actually Stop the Model?

Because no individual has any reason to fix it. Doshi and Hauser name it as a social dilemma: if writers learn their AI-assisted work is rated as more creative, they have an incentive to use AI more, and collective novelty may fall further. Each person is correctly optimising the thing they can see, which is their own output, and each is succeeding. The narrowing is only visible to whoever holds the full set of options. That vantage point is the leader's, which makes the remedy structural rather than personal. From AI Is Narrowing Your Team's Idea Pool. Only You Can See It.

Because preference is not performance. In a peer-reviewed field experiment with 758 consultants, AI improved speed and quality across 18 tasks inside the model's capability frontier, yet on a task selected to be outside that frontier AI users were 19 percentage points less likely to reach the correct solution. The same person can be helped by AI on one task and harmed on another, so usage counts tell you almost nothing about judgement. From The AI Enthusiast and the AI Refuser Need the Same Test

Drafting the message is the low-value use of AI here. A hard conversation rarely fails on the words you planned. It fails on your reaction to the response you did not. Rehearsing lets you feel the resistance before it is real, so you spend your composure in practice and arrive with it intact. From Rehearse the Hard Conversation Before You Have It

Because averaging lets a strong presentation carry a fatal flaw. A hard reject is a failure that cannot be offset, so it has to be applied before any total is calculated. Ranking first invites the most fluent option to accumulate support while its evidence gap stays unexamined. From AI Made Drafting Cheap. Build a Rejection Rubric.

Because every step towards sovereignty costs something a risk function cannot trade away on its own: capability, speed, cost, or the size of the provider pool. Only the person accountable for the outcome can decide that a slower or more expensive option is worth it for one use case and not another. Delegated downwards, the decision defaults to whichever answer is easiest to defend. From Sovereign AI Is a Strategy Choice, Not a Compliance Box

Because the anxiety is already there. Around 30 per cent of Australian workers are worried AI will replace their job, per Finder's April 2026 survey. If leaders say nothing, the silence does not read as reassurance; it reads as bad news being withheld, and rumour fills the vacuum. A leader who raises it first shapes an honest conversation instead of managing a panic. From Talk to Your Team About AI Before the Rumours Do