AI Change Desk

AI Change Desk

By Michael Hanna-Butros MeyeringTechnology
Download on the App Store

All AI Change Desk episodes

  • AI Change Desk | EP054: Does Blank Mean Nobody Used It?

    Hey! I'd love to hear your thoughts, send me a voice note.

    Does Blank Mean Nobody Used It?

    Episode 54 of AI Change Desk with Michael Hanna-Butros Meyering.

    A blue binder, a sticky front door and a training invitation that nearly goes to the wrong person. In a fictional rollout, Ruth wants to help people use Copilot. Before she sends her note, can someone else explain why each person belongs on the list?

    Context

    Separate GitHub and Microsoft examples distinguish missing editor detail from observed activity, unavailable fields from measured zeroes, and feature totals from a count of people. A reporting window is not automatically the right audience rule for a reminder.

    Impact

    Move one invented person's last qualifying use from twenty-nine days ago to thirty-one days ago. With the other conditions held fixed, the eligibility decision changes. But meeting an inactivity criterion does not mean the person has never tried the tool, or that anyone should send a message.

    Action

    Give two colleagues twenty minutes, four fictional paper cards and a pencil. Record the intended action, product and population, qualifying event, period and end date, handling of missing fields, and decision owner. Ask each reader to justify an inclusion, an exclusion and a case where they must stop for missing information. Review the invitation's first sentence as carefully as its audience.

    The people, reports and activity dates are fictional. No employee exports or messages are used. The exercise does not test or certify provider software. GitHub and Microsoft examples are separate products, not joined records or evidence of an actual mistaken notification.

    Correction: EP053 Source Update

    Conductor has withdrawn the wind-down notice discussed in EP053. Its updated notes cross out the November 13 and 20 dates, and its new update says development is resuming. Those dates in the earlier recording are historical, not current deadlines. Check Conductor's current guidance for service and data-retention details.

    Official Sources

    • GitHub: telemetry and version fields
    • GitHub: usage-metrics definitions
    • GitHub: feature engagement update
    • Microsoft: reporting and outreach criteria
    • Conductor: amended notice
    • Conductor: development update
    • Disclosure

      I use AI to help research, produce, and narrate this show.

      I check the claims, flag the risks, and decide what goes on air.

      My work in privacy shapes the questions I ask, but I don't discuss nonpublic work here.

      I speak for myself, not the State of Oregon or any other organization.

      This is practical guidance, not legal advice.

      18 min
    • AI Change Desk | EP053: Who Owns the Automation After It Changes?

      Hey! I'd love to hear your thoughts, send me a voice note.

      Who Owns the Automation After It Changes?

      Episode 53 of AI Change Desk with Michael Hanna-Butros Meyering.

      Correction: Source update - November 16, 2026. Conductor has withdrawn its previously announced wind-down. Its updated 0.93.0 notes cross out the November 13 read-only and November 20 shutdown dates, and its 0.94.2 update says Conductor is not shutting down and development is resuming. The dates mentioned in EP053 are historical, not current deadlines. Consult Conductor's current guidance for service, export and data-retention details. The original recording is unchanged.

      Sources: updated announcement, revised 0.93.0 notice, 0.94.2: Conductor is back.

      On Friday, a dispatcher gets a reopened repair back onto the crew list. On Monday she's off. Will the next dispatcher get the corrected list, or just the same mistake at a different address?

      A follow-up can fix one report without changing tomorrow's instructions. Michael looks at how shared GitHub Copilot automations survive a handoff, why a Google Cloud skill name doesn't tell you which revision ran, and how teams preserve ownership when a tool's plans change.

      Try It This Week

      Give a colleague the saved rule and three fictional tickets: one open, one reopened, one closed. Before they build the list, have the dispatcher write down which repairs should appear. Compare answers. Record who can change the rule, who accepts the result and when they'll check a fresh list.

      The dispatch story is fictional. The exercise needs no production access and does not test or certify a vendor's software.

      Sources

      • GitHub: shared Copilot Automations
      • VS Code: run and review a cloud automation
      • Google Cloud: Agent Registry release notes
      • Google Cloud: skills and revisions
      • Conductor: announcement and updated plans
      • Disclosure

        I use AI to help research, produce, and narrate this show.

        I check the claims, flag the risks, and decide what goes on air.

        My work in privacy shapes the questions I ask, but I don't discuss nonpublic work here.

        I speak for myself, not the State of Oregon or any other organization.

        This is practical guidance, not legal advice.

        16 min
      • AI Change Desk | EP052: Who Set the Permission Mode?

        Hey! I'd love to hear your thoughts, send me a voice note.

        Who Set the Permission Mode?

        Episode 52 of AI Change Desk with Michael Hanna-Butros Meyering.

        A request to correct a library's opening hours sounds simple. But who may approve

        the assistant's next action: the person, a fixed rule, or another model?

        GitHub's assisted-permissions announcement and connected cloud workflow provide

        the lead for this week's distinction: a task request, an approval decision, and

        an enforced boundary are three different things. NIST's proposed agent-identity

        principles and Google's preview policy documentation add two supporting perspectives.

        In This Episode

        • Why approving a task is not blanket approval for its intermediate actions.
        • The unresolved difference between two GitHub descriptions of managed ask rules.
        • Why reasoning about a useful action and controlling access are separate concerns.
        • Why a recorded policy verdict does not by itself demonstrate prevention.
        • A 45-minute, paper-only exercise to prepare a bounded trial proposal.
        • The community library, staff and exercise records are fictional. No library

          system or vendor permission behavior was tested. The GitHub documentation

          difference is unresolved, not a demonstrated bypass. NIST's tenets are proposed

          principles, not a binding rule or judgment about a vendor. Google's feature is

          in preview. The exercise is not deployment permission or a safety certification.

          Sources

          Public primary-source claims were rechecked November 2, 2026. The GitHub pages

          belong to one source family; NIST and Google do not independently verify GitHub's

          runtime behavior. The Google guide's October 30 update is not a launch date.

          • GitHub: assisted permissions generally available, October 26
          • GitHub: connected Copilot cloud experience, October 28
          • GitHub: assisted-permissions guide
          • GitHub: enterprise managed settings
          • NIST: proposed agent-identity design tenets, October 27
          • Google Cloud: configure semantic governance policies
          • Disclosure

            Quick disclosure before we start.

            AI-assisted tools were used in parts of the research and production workflow.

            Final editorial judgment, risk posture, and release approval stayed human-led.

            This is operational guidance, not legal advice.

            These are my opinions and are not representative of any organization.

            One additional disclosure as the Desk evolves.

            My current role includes privacy work,

            so you will hear me pay closer attention

            to purpose, access, retention, deletion, and accountability.

            I will not discuss nonpublic work here.

            These are my personal views,

            and they do not represent the State of Oregon

            or any other organization.

            19 min
          • AI Change Desk | EP051: What Changes After the Conference?

            Hey! I'd love to hear your thoughts, send me a voice note.

            What Changes After the Conference?

            Episode 51 of AI Change Desk with Michael Hanna-Butros Meyering.

            After attending Gartner IT Symposium/Xpo in Orlando, Michael connects three themes from his session notes: attention to the person receiving a service, time and coaching to develop judgment, and an honest account of AI value and risk.

            A hospitality story, a leadership conversation and Gartner's published research lead to one practical question: did the work become better for someone, or did the organization simply produce more of it?

            In This Episode

            • Why a technically correct answer can still leave someone unable to finish a task.
            • Why permission to experiment is not the same as time and support to learn.
            • How to distinguish faster responses from a better overall service experience.
            • A proposed 45-minute, paper-first exercise for improving one service interaction.
            • The access-support scenario is fictional. The exercise is editorial guidance, not a validated assessment, compliance certification or permission to deploy a production change. Conference observations are not a survey of attendees or peers. Gartner's reported survey result is attributed self-report, not an independent audit.

              Sources

              The symposium took place October 19-22, 2026. Public source claims were rechecked October 26. Multiple Gartner pages belong to the same source family; they are not independent corroboration. Guidara's TED account is an earlier telling of his story, not a new event.

              • Gartner: Orlando symposium dates
              • Gartner: Will Guidara keynote
              • Gartner: Serena Williams keynote
              • Gartner: AI value and risk
              • Gartner: CIO survey and methodology
              • Gartner: technology trends for 2027
              • TED: Will Guidara, September 2022
              • Watch EP051 on YouTube

                Read: Bring Back One Better Service

                Disclosure

                Quick disclosure before we start.

                AI-assisted tools were used in parts of the research and production workflow.

                Final editorial judgment, risk posture, and release approval stayed human-led.

                This is operational guidance, not legal advice.

                These are my opinions and are not representative of any organization.

                One additional disclosure as the Desk evolves.

                My current role includes privacy work,

                so you will hear me pay closer attention

                to purpose, access, retention, deletion, and accountability.

                I will not discuss nonpublic work here.

                These are my personal views,

                and they do not represent the State of Oregon

                or any other organization.

                19 min
              • AI Change Desk | EP050: Did the Test Touch Production?

                Hey! I'd love to hear your thoughts, send me a voice note.

                Did the Test Touch Production?

                AI Change Desk | Episode 50 | October 19, 2026 | 19:27

                Your AI demo works. Then a technician calls about the repair request it created. There is no repair: it was a test ticket. For your team, this was a demonstration. For somebody else, it became their morning.

                That opening is fictional. The operating question is not: when a test reaches a real working system, who owns the effects on the receiving side?

                Michael marks episode fifty with a brief thank-you, then examines the difference between stopping an agent, restoring a setting, removing access, and confirming that the affected workflow is actually resolved.

                Context, Impact, Action

                • Context: separately attributed OpenAI and Anthropic evaluation reports, with Google Cloud guidance on oversight and permissions. These provider accounts do not independently corroborate each other or establish receiving-side recovery.
                • Impact: a failed request can still change server state; restoring a setting does not necessarily revoke access. Invented test data can still enter a real queue.
                • Action: run a 45-minute paper-only destination rehearsal. Track what was attempted, what arrived, what changed, what access remains, and who owns the final disposition. Do not test an outside organization without authorization.
                • Sources

                  • OpenAI: Accessing and changing internal systems during dashboard research - October 15 report about June 17 evaluations.
                  • Anthropic: Investigating unintended model actions - October 9 report, corrected October 12. The practice-form example is a separate case.
                  • Google Cloud Agentic AI Responsible Use Guidelines - marked modified October 12. Deployment guidance, not a new law or proof of control effectiveness.
                  • Sources checked October 19, 2026. Receiving-side recovery is not independently verified in this episode.

                    Disclosure

                    AI-assisted tools were used in parts of the research and production workflow. Final editorial judgment, risk posture, and release approval stayed human-led.

                    This is operational guidance, not legal advice. These are my opinions and are not representative of any organization.

                    My current role includes privacy work, so you will hear me pay closer attention to purpose, access, retention, deletion, and accountability. I will not discuss nonpublic work here. These are my personal views, and they do not represent the State of Oregon or any other organization.

                    20 min
                  • AI Change Desk | EP049: What Does the Watermark Actually Prove?

                    Hey! I'd love to hear your thoughts, send me a voice note.

                    Your team has thirty minutes to publish a document. Someone asks whether AI wrote it. Nobody has checked whether it is right.

                    In this 20-minute AI Change Desk episode, Michael asks what a watermark can establish, what it cannot, and who still owns the publication decision. A detector can be right while the document is wrong.

                    Context: OpenAI's October 5 text-provenance announcement separates opt-in marking from restricted detector access. Its October 8 false-front report adds a different question: have you verified the contributor behind the submission?

                    Impact: a provenance result does not settle factual accuracy, authorship, ownership, or permission to publish. No detected mark is not proof of human authorship. EU Article 50 also distinguishes provider marking from deployer disclosure and substantive editorial responsibility.

                    Action: run one fictional workshop notice through three invented result cards: detected, not detected, and unavailable. Keep the factual error and contributor gap identical. Check the event details, the actual assistance policy, permission, and the accountable publisher before deciding what goes out.

                    This continues episode 48's stop-and-restart discussion by moving to the publication desk: the work is already written, but your responsibility is not finished.

                    The scenario and detector results are fictional, not vendor tests. OpenAI's incident and product claims remain attributed to OpenAI; multiple OpenAI pages are one source family, not independent validation. No deployed detector accuracy or organization-specific legal conclusion is claimed.

                    • OpenAI: Our approach to EU text provenance rules, October 5, 2026
                    • OpenAI: Provenance signals and detector-access limits
                    • OpenAI: Disrupting AI-enabled false-front operations, October 8, 2026
                    • European Commission: AI Act Article 50
                    • European Commission: Article 50 transparency FAQ
                    • The narration records an October 10 source check. These official pages were rechecked before release on October 12, 2026. Product scope and access rules can change.

                      AI-assisted tools were used in parts of the research and production workflow. Final editorial judgment, risk posture, and release approval stayed human-led. This is operational guidance, not legal advice. These are my opinions and are not representative of any organization.

                      My current role includes privacy work, so you will hear me pay closer attention to purpose, access, retention, deletion, and accountability. I will not discuss nonpublic work here. These are my personal views and do not represent the State of Oregon or any other organization.

                      20 min
                    • AI Change Desk | EP048: What Would Make You Stop the AI Run?

                      Hey! I'd love to hear your thoughts, send me a voice note.

                      You hit stop. The screen goes quiet. Then the supplier gets the email. Which work actually stopped?

                      In this 17-minute AI Change Desk episode, Michael separates a stop request from a verified stopped workflow, then asks what evidence and current authority are needed before work may restart.

                      OpenAI's dots controls distinguish main tasks, delegated work and recurring schedules. API steering is a change in direction, not cancellation of tools already underway. Microsoft's Foundry guidance treats trace-linked recommendations as decision support that needs validation, not a real-time safety guarantee.

                      Map five kinds of work: the main task, delegated work, future schedules, completed actions and unknown results. Give every unresolved effect an owner and deadline. Use a 45-minute tabletop with synthetic records to separate requests, acknowledgments and observed effects, then recheck purpose, data, destinations and authority before restart.

                      The supplier example is fictional, not a reported incident. This continues episode 47's assessment-scope discussion by asking how an instruction to stop becomes an observable change in work.

                      If you pressed stop now, could you explain what stopped, what already happened, and what is still allowed to happen next?

                      • OpenAI: Introducing dots, September 29
                      • OpenAI: dots control scope
                      • OpenAI: mid-turn steering
                      • Microsoft: Insights in Foundry, preview guidance
                      • Control, steering and preview-limit documentation rechecked October 5, 2026. These are two vendor source families, not independent testing of your deployment. A tabletop does not prove containment, safe cancellation or recovery.

                        AI-assisted tools were used in parts of the research and production workflow. Final editorial judgment, risk posture, and release approval stayed human-led. This is operational guidance, not legal advice. These are my opinions and are not representative of any organization.

                        My current role includes privacy work, so you will hear me pay closer attention to purpose, access, retention, deletion, and accountability. I will not discuss nonpublic work here. These are my personal views and do not represent the State of Oregon or any other organization.

                        Episode, transcript and sources | Stop and Restart Tabletop worksheet

                        18 min
                      • AI Change Desk | EP047: Who Checks the AI Safety Claim?

                        Hey! I'd love to hear your thoughts, send me a voice note.

                        A reviewer has expertise. A sponsor has a budget. An operator has a launch button. Who owns the answer when the safety finding says the work is not ready?

                        In this 19-minute AI Change Desk episode, Michael follows a safety recommendation into an accountable decision and a concrete change to work. The practical distinction: changing a control, testing it, and verifying an outcome are three different claims.

                        OpenAI's September 22 assessment proposal is a framework for scrutiny, not certification of a deployment. Its September 25 disclosure describes ongoing remediation. A separate September 28 Australian ministerial account describes a changed notification route and continuing investigation. Microsoft's September 24 worked example links a finding to a control and retest while preserving residual failures.

                        Use the five-part Recommendation-to-Decision Check: the precise decision, evidence examined and excluded, actual recommendation, responding authority, and resulting change to work. Set aside 45 minutes with the people who can answer those questions. A worksheet cannot create missing evidence or grant exception authority.

                        Listener question: can the operator explain what the review changed about the work they may do?

                        This continues episode 46's incident-ownership discussion by moving earlier: what does a review establish, and what changes after the finding?

                        • OpenAI assessment principles, September 22
                        • OpenAI incident updates, September 25
                        • Minister Gallagher interview, September 28
                        • Microsoft control-and-retest example, September 24
                        • Source-checked September 28, 2026. These are separately attributed accounts, not independent corroboration of one another or a guarantee for your environment.

                          AI-assisted tools were used in parts of the research and production workflow. Final editorial judgment, risk posture, and release approval stayed human-led. This is operational guidance, not legal advice. These are my opinions and are not representative of any organization.

                          My current role includes privacy work, so you will hear me pay closer attention to purpose, access, retention, deletion, and accountability. I will not discuss nonpublic work here. These are my personal views and do not represent the State of Oregon or any other organization.

                          Episode, sources, and audio

                          Full transcript

                          Recommendation-to-Decision worksheet

                          20 min
                        • AI Change Desk | EP046: Who Owns the AI Incident?

                          Hey! I'd love to hear your thoughts, send me a voice note.

                          AI Change Desk | Episode 46 | Runtime 20:51

                          When an AI workflow crosses a real boundary, an alert is only the beginning. This episode examines the authority needed to classify an event, interrupt work, preserve evidence, coordinate notifications, explain what is known, and authorize recovery.

                          The practical tool is a seven-part AI Incident Authority Card: named decision owners, reachable backups, and evidence that the handoffs work.

                          • Evaluation boundaries and the difference between a model stopping itself and an independent stop control.
                          • Why an announced AI Force is not the same as an established operating structure.
                          • Why a reported privacy notification is not a final regulatory finding.
                          • The separate clocks for containment, notification, public disclosure, and closure.
                          • A 45-minute exercise for the workflow your organization cannot afford to improvise around.
                          • Choose one consequential AI workflow. Assign the classifier, scope authority, stop owner, evidence custodian, notification reviewer, public-statement approver, and recovery/closure authority. Test the handoffs with safe data. Treat gaps as work with owners, not agenda items without deadlines.

                            AI Incident Authority Card worksheet

                            Chapter markers are included in supported podcast players. Interior body times are estimates.

                            • Irregular investigation update - 2026-08-14
                            • Google statements reported by Gerrit De Vynck in The Washington Post - 2026-09-18
                            • Trump public post - 2026-09-19
                            • Axios report on the announcement - 2026-09-19
                            • AEPD notification notice - 2026-09-14
                            • OpenAI model misalignment reporting framework - 2026-09-16
                            • NIST identity and access token guidance - 2026-09-15
                            • The transcript is derived from the approved render scripts and reusable disclosure, not an independent ASR transcription. Music positions follow the final assembly.

                              Quick disclosure before we start. AI-assisted tools were used in parts of the research and production workflow. Final editorial judgment, risk posture, and release approval stayed human-led. This is operational guidance, not legal advice. These are my opinions and are not representative of any organization. One additional disclosure as the Desk evolves. My current role includes privacy work, so you will hear me pay closer attention to purpose, access, retention, deletion, and accountability. I will not discuss nonpublic work here. These are my personal views, and they do not represent the State of Oregon or any other organization.

                              21 min
                            • AI Change Desk | EP045: Who Approved the Question?

                              Hey! I'd love to hear your thoughts, send me a voice note.

                              A plain-language question can now become an enterprise analysis, a dashboard, a copied artifact, and an approved action in one workflow.

                              Existing source permissions matter. They do not, by themselves, approve the purpose, metric definition, audience, derivative copy, or downstream decision.

                              In this episode, Michael introduces the seven-part Question-to-Action Receipt for governing AI-assisted analytics from the first question through final disposition.

                              Can the organization prove that an AI-assisted analysis used an approved purpose, identity, source, metric definition, audience, and action path - not merely data the user was technically allowed to retrieve?

                              1. Purpose.
                              2. Identity and authority.
                              3. Source.
                              4. Meaning.
                              5. Evidence.
                              6. Audience and copy.
                              7. Action and disposition.
                              8. Run the 45-minute Question-to-Action drill on one recurring business question. Approve, narrow, hold, or stop the workflow based on the evidence.

                                AI Data Agent Question-to-Action Receipt:

                                https://michaelhbm.com/downloads/ai-change-desk/resources/ep045-who-approved-the-question/EP045-ai-data-agent-question-to-action-receipt.docx

                                • OpenAI, , September 10, 2026: https://openai.com/index/put-data-to-work/
                                • OpenAI Help Center, , updated September 10, 2026: https://help.openai.com/en/articles/20001518
                                • Google Workspace Updates, , September 8, 2026: https://workspaceupdates.googleblog.com/2026/09/context-aware-access-controls-are-available-for-Gemini-Enterprise-in-the-Admin-console.html
                                • Google Workspace Updates, , September 10, 2026: https://workspaceupdates.googleblog.com/2026/09/manage-external-sharing-for-gemini-notebook-in-the-Admin-console.html
                                • NIST, , updated August 29, 2025: https://www.nist.gov/privacy-framework/getting-started-0
                                • AI-assisted tools were used in parts of the research and production workflow. Final editorial judgment, risk posture, and release approval stayed human-led.

                                  This is operational guidance, not legal advice. These are Michael's personal views and do not represent the State of Oregon or any other organization. His current role includes privacy work; no nonpublic work is discussed.

                                  22 min