Watch the full class.

Give your team a reusable design brief. A fast first draft is useful; a presentation your organization can keep using is worth more.

In this episode, I show a presentation created with Claude Design and explain why shared brand assets matter. The draft still needed cleanup. The useful part is giving people the same starting point, including the company's existing layouts and approved assets.

What matters for your organization

Start with a presentation you trust. Use an approved deck to show what good work looks like. A CEO may need a consistent operating review; a family office may need reporting that is easy to compare across entities. Give AI the template and the source material before asking it to design.

Keep the numbers separate from the design decisions. A polished chart can make an unsupported assumption look settled. In a board or investment committee pack, keep reporting periods and source references visible. Have the person responsible for the figures review them before the deck goes out.

Make the standard reusable. The episode describes sharing a design framework across a team. Assign someone to maintain it. When the approved template changes, update the shared version so the next presentation starts from the right files.

Put it to work this week

Take one recurring presentation. Supply an approved example and ask AI to produce a five-slide draft from a small set of verified material. Review whether it follows the standard and preserves the meaning of the source.

Pick a small task that keeps interrupting your day. Explain the result you need, let an assistant prepare it, and review the work before it reaches someone else.

This episode demonstrates HeyClicky, a voice assistant that works with what is on the screen. I ask it to turn the visible material into presentation documentation. The initial result is a reminder that a short instruction still needs a clear objective.

What matters for your organization

Describe the deliverable. For a CEO, this might be an operating brief from an approved dashboard. A family office could ask for a list of unanswered questions from a report. Specify the audience and what they need to decide; otherwise the assistant may produce a document that misses the point.

Separate preparation from action. Reading a screen and drafting an outline are different from changing records or sending a message. Set that boundary in the instruction. Use a limited task first, especially when the assistant can control applications connected to company information.

Review the result and the changes. Ask which sources it used and whether it edited anything. A task is complete when the output is useful and its effect on the underlying records is understood. The time saved should include the time you spend checking it.

Put it to work this week

Try one read-only task on approved material. Ask for a brief with source references and a list of gaps. Compare it with what you would have prepared yourself before expanding the assistant's responsibilities.

Ask AI to help define the assignment before you ask it to execute. On a complex project, a few useful questions can save a lot of work in the wrong direction.

In this episode, I use planning mode to turn a vague presentation request into a more complete instruction. I also discuss separating planning from execution and choosing different models for different parts of a project.

What matters for your organization

Let the model interview you. You do not need to start with a perfect prompt. Explain the outcome and let the assistant ask about missing inputs. For a portfolio review or an operating project, clarify who will use the result and what evidence they need before the model builds anything.

Match effort to the work. A difficult planning problem may justify more reasoning than a routine formatting task. Ask for a proposed division of work, with estimated cost and uncertainty. The model names and prices mentioned in the recording describe that session; evaluate today's options against your own workload.

Put a limit on the experiment. Define a budget and a checkpoint. Compare total effort, including revisions and review, before increasing the scope. A long run or a large volume of output does not establish that the project is valuable.

Put it to work this week

Use one stalled project. Ask AI to help write the brief, then approve the plan and a bounded first deliverable. Assign a person to accept that deliverable before the rest of the work proceeds.

Before connecting AI to company files, decide what it may read and where it may write. That decision becomes more important when several tools are working on the same project.

This episode shows how connectors and project folders can let work move between applications. I also discuss unexpected usage charges and the conflicts that arise when two agents edit the same files.

What matters for your organization

Approve the connection and its scope. A working connection is only the starting point. For a family office, one folder may contain records for several entities. Confirm the account and the permitted material before giving an assistant a task. A connector does not replace ownership of access decisions.

Give parallel work separate space. If one assistant prepares a plan and another executes it, write down the handoff. Avoid simultaneous edits to the same document. The episode discusses separate folders and Git branches; either approach still needs a person to review the combined result.

Treat usage as an operating expense. Set limits before long tasks begin. The recording includes a personal example of an unexpected bill after selecting an unlimited setting. Use that as a prompt to check your own controls, not as a prediction of what another account will cost.

Put it to work this week

Choose one recurring reporting workflow. Map its approved sources, name the person who owns each output, and test the connection with a read-only check. Keep the first draft separate from the approved records.

Use AI to find out where your explanation breaks down before you are in the room. A rehearsal is most useful when the questions force you back to the evidence.

In this episode, I use ChatGPT voice to practice for a patent examiner discussion. I interrupt the response, ask about a technical concept, and request a simpler explanation. The exercise exposes what I remember and what I need to revisit.

What matters for your organization

Practice against your actual material. For a CEO, the next conversation might be a board review. An investor may need to explain a diligence concern; a family office may need to discuss an allocation proposal. Use approved documents and ask the assistant to challenge the reasoning in those documents.

Ask one question at a time. Answer out loud before asking for help. If the response is unclear, stop and ask for a simpler explanation. Then explain it back in your own words. That is a better test of understanding than accepting an answer because it sounds confident.

Turn weak answers into a review list. Capture unsupported claims and questions that need a specialist. A simulated examiner or committee member cannot establish that the underlying analysis is correct. The value is discovering what to resolve before the real discussion.

Put it to work this week

Run a ten-minute rehearsal using a document you are authorized to use. Finish with a short list of evidence gaps and assign each one to the person who can answer it. Keep this exercise separate from any live external meeting.

Treat a voice agent like an operating process that someone owns. A natural voice is useful, but the work still needs a clear service standard.

The guest demonstration in this episode uses Retell AI and company training material to build a call-handling agent. It shows prompt preparation in Claude, conversation settings, and a test call with an interruption. The discussion then returns to ongoing quality review.

What matters for your organization

Start with the approved playbook. Use the material your team is expected to follow. The demonstration converts training documents into structured text before building the prompt. Check that the conversion preserves exceptions and escalation rules; a cleaner file is only useful if the instructions remain accurate.

Test the difficult parts of a call. Try interruptions, ambiguous requests, and questions the agent should hand to a person. A smooth demonstration is not evidence that every workflow is ready. Start with a narrow service task and make the route to human help clear.

Measure the service outcome. For an operator reviewing a portfolio company, ask about incorrect answers, repeat calls, and failed handoffs alongside cost. Assign someone to review calls and correct the playbook. The episode's reported experience is one operator's account, not a benchmark for your business.

Put it to work this week

Take one approved FAQ or reception workflow and build a test script with common questions and exceptions. Use fictional callers. Have the service owner score the answers and handoffs before considering a customer-facing pilot.

Write the acceptance criteria before you give an agent a long assignment. If you cannot describe a finished result, it is hard to know whether more work is helping.

This episode explains goal-driven work through a dashboard example and connects it to ordinary leadership. A clear objective matters, but so do the evidence of completion and the limits on how the agent may get there.

What matters for your organization

Describe a result you can inspect. For a reporting dashboard, define the approved data and the checks a reviewer must perform. For a diligence brief, require sources and an unresolved-questions section. Asking for the best possible answer leaves too much room for interpretation.

Bound the effort. Set a budget or time checkpoint, and identify the conditions that require the agent to stop. Repeated attempts should produce something you can assess. If a required source is unavailable, the agent should report the gap instead of silently changing the objective.

Keep ownership clear when teams and agents work together. The later discussion explores shared conversations with humans and AI agents. Decide who can assign work and who accepts the result. Access to a team conversation should not be treated as permission to change every system connected to it.

Put it to work this week

Rewrite one open assignment with a named owner, a testable definition of done, and a stopping point. Ask AI to identify ambiguities before starting. Keep the instructions with the output so the next reviewer can see what was requested.

AI training should change how work gets done. Pick a recurring decision or process that consumes your team's time; build something useful around it.

In this episode, I preview a hands-on training program built around practical tracks. Those include financial intelligence dashboards, an AI chief of staff, and family office management systems. The program described moves from prompting and context to agents and workflows, with security included in the training.

The leadership question is where to start.

What matters for your organization

Choose a business problem you already understand. A CEO might start with a weekly operating review. A family office could focus on reporting across entities. For an investor, organizing diligence questions may be more useful than generating another market summary. These are ways to apply the episode's training tracks.

Build the review process into the workflow. A financial dashboard needs a clear source for each number and an owner who can resolve discrepancies. Decide what AI can prepare and what still needs a person's approval. Faster reporting only helps if the underlying information holds up.

Measure useful work. For a first pilot, track time saved after review and the errors that still need correction. In capital-related work, use the output to surface unanswered questions and assumptions that need checking before a decision is made.

Give people time to practice. The episode makes the case for deeper workshops alongside monthly classes. Bring a real process, with information approved for use in your AI environment. Have the person responsible for that process participate.

Put it to work this week

Choose one recurring workflow. Name its owner and define what a successful output looks like. Start with a small sample; review the result before extending its use.