--- type: talk status: draft date: 2026-05-16 updated: 2026-05-16 project: Seafood NZ Conference client: Seafood New Zealand domain: ai-capability due: 2026-05-27 summary: > Editable slide outline for Nadia Ellis' Seafood New Zealand Annual Conference keynote, AI in the Age of Acceleration: Leading Through the Shift. Follows the agreed session flow: pace of change, the gap, integrated demo examples across conversational AI, AI automation, and agentic AI, why organisations stall, leadership's role, and three things to do this month. tags: [session-design] --- # Seafood NZ: Slide Outline v1 **Working title:** AI in the Age of Acceleration: Leading Through the Shift **Session length:** 45 minutes including Q&A **Audience:** Seafood sector leaders, operators, government, service providers **Demo approach:** Three short pre-recorded AI demo clips folded into the three AI layers **Status:** Working draft for editing with Codex --- ## Core Argument AI is moving faster than normal organisational planning cycles. Most organisations are underusing it because they are treating it as a tool problem. The real work is leadership, briefing, process clarity, and culture. For seafood, AI does not create more fish. It makes everything around the fish faster, clearer, and cheaper to manage: compliance, reporting, communication, decision prep, documentation, market intelligence, and workforce knowledge. --- ## Run Sheet | Section | Timing | Purpose | | --- | ---: | --- | | 1. The pace of change | 6 mins | Create urgency and name why this feels different | | 2. The Gap: what is possible right now | 17 mins | Show the three layers of AI available now, each with a short demo | | 3. Why most organisations stall | 7 mins | Reframe weak AI results as a briefing and adoption problem | | 4. Leadership's role | 8 mins | Make this a leadership responsibility, not an IT project | | 5. Three things to do this month | 4 mins | Give practical next actions | | Q&A / buffer | 3 mins | Questions or time protection | --- ## Slide Outline ### Slide 1: Title **On slide:** AI in the Age of Acceleration Leading Through the Shift **Speaker note:** Kia ora. I am Nadia Ellis, founder of Curiosity. I help organisations move AI from interesting experiments into practical capability. **Purpose:** Set a calm, credible, practical tone. No hype. No doom. Useful from the first few minutes. --- ## 1. The Pace Of Change ### Slide 2: Our Brains Are Not Wired For This Curve **On slide:** Our planning is linear. AI progress is not. **Speaker note:** Most business planning assumes we can see the next few steps clearly enough to plan around them. AI is not moving like that. The pace is exponential, which is why even experienced leaders can feel like the ground is shifting. **Visual idea:** Simple exponential curve. --- ### Slide 3: A Way To Measure The Shift **On slide:** How long can AI stay useful on a task? **Speaker note:** One useful way to track the pace is METR's "time horizon" research. METR measures the length of software, machine learning, and cybersecurity tasks that AI agents can complete at a given success rate. The yardstick is how long a human expert would take to do the same task. **Plain-English explainer:** If METR says an AI agent has a 50% time horizon of 1 hour, that does not mean it thinks for exactly one hour. It means the agent is predicted to succeed about half the time on tasks that would take a qualified human roughly an hour. **Caution to say out loud:** This is not a measure of all jobs. It is a clean, measurable signal from technical tasks. Real organisational work is messier, more social, and harder to score. That is exactly why leaders need to understand the direction of travel, then apply judgement. **Source:** [METR: Task-Completion Time Horizons of Frontier AI Models](https://metr.org/time-horizons/) (last updated 08/05/2026) --- ### Slide 4: What The Latest METR Data Shows **On slide:** From minutes to hours to working-day-scale tasks **Speaker note:** The latest METR dashboard shows frontier agents moving from short tasks toward multi-hour tasks. On METR's 50% time-horizon view, the current frontier is already pushing against the edge of what their task suite can measure. The dashboard shows Claude Mythos Preview (early) at roughly 17 hours, and METR explicitly warns that measurements above 16 hours are unreliable with the current task suite. **Key line:** Do not obsess over the exact number. Watch the direction: AI agents are moving from helpful answers into sustained work. **Speaker credibility note:** Do not oversell this as "AI can do every 17-hour job." METR itself says the task suite is mostly software engineering, machine learning, and cybersecurity, and that measurements above 16 hours are currently unreliable. **Animation idea for HTML level:** Animate a time bar across 2023 -> 2024 -> 2025 -> 2026, with markers moving from seconds/minutes to hours. End with a warning band after 16 hours: "measurement frontier". --- ### Slide 5: Coding Agents As The Signal **On slide:** Measured in software. Relevant everywhere. **Speaker note:** Coding agents are a useful signal because the tasks are measurable. Seafood leaders do not need to care about code. The point is that the capability curve is now visible in a domain where we can test it. AI systems are getting better at multi-step work that requires context, tool use, and persistence. **Seafood translation:** If agents can sustain longer technical tasks, leaders should expect the same direction of travel in business work: board preparation, compliance support, market analysis, stakeholder feedback, and project coordination. --- ### Slide 6: The Leadership Question **On slide:** The question is not whether AI is coming. The question is how you lead through it. **Speaker note:** Waiting for certainty is not a strategy. The organisations that move well now will build confidence, capability, and judgement while others are still trying to decide where AI belongs. --- ## 2. The Gap: What Is Possible Right Now ### Slide 7: The Gap **On slide:** What AI can do vs what most people have experienced **Speaker note:** Most people have tried AI, got a polished but average response, and quietly concluded it is overhyped. Usually, the technology is not the main problem. The brief is. **Key line:** Most people are asking a brilliant new team member to do useful work without onboarding them. --- ### Slide 8: Three Layers Available Now **On slide:** 1. Conversational AI 2. AI automation 3. Agentic AI **Speaker note:** These are not distant future categories. Organisations can work with all three now, at different levels of maturity and risk. **Visual / demo video treatment:** Use this as a quick "show the work" moment, not a static list. Three short clips or animated mock-ups: 1. **Conversational AI:** voice briefing into an AI chat, with the output becoming a useful leadership note. 2. **AI automation:** one trigger creates a routed workflow, for example meeting notes turning into decisions, actions, risks, and follow-ups. 3. **Agentic AI:** an agent-style org chart or operating system view, where different assistants have roles, boundaries, and escalation paths. **Design inspiration from screenshots:** - Agent org chart: CEO / CMO / COO style structure, translated into "Leader / Compliance / Market Intel / Comms / Board Prep" assistants. - Personal operating system: grouped playbooks under operating areas, translated into "Compliance, Operations, Market, People, Governance". - Playbook card: trigger, inputs, steps, outputs. Use this for one seafood workflow example. **Seafood visual concepts:** - **Seafood AI Operating System:** Compliance, Operations, Market Signals, People, Governance. - **Daily Leadership Brief:** calendar + inbox + meeting notes + sector news -> risks, decisions, follow-ups. - **Compliance Reporting Playbook:** trigger, inputs, steps, outputs, human sign-off. --- ### Slide 9: Conversational AI **On slide:** More than polishing emails **Speaker note:** Conversational AI is useful as a thought partner. It can support strategic thinking, analysis, decision prep, and communication. A CEO can use AI to stress-test a new strategy against different stakeholder perspectives before taking it to the board. **Example prompt:** "If this fails in 12 months, why did it fail?" **Demo clip 1: Great brief into conversational AI** Show the difference between a vague prompt and a useful strategic-thinking brief. **Weak prompt:** "Help me think about this strategy." **Strong prompt:** I am the Chief Executive of a New Zealand seafood organisation preparing to take a strategic proposal to the board. The sector is dealing with fixed harvest, rising costs, compliance burden, workforce pressure, and market uncertainty. Act as a strategy adviser who can think from three stakeholder perspectives: CE, Compliance, and Operations. Stress-test this proposal before I take it to the board. Start by asking me up to three clarifying questions, then give me the strongest risks, assumptions, and questions I should be ready to answer. **Key line:** Same model. Better brief. Completely different usefulness. --- ### Slide 10: AI Automation **On slide:** Meeting follow-up that does not disappear. **Speaker note:** AI automation works best when the process is understood. Take a meeting-follow-up workflow that currently relies on memory, inboxes, and people chasing each other. Map the steps, identify the handoffs, and use AI to compress the admin while keeping people responsible for judgement and sign-off. This is useful for the work nobody quite owns: chasing people up, aggregating feedback, keeping the latest version of a plan current, and turning scattered inputs into something a decision-maker can use. **Seafood grounding:** Compliance, reporting, documentation, internal updates, export paperwork, meeting actions, stakeholder feedback, risk registers, seasonal planning. **Demo clip 2: Capture -> Triage -> Surface -> Retrieve** Use a sanitised shared operating layer example. **On-screen workflow:** Meeting transcript + email thread + Teams chat + paper/voice note -> AI triage -> human review -> role-specific action files and context log. **Triage buckets:** - My commitments - My delegations - Decisions made - FYI-searchable - Noise **Outputs:** - `Active_Work.md` for actions, waiting items, decisions, and backlog - `Context_Log.md` for decisions, rationale, facts, quotes, and open questions - `Named_Entities.md` so the system knows which people, projects, and open questions matter **Speaker note add-on:** This is not glamorous. That is the point. A lot of AI value sits in the boring handoffs where work currently leaks. --- ### Slide 11: Agentic AI **On slide:** AI that can work across a folder, not just a prompt. **Speaker note:** Agentic AI is the newest frontier. This is AI that goes past answering a question and works through a task. A CEO could point an AI agent at a folder containing board papers, last quarter's financials, and a strategic proposal. The agent cross-references the material, flags inconsistencies, pressure-tests assumptions, and produces a consideration document. Another way to think about it is a chief of staff for information work. It can scan the calendar, inbox, meeting notes, planning documents, and live project updates, then surface what needs attention: risks, decisions waiting on someone, deadlines slipping, and today's three most important things. **Demo clip 3: Board-pack agent** Point AI at a sanitised folder with: - Board papers - Last quarter's financials - Strategic proposal - Compliance obligations / policy notes **Output to show:** A consideration document with three lenses: - **CE:** strategic direction, stakeholder confidence, board-level risk - **Compliance:** obligations, gaps, evidence, review points - **Operations:** feasibility, resourcing, sequencing, frontline impact **Key line:** This is where AI starts to look less like a chatbot and more like a junior analyst with access to the working folder. **Guardrail line:** More autonomy does not mean less accountability. --- ### Slide 12: The Seafood Relevance **On slide:** Fixed harvest. Rising costs. Compliance burden. AI does not make more fish. **Speaker note:** This is where the "do more with less" message lands. AI does not make more fish. It makes the work around the fish faster and cheaper to manage. **Optional line:** The opportunity is in the work around the fish. **Examples to draw on:** - Market and sentiment research condensed from days to hours - Daily leadership briefing from email, calendar, and meeting notes - Launch or project coordination across functions - Feedback aggregation from stakeholders, customers, or members - Compliance and policy updates summarised against operational context - Reducing the cognitive load of low-importance, high-friction admin --- ### Slide 13: Clear Outcomes Change The Work **On slide:** Vague instruction creates hidden work. Clear outcome reduces it. **Speaker note:** When we talk to humans, we rely on shared assumptions. If I say, "Go to the store and pick up some milk," a person knows the hidden steps: go to the store, pick up the milk, pay for it, bring it home, put it in the fridge. AI does not share all of that context unless we make the outcome clear. **Example to test / verify before final deck:** Reasoning models can spend far more hidden effort on vague constraints than concrete ones. "Answer in as few words as practical" is still vague. "Answer in 10 words or less" gives the model a defined target. **Practical translation:** The same applies to slide decks, briefing notes, reports, and plans. "Make me a presentation" is weak. "Create a 10-slide deck with these five sections, for this audience, using these examples, with one decision slide at the end" is a usable brief. **Key line:** Clear is not dumbing it down. Clear is how you get better work. --- ## 3. Why Most Organisations Stall ### Slide 14: Why AI Feels Underwhelming **On slide:** Expectation: magic Reality: mediocre first draft **Speaker note:** Most people try AI, get mediocre results, and conclude it is overhyped. That is like hiring a new team member, giving them no context, and deciding they are useless after the first task. **Key line:** The first response is the beginning of the conversation. --- ### Slide 15: The Human Teammate Frame **On slide:** Onboard it. Brief it. Question it. Give feedback. **Speaker note:** Use a familiar frame. If a new person joined your organisation, you would give them context, a role, examples, expectations, and feedback. AI needs the same treatment. **Simple framework:** - Context: what does it need to know? - Role: what expertise should it bring? - Task: what do you want back? - Questions: what should it ask before proceeding? - Feedback: what needs to improve? --- ### Slide 16: The Wrong First Move **On slide:** Tool. Training. Policy. None is the right first move. **Speaker note:** These are all reasonable responses. Buy a tool. Run AI training. Write a policy. None of them is stupid. They are just not the right first move if you do not yet know what is already happening. **Build-out note:** The tool can add another layer on top of invisible usage. The training can improve general awareness without improving how this team uses AI on this work. The policy can describe what should happen while nobody has mapped what is actually happening at the moment of use. **Key line:** Most organisations reach for the solution before they have looked at the problem properly. --- ### Slide 17: Map What Is Already Happening **On slide:** Map the real usage first. **Speaker note:** Not what has been approved. Not what has been assumed. What is actually happening across the team: which tools, on which accounts, with what data, under what review habits, and with what consistency across functions and seniority levels. **Speaker note add-on:** That real picture nearly always changes what leaders think they need next. The tool question does not go away. The training question does not go away. They just become answerable once there is something honest to build on. **Possible seafood link:** The stakes are higher in sectors dealing with regulation, client trust, confidential data, and public scrutiny. You need to know what is happening before someone asks without warning. **Coordination angle:** Also map where the coordination load sits. Where are people chasing updates? Where do plans go stale? Where is the same information copied across email, spreadsheets, documents, Teams, and meeting notes? Those are often the first places AI creates practical relief. --- ### Slide 18: Faster Chaos **On slide:** Point AI at chaos, get faster chaos. **Speaker note:** AI amplifies what is already there. If the process is messy, undocumented, or politically unclear, AI will not magically fix it. It will move the mess faster. Process clarity comes first. --- ## 4. Leadership's Role ### Slide 19: 90% People, 10% Tech **On slide:** AI adoption is 90% people, 10% tech. **Speaker note:** This cannot be delegated to IT. IT has a serious role, especially for security and systems. The adoption challenge is leadership, behaviour, trust, and work design. **Key line:** You cannot outsource understanding. --- ### Slide 20: AI Transformation Is Management **On slide:** AI transformation is management. **Speaker note:** The more capable AI becomes, the more this starts to look like management. Who is responsible for what? What can be delegated? Where does ambiguity go? Who has authority to override? Where does judgement sit? **Speaker note add-on:** With agentic AI, a lot of the language comes from software: control planes, hooks, loops. Useful, but incomplete. Management and organisational design have plenty to teach us: spans of control, decision rights, escalation paths, functional and divisional structures. **Key line:** Most teams building agentic systems are accidentally rebuilding management problems that organisations solved decades ago. **Seafood translation:** If an AI assistant is helping with compliance, market research, or board prep, the stronger question is: "what is this assistant allowed to do, what must it never do, who checks its work, and where does uncertainty go?" --- ### Slide 21: Draw The Agent Org Chart **On slide:** Before the workflow, draw the org chart. **Speaker note:** The fix is not importing management theory into prompts. It is designing the agent system the way a competent operator would design a human team: who reports to whom, who can override whom, what each agent is allowed to decide, and where ambiguity stops. **Seafood examples:** - A compliance-support agent should not have the same authority as a final human approver. - A board-briefing agent can flag inconsistencies, but leaders still own recommendations. - A market-intelligence agent can summarise signals, but commercial decisions need human judgement. - A daily-briefing agent can shield noise, but it needs clear rules for what must always reach a human. --- ### Slide 22: What The Pulling-Ahead Organisations Do **On slide:** Visible leadership. Safe learning. Useful first use cases. **Speaker note:** Buying tools is the easy part. The organisations pulling ahead have leaders visibly using AI, making it safe to learn, and giving people a better reason than "be more productive". **Example:** Christchurch Airport started with People & Performance. --- ### Slide 23: The Vision Has To Be Better Than Productivity **On slide:** Nobody gets excited about being 10% more productive. **Speaker note:** People need to understand what the freed-up capacity is for. Better decisions. More time with customers. Less admin pressure. Stronger compliance confidence. Faster response to market change. --- ## 5. Three Things To Do This Month ### Slide 24: Three Practical Moves **On slide:** 1. Pay for a capable tool 2. Talk to it 3. Brief one AI teammate **Speaker note:** These are immediate actions, wherever an organisation is starting from. **Details:** - Pay for a capable tool and set privacy controls properly. - Use voice to brief AI like a person, not a search box. - Pick one recurring piece of work and build a simple AI teammate around it. **Example first teammates:** - Daily chief-of-staff brief: what needs attention today, what is waiting, what is at risk - Board prep assistant: summarise papers, flag inconsistencies, prepare questions - Compliance update assistant: summarise new information against internal context - Meeting intelligence assistant: turn discussions into decisions, actions, and risks - Market signal assistant: condense public sentiment, sector news, and customer feedback --- ### Slide 25: Choose One Starting Point **On slide:** Start where the work already hurts. **Speaker note:** Choose one practical starting point: board prep, compliance reporting, internal communications, meeting notes, staff onboarding, policy updates, market research, or daily coordination. Do not start with an abstract AI strategy. Start with one real piece of work. **Alternative phrasing to test:** Start by mapping one real piece of work before choosing the tool, training, or policy response. --- ### Slide 26: Close **On slide:** Own the playbook. Rent the tech. **Speaker note:** The tools will keep changing. Your processes, judgement, standards, relationships, and sector knowledge are the asset. Build the playbook around those. **Final line option:** AI will not make more fish. It can help you make better use of the time, expertise, and judgement around the fish. --- ### Slide 27: Q&A **On slide:** Questions **Likely questions:** - What about privacy and confidential data? - Which tool should we start with? - How do we stop AI making things up? - Is this replacing people? - Where should a seafood business start? - What should stay human? --- ## Demo Recording Checklist - [ ] Record Demo 1: conversational AI stress-testing a strategy from a strong brief. - [ ] Record Demo 2: meeting follow-up / operating-layer workflow using Capture -> Triage -> Surface -> Retrieve. - [ ] Record Demo 3: agentic board-pack review using a sanitised folder. - [ ] Fully sanitise all source context before recording. - [ ] Replace Highlands names/details with seafood-sector roles and generic files. - [ ] Export three short clips or one continuous recording with clear chapter cuts. - [ ] Take screenshots as backup in case video playback fails. --- ## Source Ideas To Translate, Not Name-Drop **Thibault Sottiaux / Chris Nicholson source themes:** - AI as a chief of staff for information work: daily briefs, risk flags, calendar/inbox/document synthesis. - AI reducing coordination load: chasing updates, aggregating feedback, keeping plans current. - AI compressing research: market analysis, sentiment analysis, source condensation. - AI lowering cognitive load: handling tedious, never-quite-important tasks that otherwise sit undone. - Natural language and voice as the path to "home-cooked" software and workflows for non-coders. - Agentic work needs management design: decision rights, escalation paths, authority, review, and boundaries. **Seafood-specific translation:** Use these as examples of practical AI capability in sector leadership work, not as startup/coding examples. Frame them around board prep, compliance load, market uncertainty, stakeholder communication, member services, seasonal planning, and reducing information overload for leaders. --- ## Open Decisions 1. Do we keep all three demo clips, or record all three and use only the strongest two on the day? 2. Do we keep the coding-agent stat as the visual proof point, or swap for a more general AI progress benchmark? 3. Is Christchurch Airport strong enough as the named example, or do we need a second example? 4. Should the close be "Own the playbook, rent the tech" or "AI does not make more fish"? 5. Do we keep the "tool, training, policy" sequence in the main talk, or save it for Q&A? 6. Do we use "AI transformation is management" as a named slide, or keep it as a speaker-note insight? 7. Do we use the "chief of staff" frame explicitly, or translate it into "daily leadership brief" for this audience?