Are We Bold Enough? A Practical AI Field Guide for Seafood NZ

Waiting For Certainty Is Not A Strategy

Why This Matters Now

The coding-agent example in the keynote was a signal about pace.

In twelve months, AI coding agents showed a capability jump of more than 500 percent on benchmarked technical tasks. They can now complete some tasks that take humans around three hours, at roughly 80 percent accuracy, much faster and cheaper.

The useful point for seafood leaders is the speed of change. A lot of people are still meeting AI through:

• Free tools.
• Simple prompts.
• Copy and paste.
• The same prompt every time.
• Smart Google-style questions.

That gives a thin view of what is already possible.

Five Things Worth Remembering

1. AI reactions are human. Excitement, scepticism, fear, curiosity, resistance, and hope can all be valid.
2. Most people are still underusing AI. Simple prompts usually produce simple value.
3. The shift is from prompting to delegating. Better AI work needs better briefs, examples, context, and boundaries.
4. Voice changes the game. Talking gives AI richer context faster than typing, especially when the work is messy.
5. AI adoption is a leadership challenge. Tools matter. People, trust, workflow design, and judgement matter more.

Lines Worth Keeping

Waiting for certainty is not a strategy.

If you point AI at chaos, you get faster chaos.

The record button is your friend.

AI is ninety percent people and ten percent tech.

The Big Idea

AI changes the old equation between effort, time, expertise, and value.

In the past, it was easy to assume that a piece of work was valuable because it took a long time, required a specialist, or needed a lot of manual effort. AI unsettles that. It can draft, summarise, compare, classify, extract, and organise work in seconds.

That raises the value of human judgement.

For seafood businesses and industry teams, start with this question:

Where is useful human attention being wasted on repeated, explainable work?

Good places to look:

• Crew onboarding and training notes.
• Maintenance logs and follow-up actions.
• Compliance documentation and evidence gathering.
• Customer emails, export paperwork, and product information requests.
• Meeting summaries across operations, sales, logistics, leadership, and industry working groups.
• Turning messy operational knowledge into usable internal guidance.
• Analysing incident reports, customer feedback, or recurring admin patterns.

If You Point AI At Chaos, You Get Faster Chaos

AI amplifies the way work already happens.

Clear processes become faster. Messy processes become faster mess.

Before you automate a task, map it:

• What triggers the work?
• What information comes in?
• Who touches it?
• What decisions are made?
• What output is needed?
• Where does human approval belong?
• What could go wrong?

Start with the work, then choose the tool.

Chat, Automation, And Agents

AI capability is moving from one-off help to more delegated work. A simple way to think about it:

The shift is from "answer this question" to "help me do this piece of work."

Brief AI Like A Digital Teammate

AI works better when it has the same kind of briefing you would give a capable person joining your team.

Give it:

• Context about the business, market, customer, team, and constraints.
• A clear job description for the role you want it to play.
• Task briefs matched to the stakes of the work.
• Examples of what good looks like.
• Feedback when the output misses.

The live demo used a seafood-sector CEO trying to build support for an AI capability programme. Marketing was excited. Operations was under pressure. The CFO thought AI was overhyped and too expensive.

That is the kind of context AI needs. "Write me an AI plan" will get a bland answer. "Help me prepare for these three different reactions inside my business" gives it something useful to work with.

Use YETI When You Brief AI

The record button is your friend. Use voice to give AI the messy context you would never bother typing.

YETI is a simple way to brief AI properly:

Your 20-Minute AI Experiment

Pick one task that is:

• Repeated.
• Annoying.
• Low-risk.
• Easy to explain.
• Not dependent on perfect data.

Then fill this in:

Task:

Who does it now:

What triggers it:

Inputs needed:

Steps:

Output:

Human approval point:

What could go wrong:

What would success look like:

Good first experiments for seafood-sector work:

• Summarise a meeting transcript into decisions, actions, owners, and follow-ups.
• Turn handwritten or messy notes into a training checklist.
• Draft a customer or buyer response from approved product information.
• Pull recurring issues from incident reports or maintenance notes.
• Create a first draft of a compliance evidence checklist.
• Sort a shared inbox into enquiry types and draft replies for approval.

AI Is Ninety Percent People And Ten Percent Tech

AI cannot simply be handed to IT, a vendor, or the keenest person in the room.

It needs executive direction and bottom-up champions. The people closest to the work often know where the best opportunities are.

Practical leadership moves:

• Give people time to learn.
• Find the people already experimenting.
• Ask what is working, what feels risky, and what they wish leadership understood.
• Create simple guidance for privacy, data, review, and tool use.
• Start small enough that people can learn while keeping the business protected.

Risks And Guardrails

Use AI with judgement.

Some work is low-risk: drafting, summarising, organising, brainstorming, and preparing first cuts.

Some work needs tighter control: safety, compliance, quota, employment, external communications, contracts, environmental decisions, regulatory work, and commercial decisions.

Good guardrails:

• Turn off model training where appropriate.
• Upload sensitive documents only after checking tool settings and permissions.
• Keep human review before anything goes to customers, regulators, staff, or the public.
• Use approved source material when drafting factual answers.
• Ask AI to show assumptions, gaps, and questions.
• Record decisions about what tools are allowed and what information can be used.

Tools Mentioned

These tools were mentioned as examples. Start with the work you want to improve, then choose the tool that fits your ecosystem, risk level, and budget.

• ChatGPT and Microsoft Copilot: Everyday chat-based AI tools for drafting, planning, summarising, and testing ideas.
• Claude and Claude CoWork: Used as examples of file-based and agent-style AI work. Dispatch is the mobile function used to send instructions to a Claude CoWork agent while away from the desk.
• Wispr Flow: Voice dictation tool Nadia uses to talk to AI by voice.
• Relay.app: No-code automation tool mentioned as one way to experiment with repeated workflows.
• Teams, SharePoint, and Outlook: Workplace tools that can form part of AI-supported meeting, document, and follow-up workflows.
• Skills: Saved task instructions that help AI produce more consistent outputs.

Some resource links may include Curiosity referral or bonus links. Use them if useful. Ignore them if your organisation already has preferred tools.

Glossary

• Chat-based AI: AI you interact with through a conversation. Useful for questions, planning, drafting, summarising, and testing ideas.
• Smart Google: A limited way of using AI where you ask one question, get one answer, and move on.
• AI automation: AI connected to a repeatable process, usually triggered by something happening, such as a file arriving or a meeting ending.
• Agentic AI: AI that can work towards a goal, make a plan, use approved tools, check progress, and report back.
• Digital teammate: A practical way to think about AI. It needs context, clear work, examples, feedback, and boundaries.
• AI slop: Generic or poor-quality AI output, usually caused by weak briefing or accepting the first answer.
• Tokens: Units of AI usage. Larger or more complex requests usually use more.
• Sandbox: A boundary that limits where an AI tool or agent is allowed to work.
• Shadow AI: Unofficial AI use inside an organisation, often happening before leadership has created guidance.

Q&A Hub

Resources

• Seafood NZ keynote resource page
• Curiosity
• Nadia Ellis on LinkedIn
• Growth Medium AI
• Joe Sutheran on LinkedIn

Prepared from Nadia Ellis' Seafood New Zealand Conference keynote. For help turning these ideas into practical AI experiments, contact Nadia at nadia@curiosity.nz or visit curiosity.nz.