EP 26: Prompt Engineering - How to Actually Prompt AI and Stop Getting Average Answers

EP 26: Prompt Engineering - How to Actually Prompt AI and Stop Getting Average Answers

Why does AI give some people sharp, useful output and everyone else a generic, average answer? In Episode 26 of ⁠ ⁠Zero Shot⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ [https://zeroshot.com.au/] , ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠James Gauci⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ [https://www.linkedin.com/in/jamesrgauci/] and ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Llew Jury ⁠ [https://www.linkedin.com/in/llewjury/] go deep on prompt engineering, the skill that separates people getting real value from AI and people quietly disappointed by it. The honest truth is it is almost never the model. The models are astonishing. It is the brief. This episode covers why prompts underperform at a technical level, the four pillars of a strong prompt, and how everything changes when you move from a single prompt to designing the full context environment of an AI agent. Key Highlights * Why weak output is a briefing problem, not a model problem * How a prompt reshapes the model's internal landscape of weights * Making the implicit explicit is now a core professional skill * The four pillars: role, context, task, format * Set the role at world-class level to lift the output * Less context done well beats more context done poorly * The four parts of an agent brain: short memory, long memory, instructions, knowledge * Prompting an agent is continuous improvement, not a one-time write Tools and Frameworks Mentioned * RCTF Framework (Advancer) [https://advancer.com.au] - Llew's prompt structure covering Role, Context, Task, and Format, designed to be printed and reused across a team. * ChatGPT Work with Sol - OpenAI's flagship high-power model in its new work mode, reviewed live against Fable 5 for leadership briefing tasks. * Hermes [https://github.com] - An agent harness that prompts and improves itself between sessions, tuning its own skills and memory as a continuous improvement loop. * Retrieval Augmented Generation and Data Lakehouses, [https://www.databricks.com/glossary/data-lakehouse]The shift from classic RAG toward trusted single-source-of-truth architectures like Snowflake and Databricks for real business context. Shape the show: https://zeroshot.com.au/shape Connect with the Hosts: * ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠James Gauci:⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ [https://www.linkedin.com/in/jamesrgauci/] CEO of Cadent. * ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Llew Jury:⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ [https://www.linkedin.com/in/llewjury/] Managing Director of Advancer at the AI Agency. * ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Zero Shot:⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ [https://zeroshot.com.au/] Follow us on LinkedIn for clarity in a complex landscape. 🤠 Sponsor: This episode is brought to you by ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Prompt Cowboy⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ [https://www.promptcowboy.ai/], the agentic prompting tool that helps teams build structured, precise prompts before they hit whichever model they are using. Head to promptcowboy.ai and stop leaving results on the table. Produced by ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Yennia La Rotta⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ [https://www.linkedin.com/in/zarich-la-rotta/].⚡

EP 25: What AI Should Actually Be Doing in Your Business

How do you tell the difference between a task worth handing to AI and one that should stay human? In Episode 25 of Zero Shot⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ [https://zeroshot.com.au/] , ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠James Gauci⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ [https://www.linkedin.com/in/jamesrgauci/] and ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Llew Jury [https://www.linkedin.com/in/llewjury/] tackle the decision most businesses get wrong: choosing what to actually point AI at. The instinct is to reach for the shiny, visible task: customer service, the website, email, but that is rarely where the value lives. This episode gives you a practical way to sort any task in your business into three piles: hand it to AI, keep it human, or a bit of both. Five fast triggers to spot the real candidates, seven questions to run each one through, and an honest look at the cost of automating the wrong thing. Key Highlights * Why your gut instinct to automate is usually the wrong target * Automating a broken process just makes bad permanent, faster * Failed early automations burn organizational trust and kill momentum * The five triggers: repeat, sentence, dread, messy input, apology * Let the AI do the dread work, not just the grunt work * Seven questions that separate worth it from should we * In regulated sectors, accountability is the governance question that matters * The should conversation is where the real value lives Tools and Frameworks Mentioned * Claude Fable [https://claude.ai] - Anthropic's high-capability model, restored last week in a stripped-back format outside the US, referenced for producing report and website work that would take an analyst weeks. * Gemma 4 (Google DeepMind) [https://deepmind.google/technologies/gemma] - Google's open source model, cited as the kind of on-hardware model deployed in high-classification defense settings where public models are not permitted. * The REWIRE Framework (Advancer) [https://advancer.com.au] - Advancer's method for AI transition, moving through Reveal, Envision, and Weave to test viability and feasibility before building. * System 1 and System 2 Thinking [https://en.wikipedia.org/wiki/Thinking,_Fast_and_Slow] - Daniel Kahneman's model, referenced by James to frame when to let AI steer a fast call versus when deep human judgment is required. Connect with the Hosts: * ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠James Gauci:⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ [https://www.linkedin.com/in/jamesrgauci/] CEO of Cadent. * ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Llew Jury:⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ [https://www.linkedin.com/in/llewjury/] Managing Director of Advancer at the AI Agency. * ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Zero Shot:⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ [https://zeroshot.com.au/] Follow us on LinkedIn for clarity in a complex landscape. 🤠 Sponsor: This episode is brought to you by ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Prompt Cowboy⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ [https://www.promptcowboy.ai/], the agentic prompting tool that helps teams build structured, precise prompts before they hit whichever model they are using. Head to promptcowboy.ai and stop leaving results on the table. Produced by ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Yennia La Rotta⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ [https://www.linkedin.com/in/zarich-la-rotta/].⚡

EP 24: Process Automation with AI Agents - How to Think Before You Build

What actually changed between the automations businesses built ten years ago and the AI agents being built today? In Episode 24 of Zero Shot⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ [https://zeroshot.com.au/] , ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠James Gauci⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ [https://www.linkedin.com/in/jamesrgauci/] and ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Llew Jury [https://www.linkedin.com/in/llewjury/] go back to first principles on process automation. They trace the shift from rigid, rules-based workflows to AI agents that reason over messy, unstructured data, and unpack what that unlocks for automation, autonomy, and agentic workflows inside real organisations, including regulated and compliance heavy sectors. It is a durable mental model for deciding what to build and when, not a tool tour that dates in six months. Key Highlights * Automation no longer requires structured data to be useful * Deterministic rules have given way to systems that reason * Automation and autonomy are different capabilities that nest together * Data lakehouses feed agents directly through MCP connections * Real time finance agents end month end reconciliation waits * Regulated sectors can build sovereign, onshore, fully controlled AI * The four Ps framework leaves tool selection until last * Start with the messiest high value process first Tools and Frameworks mentioned in this episode * OpenWhispr [https://openwhispr.com], open source dictation that runs speech models locally with no data egress. * Model Context Protocol (MCP) [https://modelcontextprotocol.io], the open standard for connecting agents to data sources like lakehouses. * Microsoft 365 Copilot Cowork, [https://www.microsoft.com/en-us/microsoft-365-copilot] the agentic layer inside Copilot for scheduled, delegated work. * Data lakehouse architecture, the blended structured and unstructured data foundation agents feed on. Widely documented across major cloud platforms. * The Four Ps, Llew's sequence of problem, people, process, then product for automation projects. A working framework from the episode. Connect with the Hosts: * ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠James Gauci:⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ [https://www.linkedin.com/in/jamesrgauci/] CEO of Cadent. * ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Llew Jury:⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ [https://www.linkedin.com/in/llewjury/] Managing Director of Advancer at the AI Agency. * ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Zero Shot:⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ [https://zeroshot.com.au/] Follow us on LinkedIn for clarity in a complex landscape. 🤠 Sponsor: This episode is brought to you by ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Prompt Cowboy⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ [https://www.promptcowboy.ai/], the agentic prompting tool that helps teams build structured, precise prompts before they hit whichever model they are using. Head to promptcowboy.ai and stop leaving results on the table. Produced by ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Yennia La Rotta⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ [https://www.linkedin.com/in/zarich-la-rotta/].⚡

EP 23: Beyond Automation: AI and the Future of Creative Work with Ben Cooper, R/GA

What if the way most people are using AI is actually the lowest value thing it can do? In Episode 23 of ⁠Zero Shot⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ [https://zeroshot.com.au/] , ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠James Gauci⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ [https://www.linkedin.com/in/jamesrgauci/] and ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Llew Jury [https://www.linkedin.com/in/llewjury/] sit down with their second-ever guest, Ben Cooper, Global Executive Director of AI Products at R/GA and one of Australia's most original applied AI builders. From the world's first AI shark detection system to brand context protocols that make companies machine-readable, Ben shares a genuinely contrarian view: that the real value of AI is not generating faster and cheaper, but treating intelligence as a raw material you build with. This is an honest conversation about what survives contact with a real client, how creative work is priced and protected, and the human reckoning underneath the whole AI debate. Key Highlights * Why treating intelligence as a material beats treating it as a tool * 68 percent of brands now receive zero click traffic * Large language models trust reviews and reputation over marketing speak * Setting up AI to push back on you produces better outcomes * Time, tools, and tokens are the new pricing variables * Tangible demos make ideas unstoppable in the client room * The backlash against AI is a needed correction, not just noise * Personal agency, not the tool, is what takes you somewhere Tools and Frameworks Mentioned * NotebookLM: [https://notebooklm.google.com⁠] Google's context-building tool for organising and connecting knowledge across projects * Perplexity: [https://www.perplexity.ai⁠] AI-powered search and news discovery platform, personalised by topic and interest * ElevenLabs: [https://elevenlabs.io⁠] Voice AI platform used to train, license and scale branded voiceover at the production level * Grok: [https://grok.com⁠]xAI's model with native access to X (Twitter) data, useful for real-time knowledge mining * Schema and Markdown Files: [https://schema.org] Foundational web standards that make a brand machine-readable and discoverable inside answer engines. Connect with the Hosts: * ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠James Gauci:⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ [https://www.linkedin.com/in/jamesrgauci/] CEO of Cadent. * ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Llew Jury:⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ [https://www.linkedin.com/in/llewjury/] Managing Director of Advancer at the AI Agency. * ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Zero Shot:⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ [https://zeroshot.com.au/] Follow us on LinkedIn for clarity in a complex landscape. 🤠 Sponsor: This episode is brought to you by ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Prompt Cowboy⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ [https://www.promptcowboy.ai/], the agentic prompting tool that helps teams build structured, precise prompts before they hit whichever model they are using. Head to promptcowboy.ai and stop leaving results on the table. Produced by ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Yennia La Rotta⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ [https://www.linkedin.com/in/zarich-la-rotta/].⚡

EP 22: Frameworks for Using AI Right - The Questions to Ask Before You Start

What are the questions worth asking before you trust AI with anything important in your business? In Episode 22 of Zero Shot⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ [https://zeroshot.com.au/] , ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠James Gauci⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ [https://www.linkedin.com/in/jamesrgauci/] and ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Llew Jury [https://www.linkedin.com/in/llewjury/] lay out the practical frameworks they wish someone had handed them when they started: the questions to ask before you choose a tool, before you trust an output, and before you scale AI across an organisation. Broken into three clear acts, before you start, before you trust, and before you scale, this episode is a working checklist for making AI decisions that protect your business, your data, and your customer relationships. Honest, specific, and grounded in real consulting experience across regulated and high-stakes sectors. Key Highlights * Start with the problem, not the tool or the model * Only two to five percent of businesses are doing real AI * Tool choice matters less than people, process, and data alignment * Turning off model training is the first trust step * Data hosted in Australia can still be subject to US law * Governance done right is a value driver, not a cost * Accountability across the team costs almost nothing to implement * Pick any decision making framework, the act of choosing matters most Tools and Frameworks Mentioned Open Whisper [https://github.com/openai/whisper⁠]: Free, open source dictation tool that runs the model locally on your device, keeping sensitive audio off the cloud. Excalidraw: [h⁠⁠ttps://excalidraw.com⁠⁠]Collaborative virtual whiteboard for diagrams and mind mapping, with an MCP connector that lets Claude design directly into the canvas. ISO 42001: [https://www.iso.org/standard/81230.html⁠]The international standard for AI management systems, referenced as a practical starting framework for businesses building AI governance. National AI Centre Frameworks: [https://www.industry.gov.au/science-technology-and-innovation/technology/national-artificial-intelligence-centre⁠]Australian government resources and policies offering free, credible starting points for small businesses building AI governance and usage policies. NextDC: [https://www.nextdc.com⁠] Australian data centre operator referenced in the discussion on data sovereignty and onshore hosting for sensitive workloads. Connect with the Hosts: * ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠James Gauci:⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ [https://www.linkedin.com/in/jamesrgauci/] CEO of Cadent. * ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Llew Jury:⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ [https://www.linkedin.com/in/llewjury/] Managing Director of Advancer at the AI Agency. * ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Zero Shot:⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ [https://zeroshot.com.au/] Follow us on LinkedIn for clarity in a complex landscape. 🤠 Sponsor: This episode is brought to you by ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Prompt Cowboy⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ [https://www.promptcowboy.ai/], the agentic prompting tool that helps teams build structured, precise prompts before they hit whichever model they are using. Head to promptcowboy.ai and stop leaving results on the table. Produced by ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Yennia La Rotta⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ [https://www.linkedin.com/in/zarich-la-rotta/].⚡