AI DUNN Right Weekly - Issue #9
- Nov 24, 2025
- 12 min read
Practical AI insights for business growth
Hey AI Innovators! 👋
Welcome back to AI DUNN Right Weekly! This week Google quietly dropped its most deployable image model yet, voice AI crossed from pilot to production with real revenue proof, and Kyle Balmer revealed why most entrepreneurs are overlearning and underearning.
Here's what matters for your business:
• Google's Nano Banana Pro generates physics-accurate images that reason before rendering
• Voice AI adoption hits 97% with $2M+ budgets, but only 21% are satisfied with current agents
• The 8 Levels of AI framework shows most people need basic user-level skills to make money, not advanced expertise
• ChatGPT problem-solving prompts that structure thinking for product, strategy, and operations decisions
• OpenAI's Codex Max handles multi-day refactoring without losing context
Read time: 5 minutes
🚀 This Week's Game Changer
Google's Nano Banana Pro - The AI Image Model Built for Universal Deployment
What happened: Google released Nano Banana Pro (Gemini 3 Pro Image), an AI image model that reasons before rendering. Instead of instantly generating visuals like diffusion models, it first checks physical and logical details like lighting, shadows, perspective, and depth. The result is dramatically more realistic multi-element scenes that look like actual photographs rather than AI-generated approximations.
The innovation: This ain't about making prettier pictures. Nano Banana Pro represents Google's strategy to deploy AI image generation across every product at planetary scale. It's tiny, efficient, and capable enough to run on phones, in browsers, and inside apps. The model supports 14 inputs simultaneously, maintains 5 characters consistently in 4K resolution, creates infographics with sharp real text, and handles multilingual typography. It's already live in the Gemini app and available across Gemini API, Vertex AI, Google Ads, and Workspace.
Business impact:
For Designers: Native image generation baked into every Google app eliminates friction that previously required third-party tools like Canva or Midjourney
For Developers: A model that runs at near-zero cost enables AI-native apps - coloring books, avatar generators, meme tools, product mockups - without prohibitive API expenses
For Corporate Teams: Automatic visual creation in Slides, Docs, Gmail, and Meet transforms how quickly teams produce professional materials
The bigger picture: Google doesn't need the flashiest model. It needs the most embeddable one. When you control 3 billion Chrome installs, 2.5 billion Android users, and 1.8 billion Gmail accounts, deploying a lightweight model everywhere beats having the best model nowhere. Nano Banana Pro follows the same playbook as Google Maps and Chrome: make it fast, make it free, make it everywhere.
Why it matters: The AI image war shifted from "who generates the best art" to "who embeds generation where people already work." Midjourney creates stunning visuals, but you need to leave your workflow to use it. Nano Banana Pro disappears into the tools you already use daily. That's how utility beats beauty in mass adoption. And when adoption becomes universal, the best standalone tool becomes the niche luxury option.
🛠 AI Tool Spotlight
Kyle Balmer's AI Proficiency Framework - Stop Overlearning, Start Earning
What it does: Kyle Balmer has developed a practical framework mapping eight distinct levels of AI proficiency that shows exactly where you need to be for different business models. Most entrepreneurs convince themselves they need advanced research-level expertise when they actually need solid user-level capabilities to start making money. This assessment eliminates the learning trap keeping brilliant people stuck in courses instead of building profitable businesses.
Kyle teaches entrepreneurs how to build AI workshop businesses through his training program, showing you how to turn your current AI knowledge into paid corporate training gigs. If you're thinking about running AI workshops but worried you don't know enough yet, his approach will show you that you're probably already qualified to start.
Key features:
Beginner Stage: First Contact - Experimenting with ChatGPT, figuring out basic prompts and capabilities
User Stage: Effective Application - Using AI for real work, understanding iteration, getting consistent outputs
Advanced User: Power Implementation - Explaining AI conceptually, using advanced techniques, building custom GPTs
Builder Stage: No-Code Creation - Creating applications using tools like Lovable, Cursor, v0 without traditional coding
Integration Stage: System Connection - Connecting AI to CRMs, databases, email through Zapier, Make, APIs
Developer Stage: Custom Building - Writing Python, implementing RAG systems, building production-ready applications
Advanced Development: Optimization - Fine-tuning models, running local inference, optimizing costs at scale
Research Stage: Innovation - Reading papers, training neural networks, designing new architectures
The business model reality:
AI authority and content creation - Requires basic to advanced user skills
Corporate workshops and training - Requires basic to advanced user skills
AI automation businesses - Requires advanced user to integration skills
AI applications and products - Requires builder to developer skills
Real example: You want to run AI workshops for marketing teams. Traditional thinking says you need to understand transformer architecture and neural networks. Reality says you need to be skilled at using the tools and know marketing problems inside out while communicating clearly. That's work that pays £2,000 per workshop. The additional theoretical knowledge is nice but shouldn't stop you getting started.
The workflow:
Honestly assess your current skill level based on actual capabilities, not aspirations
Decide which business model you want to pursue (authority, workshops, automation, products)
Identify the gap between your current abilities and what's actually required
Stop studying theoretical knowledge and start building at your current capability
Develop skills through doing rather than preparing for tests that don't exist
Best use cases: Breaking analysis paralysis that keeps skilled professionals stuck in learning mode, matching AI education to specific revenue goals instead of vague "become an expert" aspirations, giving yourself permission to monetize Level 2 skills while others chase Level 8 credentials.
Why it's a breakthrough: The smartest people often get in their own way. They second-guess, suffer imposter syndrome, and worry they're not smart enough. Less knowledgeable people don't have these worries - they're unburdened. That's why you know successful people in your industry who aren't particularly sharp. They knew enough to do the thing rather than preparing endlessly. Kyle's framework gives permission to start building with the AI skills you already have.
Want to learn how to turn your AI knowledge into paid workshops? Kyle Balmer teaches entrepreneurs exactly how to build AI workshop businesses. Check out his training at aiwithkyle.com to learn how to start running corporate AI training sessions even if you think you're not "expert enough" yet.
⚡ The 5-Minute AI Academy
Voice AI Crosses From Pilot to Production - What the 2025 Data Reveals
The conversation shifted from "should we test voice AI" to "how do we scale it profitably." Deepgram's 2025 State of Voice AI Report with Opus Research surveyed 400 senior leaders at $100M+ enterprises and revealed adoption patterns, budget priorities, and the satisfaction gap creating massive opportunities.
The Adoption Reality: 97% of organizations now use voice AI, with 84% increasing budgets in 2025. This ain't experimental technology anymore. But only 21% report being "very satisfied" with current legacy agents. That gap between universal adoption and low satisfaction represents the opportunity for businesses that implement human-like agents handling real tasks, reducing wait times, and lifting CSAT scores.
The 5 Strategic Insights:
1. Where Voice AI Actually Deploys
The breakthrough use cases aren't futuristic scenarios. They're practical business functions:
Customer service - Automated inquiry handling, troubleshooting, account questions
Order capture - Taking orders, processing requests, managing inventory queries
Task automation - Scheduling, reminders, data entry, basic workflows
Appointment booking - Calendar management, rescheduling, confirmation calls
These tasks share common traits: high volume, repetitive patterns, clear decision trees, and immediate ROI measurement. Companies don't start with complex negotiations. They automate the 80% of calls following predictable patterns.
2. What Separates Leaders From Laggards
The capabilities that determine success aren't about fancier AI models. They're operational fundamentals:
Latency - Response time under 1 second feels natural, over 2 seconds breaks conversation flow
Accuracy - Speech recognition above 95% prevents frustration, below 90% creates more work
Tooling - Integration with existing CRMs, databases, and business systems determines usefulness
Context handling - Multi-turn conversations that remember previous exchanges rather than resetting each response
Companies achieving high satisfaction scores nail these basics. Those struggling typically have one failing dramatically.
3. Why Legacy Agents Disappoint
Traditional IVR systems and early voice bots fail for predictable reasons:
Rigid menu trees that force callers into predetermined paths
Inability to handle nuanced requests or unexpected phrasing
No real-time adaptation based on caller sentiment or urgency
Disconnected from business systems requiring human handoff for simple tasks
The 79% dissatisfaction rate reflects technology that technically works but creates poor experiences. When your automated system frustrates customers more than helping them, adoption metrics become meaningless.
4. The Unit Economics Transformation
Voice AI economics flipped from cost center to profit driver:
Traditional call centers - £15-30 per handled call with human agents
Legacy IVR systems - £3-8 per call with 40-60% completion rates
Modern voice AI - £0.50-2 per fully automated resolution with 70-85% completion
The savings compound. A company handling 10,000 monthly calls saves £150,000-280,000 annually by automating even 70% of volume. These aren't theoretical projections - they're actual results from production deployments.
5. The Implementation Pattern That Works
Successful voice AI rollouts follow consistent paths:
Phase 1: Single Use Case (Months 1-3)Pick one high-volume, low-complexity task. Measure baseline metrics. Deploy AI agent. Compare results. Iterate based on actual call patterns rather than assumptions.
Phase 2: Expand Coverage (Months 4-6)Add adjacent use cases that share infrastructure. Customer service naturally extends to basic troubleshooting. Order capture expands to status updates. Each addition leverages existing integration work.
Phase 3: Optimize Performance (Months 7-12)Fine-tune latency, improve accuracy through training data from real calls, optimize conversation flows based on dropout analysis, integrate deeper into business systems.
Companies skipping phase 1 piloting and jumping straight to full deployment consistently struggle. Those treating initial rollout as learning opportunity build momentum through early wins.
The Lesson for Your Business:
If you're exploring voice AI, focus on these priorities:
Start narrow - One specific use case with clear success metrics beats ambitious everything-automation
Measure obsessively - Track completion rates, customer satisfaction, cost per interaction, human handoff frequency
Optimize basics - Latency and accuracy matter more than advanced features nobody notices
Integrate properly - Voice AI disconnected from business systems creates work rather than eliminating it
Action step: Identify one high-volume, repetitive task your business handles through phone calls or voice interactions. Calculate current cost per interaction. Research voice AI providers offering that specific capability. Run a 30-day pilot with 10% of volume. Compare metrics. Expand only if results beat baseline by meaningful margins.
Expected reality: Most businesses will automate 60-80% of routine voice interactions by 2026. The competitive advantage goes to companies implementing now while competitors wait for "perfect" solutions. Being first with "good enough" beats being last with "optimized." Production experience compounds faster than theoretical preparation.
💡 Prompt of the Week
ChatGPT Problem-Solving Framework
Stop reacting to symptoms and start finding root causes with structured analysis. This prompt breaks any problem into clear parts using frameworks built for decision analysis:
Act as a strategic problem solver. Create one systematic analysis framework for [PROBLEM TYPE].
Essential Details:
• Problem: [SPECIFIC CHALLENGE]
• Context: [RELEVANT BACKGROUND]
• Constraints: [TIME/BUDGET/RESOURCES]
• Stakeholders: [WHO'S AFFECTED]
• Desired Outcome: [SUCCESS LOOKS LIKE]
• Decision Timeline: [WHEN CHOICE NEEDED]
Create one problem-solving framework including:
1. Root cause analysis questions
2. Solution path generation (minimum 3 options)
3. Trade-off evaluation matrix
4. Risk and assumption mapping
5. Next step action items
6. Decision criteria for choosing path
Structured reasoning for better decisions. Keep under 200 words total.
Why this works: Most people solve problems by reacting to the loudest symptom rather than identifying underlying causes. This prompt forces you to break situations into layers: what you know, what you don't know, what must be true for solutions to work. The structure reduces cognitive noise and reveals the next best move. You shift from debating assumptions in your head to writing down actual analysis.
Best use case: Product decisions, strategy planning, operational improvements, hiring choices, customer issues, personal decisions requiring trade-off analysis. Particularly valuable when stuck between options that all seem equally valid or equally flawed. The framework clarifies which criteria matter most and which solutions actually address root causes rather than symptoms.
🧠 Quick Wins: 5 AI Tools Worth Investigating
Based on this week's newsletter coverage and emerging capabilities:
🎨 Nano Banana Pro - Google's physics-reasoning image modelUse case: Creating infographics, product mockups, architectural renders, multilingual designs, brand assets without leaving Google Workspace
🧠 Kyle Balmer's AI Workshop Training - Learn how to build a business teaching AI workshops to companiesUse case: Monetizing your current AI skills through corporate training, stopping overlearning and starting to earn, building authority in your industry
💬 Deepgram Voice AI - Production-grade speech recognition and voice agentsUse case: Automating customer service, order capture, appointment booking, task automation with sub-1-second latency
🔧 ChatGPT Problem-Solving Prompts - 20 structured frameworks for breaking down complex decisionsUse case: Product strategy, operational improvements, hiring decisions, customer issue resolution, personal choices requiring trade-off analysis
💻 OpenAI Codex Max - GPT-5.1 coding model handling multi-day refactoring autonomouslyUse case: Large-scale code refactoring, iterative debugging, test-driven development, extended reasoning tasks without context loss
📈 Business Intel: This Week's Market Movers
🎨 Google Deploys Infrastructure-Level Image AI Nano Banana Pro represents strategic positioning rather than feature competition. When you control billions of users across Chrome, Android, Gmail, and Workspace, embedding lightweight generation everywhere beats having the flashiest standalone product. The middleware businesses (Canva, Midjourney) face native competition from platforms users already inhabit daily.
🎤 Voice AI Budgets Increase Despite Satisfaction Gap 97% adoption with 84% budget increases confirms production viability, but 21% "very satisfied" rating reveals implementation challenges. The opportunity sits in the gap between universal deployment and low satisfaction. Companies solving latency, accuracy, tooling, and context handling capture enterprise spend previously locked in legacy IVR systems.
📚 Proficiency Frameworks Challenge Credential Obsession Research into AI proficiency levels shows most AI entrepreneurs chase advanced research-level expertise when revenue requires solid user-level capability. The pattern mirrors traditional education where professors with maximum knowledge often earn less than practitioners with minimum viable skills. Permission to monetize current capabilities beats preparation for imaginary standards.
🔧 Problem-Solving Prompts Formalize Decision Frameworks ChatGPT Central's systematic approach to breaking down complex situations reveals how AI amplifies structured thinking. The value ain't answering questions - it's forcing users to articulate root causes, constraints, trade-offs, and success criteria before jumping to solutions. Better inputs generate better outputs regardless of model capability.
💻 Autonomous Coding Handles Multi-Day Tasks OpenAI's Codex Max compresses session history automatically, enabling refactoring and debugging cycles that previously required constant human oversight. The compaction technology addresses the core limitation of context windows - running out of memory during extended work. When AI can maintain coherent work across days, developer productivity shifts dramatically.
📚 This Week's Curated Reading
Based on key developments from this week's AI news:
• Deployment Beats Excellence: Google's Nano Banana Pro strategy shows infrastructure advantage trumps feature superiority. When you control the platforms where people already work, embedding "good enough" technology everywhere outperforms standalone "best in class" tools requiring workflow disruption. The lesson: distribution moats matter more than capability moats.
• Satisfaction Lags Adoption: Voice AI hitting 97% adoption with only 21% high satisfaction reveals the difference between testing technology and trusting it. Universal deployment doesn't equal successful implementation. The companies solving latency, accuracy, and integration basics will capture dissatisfied enterprise budgets abandoning legacy systems.
• Proficiency Mismatches Revenue Requirements: Most AI businesses need basic to advanced user capabilities while founders chase research-level credentials. The smartest people get in their own way through imposter syndrome and preparation paralysis. Less knowledgeable people proceed unburdened by imaginary standards. Matching required proficiency to specific business models eliminates wasted learning.
• Structured Thinking Amplifies AI Value: Problem-solving frameworks work because they force articulation of constraints, trade-offs, and success criteria before solution generation. The value ain't better answers - it's better questions. When you clarify what you know, don't know, and what must be true, AI helps synthesize rather than guess.
• Context Compression Enables Autonomy: Codex Max's session history compaction solves the fundamental limitation of AI coding assistants. When systems can maintain coherent work across days without losing thread, the bottleneck shifts from "how much context fits" to "how well does it reason." Multi-day autonomy changes what's delegatable to AI.
🎯 Action Items for This Week
For Corporate Teams:
Audit your voice interaction volume (customer service, order capture, appointments) and calculate cost per interaction with current systems
If running AI workshops or training, honestly assess your skill level using a proficiency framework - basic user skills are monetizable, stop waiting for advanced expertise
Test Nano Banana Pro in Google Workspace for one visual task (infographics, mockups, presentations) and measure time savings versus current process
For Small Businesses:
Identify one repetitive phone-based task that follows predictable patterns and research voice AI providers offering that specific capability
Calculate whether your team has solid user-level proficiency (if yes, you can monetize through workshops, consulting, or automation services)
Use ChatGPT problem-solving framework for one stuck decision requiring trade-off analysis
For Entrepreneurs:
If building AI businesses, assess whether you're chasing credentials (research-level expertise) when your business model requires practical capabilities (user to builder level)
Examine one high-friction workflow and determine if lightweight AI (like Nano Banana Pro) embedded in existing tools beats standalone applications
Pilot voice AI for one customer-facing task if handling 500+ monthly calls - measure completion rates and satisfaction versus baseline
🔮 Looking Ahead
Next week's AI DUNN Right Weekly will cover:
How AI middleware businesses are responding to margin compression from model providers
Real-world ROI measurement frameworks for AI tool investments beyond hype metrics
The shift from "AI can do this" to "AI profitably does this at scale"
Why authentication and data security suddenly matter when embedding AI everywhere
Have questions or topic requests? Reply to this email. I read every message and use your feedback to shape future issues.
That's a wrap for Issue #9!
This week proved distribution advantage beats feature superiority, satisfaction gaps create opportunities despite universal adoption, and proficiency mismatches keep brilliant people overlearning while less knowledgeable competitors earn.
The businesses understanding where they actually need to be on the capability ladder rather than chasing imaginary standards will build faster, ship sooner, and monetize earlier. Everyone else prepares for tests that don't exist while opportunities pass by.
Stay innovative, Jackie @ AI DUNN Right
P.S. - That voice AI satisfaction gap? If 97% of companies deploy it but only 21% are happy, someone's about to make serious money solving latency, accuracy, and integration basics. The opportunity ain't building fancier models. It's making existing technology actually work in production. Sometimes the biggest innovations are just doing the boring fundamentals properly.
Comments