Practical AI: Episode 24
The AI Predictions Nobody Will Make, And Why They’ll Matter Most in 2026
About This Show
Practical AI is a weekly live show (Fridays 11am CT) hosted by Olga Pechnenko and Chris Pearson that cuts through AI hype to deliver news, trends, and hands-on tips for builders and founders. Unlike technical AI podcasts, Practical AI focuses on business applications and ROI—what actually works, what’s hype, and what you can implement Monday morning.
What You’ll Gain
- Learn why the 2025 AI hype cycle collapsed and what actually happened: DeepSeek’s $6M model knocked $600B off Nvidia’s valuation, agents became supervised interns instead of autonomous colleagues, and 95% of enterprise AI projects delivered zero ROI. These failures reveal what will actually work in 2026.
- Understand the four prediction tiers for 2026 ranging from safe consensus bets (agentic tools, edge AI, EU enforcement) to the unspoken risks no AI model wants to discuss: the web slop apocalypse, junior talent extinction, deepfake elections, and the breakdown of educational verification.
- Discover the competitive dynamics reshaping the AI landscape including Anthropic blocking OpenAI and XAI from using Claude (revealing who the real engineering players are), Apple’s defensive Gemini partnership, and why Meta killed the metaverse to fund nuclear-powered AI data centers.
- See a live demonstration of prototype-to-production workflow where a Claude-generated website design becomes a fully functional, editable PageMotor site in under four hours, replacing thousands of dollars of agency work and eliminating the micro-decision fatigue that kills most web projects.
- Gain a framework for surviving the 2026 ROI reckoning including what data to protect from AI training, how to install a BS detector against vendor hype, and why AGI will emerge from networked systems rather than any single company’s model.
Biggest Takeaway to Implement
Stop treating AI adoption as software installation. The companies winning in 2026 won’t be fastest adopters but those who protect their proprietary data, verify vendor claims with actual customer results over months, and build the connective tissue between AI prototypes and production systems. Your edge isn’t the AI itself—it’s shipping the prototype to something real… fast.
Free PageMotor and Practical AI Updates:
Frequently Asked Questions 🤔
What percentage of enterprise AI projects delivered ROI in 2025?
According to MIT research, 95% of enterprise AI projects delivered zero ROI in 2025. Companies treated AI like software installation rather than workflow overhaul, leading to mass failure of pilot programs.
How did DeepSeek impact Nvidia’s stock price?
DeepSeek released a model trained for only $6 million that knocked $600 billion off Nvidia’s market cap in a single day, proving that math optimization can beat massive compute spending as a competitive moat.
What percentage of companies trust AI agents to make decisions?
Only 10% of companies trust AI agents with actual decisions. While 23% are scaling agents, they function as supervised interns rather than autonomous colleagues.
What is the AI ROI reckoning predicted for 2026?
The ROI reckoning is the 2026 moment when CFOs demand production-grade AI utility or kill projects entirely. With 95% of AI projects showing zero ROI, the pilot phase is dead.
What is the slop apocalypse prediction?
By late 2026, an estimated 60% of web content will be AI-generated garbage. Search will break under synthetic content, and verification becomes the only valuable currency online.
Why did Anthropic block OpenAI and XAI from using Claude?
Anthropic blocked XAI via Cursor after discovering engineers were using Claude to build a competing coding suite. They previously blocked OpenAI when engineers were caught using Claude to help build GPT-5—revealing that serious AI engineering increasingly requires Claude.
Practical AI: The AI Predictions Nobody Will Make, And Why They’ll Matter Most in 2026
00:00 Live Intro and The Weird Plan: Four AIs vs Reality
Chris and Olga set up an experiment: interrogate four leading AI models (GPT, Claude, Gemini, Grok) about their 2025 predictions, score how badly they whiffed, then extract their 2026 forecasts. The twist? They also pushed the models to reveal predictions they’re incentivized to hide. The result is a four-tier framework separating consensus safe bets from the genuinely uncomfortable forecasts that no AI company wants to discuss publicly.
01:40 Flash News Pt 1: CES Robot Takeover and Nvidia’s Open-Source Power Move
CES 2025 was dominated by robotics. Nvidia open-sourced their entire robotics AI stack on Hugging Face, letting smaller companies build without billion-dollar infrastructure. Boston Dynamics’ Atlas heads to Hyundai factories in 2026. Notably absent: Optimus had no booth among 4,000 exhibitors. Robotics investment has surged in the past two months, confirming real money moving toward physical AI.
07:30 Flash News Pt 2: Meta Kills Metaverse, Goes Nuclear and Anthropic’s Breakup Drama
Meta killed its metaverse and cut 10% of Reality Labs to fund nuclear-powered AI data centers. The bigger story: Anthropic blocked XAI’s Claude access via Cursor after discovering XAI engineers were using Claude to build a competing coding suite. They previously blocked OpenAI in August when engineers were caught using Claude to help build GPT-5. Companies NOT blocked? Gemini and Meta.
The revealed preference that serious AI engineering requires Claude. When competitors like OpenAI and XAI secretly rely on Claude to build competing products, it signals Claude’s unique capability that even well-funded AI labs can’t replicate internally.
13:20 Flash News Pt 3: Health AI Battle, Claude Cowork Flex and Apple-Gemini Handshake
OpenAI released ChatGPT Health targeting consumers with wearable integrations. Anthropic countered with HIPAA-compliant hospital integrations for enterprise. Same week, different strategies: OpenAI chasing consumers, Anthropic solving infrastructure problems. Claude’s Cowork desktop agent launched with underwhelming use cases, but it’s a muscle flex: Anthropic built it in a week using only Claude Code. The Apple-Gemini partnership came without any demo, suggesting Apple has nothing working yet.
19:10 2025 Predictions Autopsy: What Blew Up vs What Crashed
A model trained for $6 million knocked $600 billion off Nvidia’s market cap, proving math optimization beats cash as a competitive moat in AI development.
MIT research shows 95% of enterprise AI projects delivered zero ROI. Companies treated AI like software installation rather than workflow overhaul.
Only 23% of companies are scaling AI agents, and just 10% trust them with actual decisions. Agents became supervised interns, not autonomous colleagues.
The models predicted US dominance was unassailable and billions in compute would be the only moat. Then DeepSeek released a model trained for $6 million that knocked $600 billion off Nvidia’s market cap. Math optimization beat cash. The prediction that agents would join the workforce as full employees? Reality: supervised interns.
The 78% adoption prediction hit hardest. Actual value realization: 5%. Companies treated AI like software installation rather than workflow overhaul. The shift from “more data” to “better reasoning” became necessary after models ran out of training text. Shadow AI usage exploded as employees hide their AI use. The junior talent crisis emerged: if AI handles first drafts and research, who becomes the next senior VP?
33:30 2026 Tier 1: The Safe Stuff Everyone Agrees On
All four models agreed: continued shift to agentic AI, efficiency wars driving compute costs down as wrapper companies seek alternatives to expensive API calls, edge AI boom as privacy and latency push processing onto devices, and EU AI Act enforcement in August 2026 with someone getting fined as an example.
Companies like Cursor spend 60-70% of revenue on API costs, driving a survival-level run toward open-source models.
The efficiency pressure is survival-driven. Cursor and similar companies spend 60-70% of revenue on API costs. They’re running to open-source models out of necessity. Edge computing wins on basic engineering: local reasoning means faster delivery.
42:40 Tier 2: Realism Setting In, Cracks Showing
The 2026 moment when CFOs demand production-grade AI utility or kill projects entirely. With MIT research showing 95% of enterprise AI projects delivered zero ROI, the pilot phase is dead—companies must prove real business value or lose funding.
The ROI reckoning arrives. That MIT survey showing 95% of AI projects have zero ROI means CFOs will demand production-grade utility or kill projects. The pilot phase is dead. Export controls failed; China is building competitive models on domestic silicon. The world fractures into incompatible AI ecosystems: US closed-source, China open-source, and the open-source community potentially winning long-term through sheer node count and iteration speed.
45:20 Tier 3: Wild Cards and Rogue Bets
Gemini predicted the death of the app ecosystem: generative UI creates the interface you need, when you need it, then deletes it. Static apps become obsolete. UX designers face the same disruption programmers are experiencing. ChatGPT predicted outcome-as-a-service replacing SaaS: you pay for results, not software seats. If the AI fails, the vendor pays. Risk shifts from buyer to seller.
Grok predicted robots entering kitchens and warehouses at scale. The physics wall looms: infrastructure bottlenecks (power, cooling) will dictate where AI can physically be built before we run out of algorithmic ideas.
49:00 Tier 4: The Predictions AIs Won’t Touch (The Dirty Ones)
The prediction that by late 2026, 60% of web content will be AI-generated garbage. Search engines break under the weight of synthetic content. Verification and trust become the only valuable currency online.
By late 2026, an estimated 60% of web content will be AI-generated slop, breaking traditional search and making verification the only valuable currency.
The models ignored predictions that make them look bad. The slop apocalypse: by late 2026, 60% of web content will be AI-generated garbage. Search breaks. Verification becomes the only currency. The junior talent crisis accelerates as entry-level positions evaporate. Corporate ladders lose their bottom rungs.
The 2026 midterms become the first AI-native election with deepfakes everywhere. Education fails: schools cannot assess learning when AI completes any assignment. A 16-year-old high schooler reports widespread AI cheating with no detection. The loneliness epidemic worsens as AI validation feels easier than human connection.
58:00 Okay, So What Do We Actually Do Monday Morning?
A decision-making framework for evaluating AI vendors: (1) Demand proof over vibes—require actual ROI data from multiple customers over months, not weeks. (2) If a pitch highlights benefits but skips risks or costs, they’re lying. (3) FOMO is a sales tactic, not a business strategy.
Framework for protecting your position. Know your bottleneck: Enterprise buyers demand ROI proof and compliance. Consumer products face energy costs and copyright liability. Protect your data: Before any prompt, ask if leaking to your biggest competitor would hurt. Your prompts and results become training data. Uncheck training consent everywhere.
Install your BS detector: FOMO is a sales tactic. Demand proof over vibes. Require actual data from multiple customers over months. Every pitch highlights benefits; if they skip risks or costs, they’re lying. The 2026 winners won’t be fastest adopters but those who got constraints early, protected data, and filtered noise.
1:08:40 Quick AGI Rant: It’s Not a Single Model
The argument that AGI won’t emerge from any single company’s model. Real AGI emerges from networks: reasoning models generate data, that data flows to industry-specific systems, those systems act and generate new data flowing back. Sentience is the network effect, not any single node’s power.
Chris argues AGI won’t emerge from any single company’s model. Consuming all data doesn’t create creativity or associative thinking; it enables keyword matching across a data set. Real AGI emerges from networks: reasoning models generate data, that data flows to industry-specific systems (like PageMotor for websites), those systems act and generate new data flowing back to reasoning models or other nodes.
Think of how brains work: new connections form pathways, pathways enable skills, cross-talk creates capability. Sentience is the network effect, not any single node’s power. AGI is something we all build together. OpenAI won’t accomplish it alone.
1:14:15 This Week’s Funding Firehose
AI companies captured 70% of all venture capital this week: $6.1 billion across 94 companies with 18 mega-deals. US led with over $4 billion.
AI companies captured 70% of all venture capital this week: $6.1 billion across 94 companies with 18 mega-deals. US led with over $4 billion. Top funded: Skilled AI teaching robots to work in warehouses, etched.ai building specialized chips for 10x faster inference (an edge play), Parloa replacing customer service with AI agents, Rain moving payments via blockchain, and Maul creating Sharia-compliant Islamic banking (no interest charges). Robotics and infrastructure dominate, confirming the physical AI trend.
1:22:50 Live Demo: Claude Prototype to Shipped PageMotor Site
Claude produced a complete website prototype in 90 minutes. Chris translated it to PageMotor in 4 hours. Total: 5.5 hours replacing multi-thousand dollar agency work.
Olga’s recruiting company had a basic brochure website. She dreaded redesigning it because of copy-writing fatigue and developer dependency. Using ChatGPT to generate a comprehensive prompt, she fed it to Claude, which produced a complete website prototype in 90 minutes. The prototype included hero section, about, methodology, interactive elements, testimonials, and contact forms with AI features.
Claude made thousands of micro-decisions about fonts, spacing, and color that would consume hours of human attention. Chris translated the clean HTML into PageMotor’s block system in four hours: production site with full CMS, user management, and lead generation. Multi-thousand dollar agency work replaced by 5.5 hours combined. Unlike Gemini (which remixes the same elements), Claude approaches each design fresh. It has genuine design reasoning.
1:48:00 Wrap and One Big 2026 Lever
The real 2026 edge isn’t what AI models can do. It’s how they share data with systems that do things you need. The tools connecting dots matter more than the tools doing inference. Olga’s website works now because Claude’s prototype connected to PageMotor’s production system. The prototype became real in hours, not months. That’s the lever: shipping prototypes fast, not generating them. Anyone can generate. The winners ship.