NotebookLM: Learning on Steroids

10 AI Trends Business Leaders Need to Know Right Now

Nobody Told Me About IT — EP005: Tad Doyle & Nabil Gharbieh 0:05 — THE HOOK Round-robin: ten trending AI topics, the honest read on each, what to do. No panic. No hype. 0:51 — TREND 1: THE EXPLORER TRAP 51% of small business owners are AI explorers. Only 8% have advanced adoption. Two years of exploration without commitment is diffusion. Pick one workflow, commit for a quarter, measure it. 1:37 — TREND 2: AGENTS ARE REAL, NOT MAGIC MIT Sloan: agentic AI hits Gartner's trough of disillusionment in 2026. Gartner: 40% of agent projects canceled by end of 2027. Agents hallucinate, need humans in the loop, and are hard to govern. Specific tasks, specific guardrails. Fully autonomous is a red flag. 2:35 — TREND 3: THE SKILLS GAP IS REAL 63% of employers cite skills as the primary AI barrier. 20% of small businesses feel confident adopting AI, vs. 82% of mid-sized firms. Only 12% of small businesses invest in training, despite 29% citing it as the biggest obstacle. Zero training budget means you have a software budget. 3:21 — TREND 4: THE WORKFORCE ISN'T ONE BLOCK 31% of employees push back on AI strategy. 45% worry too much AI will hurt the company's reputation. 30% of SMB workers use AI daily. Ask your team three things: what excites you, what worries you, what's missing. 4:05 — TREND 5: THE .COM ANALOGY IS INCOMPLETE MIT Sloan compares the AI moment to .com: sky-high valuations, growth over profits. Real businesses are also seeing real ROI. Both true. Some tools won't exist in three years. Don't bet on one vendor. Add a vendor audit annually. 4:46 — TREND 6: GOVERNANCE SEPARATES LEADERS FROM THE REST 68% of high-ROI AI organizations have mature governance. Among lower performers, 32%. 56% of CEOs are delaying major AI investments due to governance uncertainty. If you can't answer "Who owns AI here?" fix that first. Name the owner this week. 5:24 — TREND 7: TOP-DOWN PICKS OUTCOMES PwC: crowdsourcing AI from departments rarely produces meaningful outcomes. Senior leadership picks two or three strategic workflows and commits resources. Bottom-up gets adoption. Top-down gets outcomes. 6:01 — TREND 8: DATA IS THE REAL BOTTLENECK SAP and Snowflake advisors agree: data is the bottleneck to trusted AI. Healthcare charts average 46,000 words. Most business knowledge is scattered, duplicated, contradictory. Top adopters fix data first. Identify the one data set that, if cleaned, would unlock the most value. Beats three new tools. 6:45 — TREND 9: THE CATCH-UP IS REAL SBA: small businesses closing the AI gap with large enterprises faster than in previous cycles. Within SMB, the spread between adopters and non-adopters is widening. 96% plan to adopt, 8% have advanced adoption. Benchmark against your five closest competitors, not Microsoft. 7:30 — TREND 10: MARKET TIMING Microsoft Q1 2026: AI usage growing 1.5 points per quarter globally. US at 17.8%. UAE at 70.1%. Gartner: 40% of enterprises will use task-specific AI agents by end of 2026, up from under 5% in 2025. You're not late. You're not early. You're right on time. KEY TAKEAWAY "Every trend has an opportunity and a real friction. AI in 2026 isn't a crisis and isn't a paradise. It rewards leaders who hold both, the upside and the work." LINKS AND RESOURCES McKinsey — The State of AI: shadow AI and personal-account usage. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai Microsoft + LinkedIn — Work Trend Index: BYO-AI and talent migration. https://www.microsoft.com/en-us/worklab/work-trend-index Gartner — AI governance and shadow AI risk. https://www.gartner.com/en/information-technology/insights/artificial-intelligence IBM IBV — CEO Decision-Making in the Age of AI: 56% delaying due to governance. https://www.ibm.com/thought-leadership/institute-business-value All episodes — https://nobodytoldmeaboutit.com

Finding Your Company's AI Middle Path

Nobody Told Me About IT — EP004: Finding Your Middle Path in AI Hosts: Tad Doyle & Nabil Gharbieh  ·  Published: May 19, 2026  ·  Runtime: 8:02 0:05 — THE TWO EXTREMES Most AI conversations inside organizations are happening at one of two poles: total lockdown or no governance at all. Tad and Nabil open by naming the tension directly — and then spend the episode inhabiting each extreme as composite client characters, asking each other the questions advisors actually ask. “The fear is legitimate. You’ve got data leakage, you’ve got IP exposure, you’ve got compliance exposure. The response is an actual risk that can be a lot worse.” — Nabil Gharbieh 1:26 — THE LOCKDOWN CLIENT Tad plays a composite mid-market CEO who blocked ChatGPT, Copilot, and Claude at the firewall and told staff that AI use is grounds for termination. Nabil asks three questions: What happens when employees still need to get work done? What does this do to competitive position? And was the fear actually justified? The answers: shadow AI moves in immediately, competitors move faster, junior staff start leaving, and the lockdown just pushed the risk somewhere no one could see it. “You’re gonna get shadow AI. You’re gonna get people who use their personal cell phones, their personal accounts, their personal emails to forward to themselves — and now they’re going to work off their personal laptops. So you’ll have zero visibility, zero control.” — Nabil Gharbieh 3:00 — THE WILD WEST CLIENT Nabil plays a composite 75-person CEO who told staff to “just figure it out.” Everyone expenses their own tools with no contracts, no data protection agreements, no inventory. Tad asks three questions: Who owns the data those tools are seeing? What happens when a tool has a breach or shuts down? And what do you tell the board when they ask for the AI strategy? The answers: nobody knows, business continuity is a gap nobody has planned for, and “we’re embracing AI” is not a strategy. “There’s a lot of activity. There’s absolutely no strategy. There’s no value assessments. There’s no risk assessments. It’s just kind of ‘we’re embracing AI.’ There’s no strategy with this.” — Nabil Gharbieh 5:21 — THE MIDDLE PATH The hosts flip from adversarial to collaborative. Nabil’s two moves for the lockdown client: replace the ban with a one-page acceptable use policy and publish an approved tool list. Tad’s two moves for the all-in client: do a visibility inventory via Slack or email (no judgment), and define three data categories that can never go into an AI tool — customer PII, financial data, contracts. KEY TAKEAWAY “Neither extreme works. The conversation your organization needs to be having isn’t ‘Should we use AI?’ It’s ‘Where on the spectrum are we right now, and what’s our next move toward the middle?’”  — Tad Doyle LINKS AND RESOURCES Microsoft Copilot — https://www.microsoft.com/microsoft-365/copilot/microsoft-365-copilot [https://www.microsoft.com/microsoft-365/copilot/microsoft-365-copilot] ChatGPT Team — https://openai.com/chatgpt/team [https://openai.com/chatgpt/team] Claude for Work — https://www.anthropic.com/claude/work [https://www.anthropic.com/claude/work] All episodes — https://nobodytoldmeaboutit.com [https://nobodytoldmeaboutit.com]

NotebookLM: Learning on Steroids

Nabil Gharbieh has a certification problem, the same one most IT leaders have. Every year there is a new credential, a new framework, a new exam. In this episode he walks through the tool he uses to cut through that burden, NotebookLM, and shows exactly how he gets from zero knowledge to exam-confident without burning weekends. 0:00 Introduction and the certification treadmillNabil opens with a concrete result: when he took the ISO 42001 exam, it was the first time he felt genuinely confident going in rather than hoping for the best. 0:25 What NotebookLM isNotebookLM is a Google product that lets you upload your own sources, PDFs, URLs, videos, notes, and then interact with an AI constrained to only that material. Three panels: sources on the left, chat in the middle, Studio tools on the right. He is using the paid version, which offers more Studio options than the free tier. 1:10 Loading sources and setting scopeNabil demonstrates with ICS-100, incident command training for his volunteer fire station. He loads four course documents plus his own handwritten notes. The key design choice: by uploading only exam-relevant material, he constrains the AI to a closed knowledge set. 2:15 Why prompting changes everythingNabil writes a custom prompt for each output, for example: "I am taking the ICS-100 exam. Help me understand the knowledge in a way that you are certain I will be 90% ready to pass." He also uses AI to sharpen those prompts before pasting them in. 3:30 Audio overviewsThe audio overview feature converts source material into a two-host podcast format. Nabil generates these and listens while walking his dog. The paid tier adds an interactive mode where he can interrupt the AI conversation to ask a follow-up question. 4:53 How to sequence the toolsStart with a video course or lecture first for foundational orientation. Then bring the material into NotebookLM for the real learning work. Starting with NotebookLM before any foundational exposure produces weaker retention. 5:16 Mind mapsThe mind map gives an expandable visual of the entire subject area. After two minutes on the ICS-100 map, Tad, who had no prior exposure to incident command, could already describe the basic structure."I bet you, if you look at this long enough, you'll be able to break it down for me, and you came into this conversation not knowing what incident command was." 6:15 Flashcards and quizzesFlashcards include a green check (I know this) and a red X (ask me again) mechanic. For quizzes, Nabil always starts on medium difficulty, then creates a harder quiz once he feels ready. When he gets a question wrong, he clicks "explain" and the chat panel shows exactly which source the answer came from. 7:14 Data tables and infographicsThe data table organizes material by topic, objectives, best practices, and source document. The infographic generates a visual summary. For visual learners, this format alone can do significant conceptual work. 9:06 Regenerating outputsEvery Studio output can be deleted and regenerated with a new prompt. On the free tier, regenerations are limited to roughly three per item per day. On paid, no cap. 9:45 Sharing notebooksNotebooks can be shared directly. After Nabil passed the ISO 42001 exam, he shared his notebook with Tad so all the source loading, prompting, and output generation was already done. Key takeaway"This was the first time I ever took an exam that I felt like I actually knew the content really well. I didn't feel nervous. I just felt really confident. And it was this tool that I used." Links and resources Show notes for this episode: https://www.google.com/url?q=https://docs.google.com/document/d/1WEmp02WuFsaxJ3wzDP4uLfiqNNLvAKEwmwn9FQfVXzY/edit?usp%3Dsharing&sa=D&source=editors&ust=1778560746111589&usg=AOvVaw02nkHPG4bm9KlP22dBgqOS [https://www.google.com/url?q=https://docs.google.com/document/d/1WEmp02WuFsaxJ3wzDP4uLfiqNNLvAKEwmwn9FQfVXzY/edit?usp%3Dsharing&sa=D&source=editors&ust=1778560746111589&usg=AOvVaw02nkHPG4bm9KlP22dBgqOS] Google One 4-Month Free Trial: https://one.google.com/referral/redeem/R43DP4K4?g1_landing_page=5 [https://one.google.com/referral/redeem/R43DP4K4?g1_landing_page=5] Nobody Told Me About IT: https://nobodytoldmeaboutit.com [https://nobodytoldmeaboutit.com]

AI for Deep Research: How We Actually Use It

Most mid-market organizations have AI in use, whether approved, shadow, or embedded in vendor tools. Almost none have a governance framework for it. In Episode 02, co-host Nabil Gharbieh and Tad Doyle demonstrate live, on screen, how an experienced IT advisor actually uses Claude to research AI governance platforms. This is a structured methodology, not a product endorsement. Tad opens with a real-world scenario: a mid-market financial services organization has approved AI use but has no governance framework. Employees are using Copilot, Claude, ChatGPT, and other tools across platforms the organization has never reviewed. In regulated industries. The research question: what platforms exist to help solve this? Before the demonstration begins, Tad shares a practical tip: ask the model which version to use. Claude will tell you that Opus is overkill for most research tasks. Using the right model saves tokens and cost. Claude returns a structured framework covering research approach, timeline, tool categories, and evaluation criteria. Tad's first move is a deliberate sanity check, scrolling through the output to confirm it aligns with what he knows about NIST frameworks before going further. The second prompt adds real context: the organization runs Microsoft 365, uses Copilot as its primary AI tool, and carries SEC and FINRA obligations. Under 500 employees. Shadow IT is a known concern. Nabil surfaces a real concern. Claude has a tendency to validate rather than challenge, using phrases like "perfect" and "great research" regularly. Tad's answer: maintain your own skepticism. Ask the model what it might have missed, or how you could phrase the question better. The positivity in the interface is not the same as accuracy. The resulting report runs 18 pages after Tad requests a completeness check. It includes an executive summary, tool categories mapped to governance needs, a quick reference guide with vendor pricing, maturity assessments, detailed vendor profiles with advantages and risks, a scored comparison, and a summary recommendation with two platforms for Phase 1 evaluation. One honest caveat on pricing: enterprise software pricing from AI research is rough. Most platforms are quote-based. The research narrows the field. A real pricing conversation follows separately. The report surfaces established platforms like Microsoft Purview alongside less familiar names. Tad's approach: the presence of independently validated platforms on the same list gives you confidence Claude is comparing real tools. For the unknowns, due diligence follows, including account rep conversations and Gartner or Forrester reports. Those reports can be uploaded back into Claude to refine the document further, with citations. Nabil's closing question: you went from zero knowledge to a vendor list. You wouldn't actually recommend based on this alone? Tad's answer: correct. The output is a starting map, not a destination. When presenting to clients, he is explicit about his qualifications and reservations, especially on pricing, and brings domain knowledge to reality-check the recommendations. The third prompt stress-tests the output: which platforms are mature and field-proven at mid-market scale? Which are overhyped or early-stage? What concerns would you have about recommending each? The result confirms which platforms are well-regarded and flags the less-proven tools for additional review before moving forward. "AI doesn't replace the advisor. It makes the advisor faster, more efficient, and more productive. The value isn't in the tool. It's in knowing what questions to ask, and knowing how to evaluate what comes back." Chapters 2:10 The Starting Point 3:32 Reading the Output 6:25 How Do You Know It's Not Just Telling You What You Want to Hear? 7:13 Walking Through the Report 14:44 What to Do with Vendors You've Never Heard Of 17:20 Trust But Verify 19:20 Stress-Testing the Output Show Notes on www.nobodytoldmeaboutit.com

Anthropic Project Glasswing: When AI Finds the Bugs Before the Hackers Do

Anthropic's Mythos Found A 27-Year-Old Bug. Here's What IT Leaders Should Do Monday.