The AI Frontier Playbook

Your AI agent is quietly building a profile of you. The real question is what happens to that profile, and who would even notice if something went wrong.In this episode of The AI Frontier Playbook, I sit down with Alyssa Ofstein, Senior Technical Program Manager in Microsoft’s AI Safety and Security team, where she has trained more than 17,000 people on how to build AI agents safely. Alyssa sits right between the red team, the people trying to break AI, and the engineers who build it. Her job is to turn what goes wrong into guidance people can actually follow.Alyssa explains the risk most people never think about: agent memory. Agents don’t just store what you tell them, they make inferences about you, and those inferences, true or not, can spread across every system the agent touches. She is candid about why companies say they want responsible AI but reward speed, why trust and safety teams are still understaffed, and why she agrees that frontier labs should run like nuclear facilities.We also get practical: what makes AI guidance obvious, adaptable, and tangible enough that engineers follow it, how Medusa’s Memory Heist, the open source game she built with Joris de Gruyter, teaches AI security through play instead of fear, the first three things a company should put in place, and the three questions Alyssa asks before she hands any task to AI. She closes with an emotional answer about what a random Tuesday in 2036 could look like if we get this right.You'll learn:How Alyssa turns red team findings and incident reports into guidance engineers followWhy agent memory is the AI risk nobody thinks aboutHow agent inferences about you can spread across tools, networks, and other people’s agentsWhy companies say “responsible AI” but reward speedWhy frontier labs should run like nuclear facilitiesWhat makes AI guidance obvious, adaptable, and tangibleHow Medusa’s Memory Heist teaches AI security through play, not fearThe first three things to put in place: data hygiene, a reward system, and stakeholder relationshipsWhy you can’t rely on your AI provider aloneWhat Alyssa will never hand over to an AI agentHer three-question framework before giving any task to AIWho blows the whistle when every worker is an agent?Why there is a permanence to what we tell an LLMWhat happens when an agent’s goal isn’t your goalTen years out: what AI could mean for a random Tuesday in 2036Whether you’re building AI agents, rolling out AI across your company, or just using AI every day at work, this episode gives you a clear way to think about the risks, and how to stay in control.Resources & LinksAlyssa LinkedIn: https://www.linkedin.com/in/alyssaofstein/Medusa’s Memory Heist (free, open source): https://microsoft.github.io/medusa/site/Our Sponsor Wepoint: https://www.wepoint.com/en📬 Enjoyed this episode?Subscribe to the AI Frontier Playbook: https://samuelboulangerai.substack.com/Connect with me on LinkedIn: https://www.linkedin.com/in/samuelboulanger/Visit the website: https://frontierplaybook.aiNewsletter: https://theaifrontierplaybook.substack.com/

Most of us spend the day moving information between systems. AI can make that faster, but the real win is not needing to do it at all.In this episode of The AI Frontier Playbook, I sit down with Bryan Goode, Corporate Vice President for Business Applications and Agents at Microsoft. Bryan was part of the team that launched Microsoft Teams, and today he leads Dynamics 365 and the AI agents being built around it. His argument is simple: making each person faster is the small win. Rebuilding the process is the real one.Bryan explains why employees have become the "human middleware" between disconnected systems, and why the toggle tax of switching from app to app can cost the average information worker around four hours a week. He shares how L&G and Lifetime rethought their processes with AI, and why he believes AI makes business applications more important than ever, not less, as the value moves from the user interface down to the data, the intelligence layer, and governance.We also get practical: when to use a pre-built Dynamics 365 agent versus building your own in Copilot Studio, how multi-agent orchestration works, where a human still needs to sign off, and how Agent 365 gives IT visibility across every agent in the organization. Bryan closes by sharing the agents he uses every day, what Microsoft Scout actually is, and why he believes AI will deepen the human connection at work.You'll learn:Why making every individual faster with AI doesn't fix the underlying processWhat the "toggle tax" is and how much it costs the average information workerHow L&G and Lifetime used AI to rethink customer service and order fulfillmentWhy the "SaaS apocalypse" gets it backwards, and how business applications are changingWhen to use pre-built Dynamics 365 agents and when to build your own in Copilot StudioHow multi-agent orchestration connects agents into one business processHow to decide where human approval is required as agents start taking actionWhy the best results combine bottom-up experimentation with top-down process changeHow to choose your first process, the right sponsor, and the metric to measure relentlesslyWhy the business, not IT, should own the outcome of an AI processThe agents Bryan uses every day, and what Microsoft Scout isWhy Bryan believes AI will make work more human, not lessWhether you're leading an AI transformation, running a business function, or trying to figure out where agents fit in your organization, this episode gives you a practical roadmap for rebuilding the way work gets done.Resources & Links Bryan LinkedIn: https://www.linkedin.com/in/bryangoode/ Our Sponsor Wepoint: https://www.wepoint.com/en📬 Enjoyed this episode? Subscribe to the AI Frontier Playbook: https://samuelboulangerai.substack.com/ Connect with me on LinkedIn: https://www.linkedin.com/in/samuelboulanger/ Visit the website: https://frontierplaybook.ai Newsletter: https://theaifrontierplaybook.substack.com/

Your AI agent may not need a sophisticated exploit to be hacked. Sometimes, one emoji is enough.In this episode of The AI Frontier Playbook, I sit down with Richard Diver, Director of Story Design at Microsoft, author of Guardians of AI and Unlearn Security, and a security veteran with 30 years of experience. We explore why the biggest AI security problems aren't really technology problems — they're thinking problems.Richard breaks down how threat modeling changes when AI can act autonomously, why every AI system should be evaluated through three simple questions — asset, access, and actor — and why you should think through the "what if?" before an agent ever reaches production. We get into prompt injection, indirect attacks hidden inside documents and images, AI memory as a new attack surface, and why giving one AI access to everything may be one of the biggest mistakes organizations can make.We also explore Richard's "green team" concept — why security can't remain the responsibility of a small security department, and why the people actually using AI every day need to be part of the security process. He shares six things companies should check before turning on their first AI agent, how to limit an agent's blast radius, and why the worst attacks aren't the ones that cause immediate damage — they're the ones nobody notices for years.You'll learn: Why major security failures are usually caused by how we think, not a lack of security tools Richard's Asset, Access, Actor framework for threat modeling AI systems How a single emoji or special character can become an attack vector against AI Why prompt injection can hide inside documents, emails, images, and other content Why giving one AI access to everything creates unnecessary risk Why Richard would rather have 1,000 agents doing one job each than one AI with access to everything Why "read-only" access isn't necessarily safe when an AI can read and share information How AI memory creates a new long-term attack surface What the "green team" is and why security needs to involve everyone The six things to check before turning on your first AI agent Why reducing the blast radius may be more realistic than trying to prevent every attack Why the most dangerous AI attacks are the ones that go undetectedWhether you're building AI agents, deploying AI across an organization, or simply using AI tools in your everyday work, this episode gives you a practical way to think about security before something goes wrong.Resources & Links Richard on LinkedIn: linkedin.com/in/rdiver Book: Guardians of AI Book: Unlearn Security Our sponsor Visionet: visionet.com📬 Enjoyed this episode? Connect with me on LinkedIn: linkedin.com/in/samuelboulanger Visit the website: frontierplaybook.ai Newsletter: theaifrontierplaybook.substack.com

Customers almost never ask "can you build this" anymore. What they actually want to know is "can I trust it."In this episode of The AI Frontier Playbook, I sit down with Sapna Grover, Senior Director at Microsoft leading the Cloud, Data & AI Center of Excellence. Before this, she spent close to three decades in software, as a scientist at India's Defense Research and Development Organization, in leadership roles at Pegasystems, and in engineering leadership at Microsoft across Windows and Search. Today she's helping the world's largest companies actually put AI into production.Sapna breaks down why AI flips the traditional software path: you start with the business outcome and the data, not the code, and you have to define what a "good" answer even looks like since AI rarely gives the same response twice. Using a live airline customer-support example, she walks through what AI evaluation really takes: a golden data set, an LLM as judge, and tracing every decision so AI stops being a black box.We also get into why the most dangerous AI failures are silent, quality drifting while every dashboard stays green, and the single most expensive mistake she's seen companies make: falling for the technology before the problem. She closes by sharing the AI agents she's built for her own life, including one that drafts her Friday team update and another that tracks her protein intake better than any app could.You'll learn:Why the fundamentals of building great products haven't changed, even though AI has changed the engineering problems underneath themHow AI flips the traditional software path, starting with business outcome and data instead of codeWhy "did it produce the expected answer" is the wrong question for AI, and what to ask insteadHow to build a golden data set and use an LLM as a judge to evaluate AI at scaleWhy tracing turns AI from a black box into a glass box you can actually debugWhy the most dangerous AI failures are silent, with quality and trust eroding while every system metric stays greenThe most expensive mistake companies make with AI, and why it usually isn't a technical oneWhy AI needs the same financial discipline as cloud infrastructure, and how uncontrolled cost quietly kills ROIWhether you're a product leader shipping your first AI feature or a technology leader responsible for what happens after launch, this episode gives you a practical framework for building AI your organization can actually trust.Resources & LinksGuest LinkedIn:linkedin.com/in/sapna-grover-9170032📬 Enjoyed this episode? Subscribe to the AI Frontier Playbook:https://theaifrontierplaybook.substack.com/ Connect with me on LinkedIn: https://www.linkedin.com/in/samuelboulanger/ Visit the website: https://frontierplaybook.ai#AI #FutureOfWork #Startups #MicrosoftCopilot #EnterpriseAI

AI in retail isn't the chatbot on the website — it's the frontline worker running the store floor, and that's where the real transformation is happening.In this episode of The AI Frontier Playbook, I sit down with Ghazanfar Riaz, Senior Vice President of the Global Digital Solutions Group for the Microsoft Business Unit at Visionet, who has spent 20 years shipping enterprise technology and is now deploying AI into some of the biggest retailers in North America.We're not talking about shopping assistants or chatbots — we're talking about the person restocking shelves, running the till, and checking if your size is in the back. Ghazanfar walks through what a frontline AI co-pilot actually does in practice: starting an associate's shift with a prioritized brief instead of a stack of emails, catching shelf exceptions before a customer notices a product missing, and guiding new hires through SOPs without pulling a manager off the floor. We get into why the biggest bottleneck in AI adoption isn't the technology or the data — it's people, and specifically whether frontline workers are trained and involved early enough to actually trust the tool. Ghazanfar also shares a real example of a retailer whose shelf exceptions were happening 100 times a week, taking 25 minutes to catch and fix, with only 70% resolved in the same shift — and how closing that gap became the model for a good AI use case. We close on what a retailer risks by waiting another year, and what the frontline retail job might look like a decade from now.You'll learn:Why the real AI transformation in retail is happening on the store floor, not in the customer-facing chatbotHow a co-pilot morning briefing turns an associate's shift into a prioritized to-do list instead of guessworkWhy shelf-exception detection and auto-generated replenishment tasks are closing an expensive, invisible retail gapWhy the biggest bottleneck in retail AI adoption is people, not technology or dataWhat separates a retailer that's actually ready for AI from one whose project stalls in the first few weeksHow assisted selling and cross-store inventory visibility turn a lost sale into a substitute purchaseWhy returns are one of the highest-impact, lowest-effort use cases for frontline AIWhy waiting a year to deploy AI costs a retailer more in lost organizational learning than in outdated technologyWhether you're a retail operator trying to figure out where to start with AI, or a technology leader building the next generation of frontline tools, this episode gives you a practical framework for turning daily store friction into your first AI use case.Resources & LinksGhazanfar Riaz on LinkedIn: linkedin.com/in/gsheikh-01📬 Enjoyed this episode?Subscribe to the AI Frontier Playbook: samuelboulangerai.substack.comConnect with me on LinkedIn: linkedin.com/in/samuelboulangerVisit the website: frontierplaybook.ai#AI #FutureOfWork #Startups #MicrosoftCopilot #enterpriseai

AI gets the attention, but trust gets the PO. The coolest demo in the world means nothing if an enterprise does not believe you are safe, compliant, and serious about data.In this episode of The AI Frontier Playbook, I sit down with Tom Davis, Partner at Microsoft for Startups, who founded Solair in 2011, built it on Azure, and sold it to Microsoft in 2016.Tom has spent the last several years helping thousands of founders turn AI and cloud innovation into real business outcomes, and he brings a rare perspective: he has built and sold a company, and now he sits on the other side helping the next generation of startups scale. We talk about why speed to a working product stopped being the advantage it used to be, why plugging into something like Microsoft Copilot can matter more than the product itself, when a founder should build a tool in house instead of buying one off the shelf, and the one mistake that quietly kills startups that looked incredible in the demo. Tom also shares how his own son is building a two person startup powered by AI agents, and why he believes trust, not technology, is what closes enterprise deals.You'll learn:Why time to MVP has collapsed and what that means for founders competing on speed aloneHow to decide whether to build your own tool or buy something off the shelfWhy distribution through platforms like Microsoft Copilot can matter as much as the product itselfWhat enterprises actually check before they trust a startup with their dataWhy technical founders often overbuild and lose sight of what customers actually wantHow AI agents are changing the size and structure of a modern startup teamWhat founders should build into their architecture from day one to avoid costly rework laterHow to test an idea with a proof of concept before committing months of runway to itWhether you are a founder building your first AI powered product or an operator trying to understand how enterprise buying decisions really get made, this episode gives you a practical framework for the next 90 days.Quick thanks to our sponsor OS4 Techno Services, a Quebec based IT firm helping organizations navigate infrastructure, cybersecurity, cloud, and AI with strong Microsoft expertise. Learn more about OS4 Techno Services: https://www.os4techno.com/📌 Resources & LinksTom Davis on LinkedIn: https://www.linkedin.com/in/tom-davis-startups/📬 Enjoyed this episode?Subscribe to the AI Frontier Playbook: https://theaifrontierplaybook.substack.com/Connect with me on LinkedIn: https://www.linkedin.com/in/samuelboulanger/Visit the website: https://frontierplaybook.ai

AI projects stall when nobody names the real business bottlenecks first. In this episode of The AI Frontier Playbook, I sit down with Ryan Cunningham, Microsoft CVP of Copilot Studio and Power Platform, to talk about why, and what actually separates a good AI use case from a wasted one.Ryan leads agents, apps, and automation across the entire Microsoft stack, and in this conversation we get into why most AI projects stall before they even start, whether apps and low code still matter now that you can generate almost anything from a chat window, how to actually choose between pro code, low code, and vibe coding, and how Copilot Studio's credit based pricing is reshaping the way teams think about cost. Ryan shares the real story behind a customer who turned a 36 hour quote generation process into a 2 hour one, and why that is the model for a good AI use case. Quick thanks to Kezber, a Microsoft partner, for sponsoring this episode.You'll learn:Why most AI projects fail because nobody defined the outcome before picking a toolHow to turn a vague business problem into a measurable AI use case with a clear ROIWhy apps and structured interfaces are not going away even as chat based AI expandsWhat actually happens to custom software and SaaS as vibe coding becomes mainstreamHow to decide between pro code, low code, and vibe coding for a given problemWhy credit based consumption pricing is replacing flat per user licensing for AIHow Power Apps turning into MCP servers changes what agents can do with your dataWhat governance and oversight look like once agents start operating with less human involvementWhether you are a business leader trying to justify your first AI investment or a developer deciding which tool fits the job in front of you, this episode gives you a practical framework for making that call.Thanks to Kezber for sponsoring this episode. Learn more about Kezber: https://kezber.com/📌 Resources & LinksRyan Cunningham on LinkedIn: https://www.linkedin.com/in/rycu📬 Enjoyed this episode?Subscribe to Mastering AI for Productivity: https://samuelboulangerai.substack.com/Connect with me on LinkedIn: https://www.linkedin.com/in/samuelboulanger/Visit my website: https://frontierplaybook.ai

Learn how to automate sales and back-office processes with AI for business, without overwhelming your team. In this episode of The AI Frontier Playbook, I sit down with Eric Murray, Director of Innovation at Kezber, a Microsoft partner in Quebec, to break down AI implementation and business workflow automation with Microsoft Copilot, Power Platform, and Copilot Studio.Many companies fail when they try to automate an entire department at once. Eric's approach starts instead with repetitive tasks that are low risk for your business, owned by a champion who knows every exception. We dig into how to identify the right processes to automate first, why most AI projects stall before delivering anything, and what it really takes to move from a working prototype to a secure production system.Eric brings a practitioner's view shaped by real client deployments. We walk through a full accounts payable automation scenario, from the invoice landing in a shared mailbox to the data reaching the ERP, and break down which steps should be plain deterministic automation with Power Automate and which ones genuinely need AI reasoning in the loop. We also cover human-in-the-loop design, exception handling, data residency under Quebec's Law 25, governance with tools like Microsoft Purview and Agent 365, and why a single model swap took one of his AP solutions from 50 dollars a month to 5 dollars a month.You'll learn:Why deploying Copilot licenses is the starting line, not the finish lineHow to spot a strong first automation candidate: repetitive, low risk, and owned by a championWhat red flags signal a process is not ready to automateHow to slice a process like accounts payable into small chunks that deliver value in weeks, not monthsWhen to use straight automation versus AI reasoning inside the same workflowHow to design human-in-the-loop escalation so people trust the system instead of working around itWhy security, access control, and data residency are the real work behind any AI agentHow to measure ROI after launch and diagnose whether the problem is the tool or adoptionWhether you are a business leader wondering where to start with AI automation or a Power Platform and Copilot Studio builder taking prototypes into production, this episode gives you a concrete framework for picking the right first project and scaling it without losing trust.📌 Resources & LinksEric Murray on LinkedIn: linkedin.com/in/eric-murray-ing-18851a6Kezber: https://kezber.com/en/📬 Enjoyed this episode?Subscribe to The AI Frontier Playbook Newsletter: samuelboulangerai.substack.comConnect with me on LinkedIn: linkedin.com/in/samuelboulangerVisit my website: samuelboulanger.com#AI #FutureOfWork #MicrosoftCopilot #PowerPlatform #Automation

Why does AI adoption fail in business? In this episode of The AI Frontier Playbook, an AI podcast for enterprise leaders, I sit down with Stacey Mulcahy, Senior Director of the Microsoft Garage, to break down the real AI adoption challenges companies face and how Microsoft solves them with curiosity, hackathons, and hands-on AI skilling.Stacey leads the team behind the Microsoft Global Hackathon and the Skill Up AI program, engaging over 70,000 employees in grassroots AI learning and employee enablement across the company.We go deep on why formal learning paths are not enough, why curiosity and exploration drive real behavior change, and what Stacey calls the grok moment: the point where AI stops being a search engine and starts changing how you work. She also shares why leadership modeling makes or breaks adoption, how she runs learning in the open by streaming her own experiments, and why the GitHub Copilot CLI has completely changed her daily workflow.You'll learn: Why AI adoption gets stuck even when the tools and programs are good How to create the grok moment that turns skeptics into adopters Why leadership modeling matters more than any champion program What community skilling looks like and why learning with others beats learning alone How creativity and play outperform productivity use cases as an entry point to AI Why durable skills like adaptability, delegation, and taste matter more than tool knowledge How Stacey uses the GitHub Copilot CLI, custom agents, and MCP integrations to run her day What the future of work looks like when everyone manages a team of agentsWhether you are a leader driving AI transformation in your organization or an individual contributor trying to build AI literacy on your team, this episode gives you a practical playbook for making adoption actually stick.📌 Resources & LinksStacey Mulcahy on LinkedIn: https://www.linkedin.com/in/staceymakes/📬 Enjoyed this episode?Subscribe to Mastering AI for Productivity: https://samuelboulangerai.substack.com/Connect with me on LinkedIn: https://www.linkedin.com/in/samuelboulanger/Visit my website: https://samuelboulanger.com#AI #FutureOfWork #AIAdoption #Microsoft #AILiteracy

How to use AI for content creation without losing your voice. Jack Rowbotham earned 7.5 million LinkedIn impressions in a year at Microsoft, and in this episode of The AI Frontier Playbook he breaks down his exact AI content workflow.The real skill in the AI era is not creating more. It is having the judgment to ignore almost all of it. I sit down with Jack Rowbotham, Senior Product Marketing Manager for Copilot Studio at Microsoft, who built a personal brand of over 90,000 followers while working full-time and was named a Microsoft Frontier Trailblazer for FY26. We unpack the line between using AI tools to amplify your voice and letting them replace your thinking.You'll learn:How to use AI tools for content creation without sounding like a robot or losing your voiceWhy leading with your own perspective before touching AI is what protects authenticityHow Jack runs AI evals on his own LinkedIn posts to keep AI-generated content accurateWhat separates using AI to amplify your voice from outsourcing your thinkingHow the agentic AI agent boss model reshapes work for marketers, creators, and knowledge workersWhy prioritization and judgment are the new edge once AI removes the execution barrierWhat a personal brand agent built in Microsoft Copilot Studio can actually do for youHow to turn event photos, whiteboards, and slides into usable notes with AI in minutes📌 Resources & LinksJack Rowbotham on LinkedIn: https://www.linkedin.com/in/jackrowbotham/Jack's LinkedIn Learning course, How to be an Agent Boss: https://www.linkedin.com/learning/how-to-be-an-agent-boss-leading-ai-agents-at-workEnjoyed this episode? Subscribe to the newsletter: https://samuelboulangerai.substack.com/Connect with me on LinkedIn: https://www.linkedin.com/in/samuelboulanger/Visit my website: https://frontierplaybook.ai