About
Articles by Lee
-
Welcome to the Model Catalog in VS Code AI Toolkit
Welcome to the Model Catalog in VS Code AI Toolkit
Great blog on Azure Dev Community Blog by Vinayak Hegde on the availability of the Expanded Model Catalog for the AI…
Activity
10K followers
-
Lee Stott reposted thisAfter only two weeks of working with Carlos Castro I can definitively confirm that he is the model whispering 🐐Lee Stott reposted thisWe benchmarked the GitHub Copilot agentic harness against the harnesses that ship leading models natively. Holding the model and task fixed across SWE-bench Verified, SWE-bench Pro, SkillsBench, TerminalBench, and Win-Hill, the results were clear: ✅ Task resolution on par with model-vendor harnesses ✅ Fewer tokens across most configurations 💡 A key learning: With GitHub Copilot supporting more than 20 models, you're free to pick efficiency or peak quality per task. Explore the data ⬇️ https://lnkd.in/eq_RnAPj
-
Lee Stott reposted thisLee Stott reposted thisCOZY AI CLOSING: Tomorrow the last episode of Mr. Maeda’s Cozy AI Kitchen (CAIK 🎂) goes live! The special AI CHEF apron has had a wonderfully long run as a truly non sequitur partnership with 🧑🍳 Ross Heise as a way to make AI a little more … cozy. What is in the last episode of the Cozy AI Kitchen? There is an epic ending drone shot after me and Ross hang up our aprons that you don’t want to miss! And we bring back some SFX for those who remember the first episodes of the series. — Watch all 50+ 🍰 episodes https://lnkd.in/g6upvbGX that trace the history of AI’s evolution from its kickoff moment with gpt-3.5T 🤓
-
Lee Stott reposted thisLee Stott reposted thisYour AI agent can deploy clean and still fail. That is the uncomfortable part of shipping hosted agents. John Folberth had agents deploy successfully. The pipeline looked healthy. But the agent failed to return responses because of a dependency issue. That is where smoke tests earn their place. - Not as quality scoring. - Not as a replacement for evaluations. - Not as proof that the agent is "good." Smoke tests answer the first release question: Is the agent reachable, responding, and following its basic instructions? For Foundry Hosted Agents, that can be a JSON test catalog plus a Python runner wired into GitHub Actions after deployment. Start with the failures users notice first: on-topic response, refusal, continuity, hallucinated premise, context-dependent answer. Share this with the person approving agent releases. https://lnkd.in/eZM6zkzj #AzureAI #DevOps #GitHubActions #AIAgentsSmoke Test Microsoft Foundry Agents with GitHub Actions | Microsoft Community HubSmoke Test Microsoft Foundry Agents with GitHub Actions | Microsoft Community Hub
-
Lee Stott reposted thisLee Stott reposted thisStreaming long context to an LLM is the easy part. Managing the massive memory footprint and keeping inference fast and cost-effective is where most developers get stuck. I’m excited to share my upcoming session on KVStream. We are moving past traditional, resource-heavy context handling to look at how we can efficiently manage the Key-Value (KV) cache for long-context LLMs. We’ll dive deep into memory efficiency and see how to dramatically reduce KV cache overhead without losing model accuracy using: • KVStream framework to dynamically compress and stream cache data. • Local LLM deployment to measure latency improvements firsthand. • Real-time performance tracking to observe throughput gains and memory savings. 𝘛𝘩𝘦 𝘨𝘰𝘢𝘭? To move beyond hardware bottlenecks and build the architecture needed to serve long-form content, extensive codebases, and deep chats smoothly. We’ll look at how to maintain high accuracy and low TTFT (Time to First Token) even under heavy context loads. If you’re looking to scale your LLM applications, slash memory costs, and stop letting KV cache bloat slow down your generation, this one is for you. Catch the session: 📅: 2nd July, 4:30PM BST/8:30 AM PT/ 9PM IST 📍 Join here: https://lnkd.in/gFZSyFjq Lee Stott #GenerativeAI #LLMs #KVCache #Microsoft #OpenSource #MachineLearning #AIEngineering #DeepLearning #LLMOps #TechCommunity
-
Lee Stott reposted thisLee Stott reposted thisThought I had a masterpiece...turns out I had a relatively semi-AI slop website 😭 But that's OK because I'm not a designer by profession! 🥹 However, tools like Impeccable can help non-designer devs (like me) mitigate the chance of our vibe-coded apps and websites from looking like AI slop! Check out my review below & enable Impeccable in the GitHub Copilot App to critique your vibe coded projects! 👩🏾💻
-
Lee Stott reposted thisLee Stott reposted thisWe have version control, code review, linters, and profilers for our code. For the *design* of what we build, most of us have "looks good to me." I spent a session closing that gap on a real, already-shipping app - and it changed how I think about UX work. The tool: Impeccable, a frontend design skill by Paul Bakaus, now built right into the GitHub Copilot app. You invoke it inside a Copilot session — no separate tool, no context switch. What actually impressed me is that Impeccable brings an engineering approach to UX. It treats design quality the way we already treat code quality: - A version-controlled spec. My design system became committed files (PRODUCT-dot-md + DESIGN-dot-md), diffed and reviewed in a pull request. Design intent stopped being tribal knowledge in my head. - A linter for design. A deterministic detector flags off-spec values and known "AI-slop" patterns mechanically - one pass went from 14 findings to 0. But unlike a dumb linter, it knew intent from accident: it left my deliberate one-off choices alone and documented them instead of forcing conformance. - Profile before you optimize. The one performance change happened only after I measured frame budgets under a throttled CPU. Desktop showed zero difference; the throttled phone showed ~⅓ less frame time. The data decided - not my taste. - A prioritized, executable backlog. Every finding mapped to a specific command, ordered by real user impact. I always knew exactly what to do next, and stayed in control of every fix. Full write-up with the before/afters and the measured numbers (don't forget to subscribe to my "Agentic Engineering in Action" newsletter - the button is in the article below) 👇 🔗 Impeccable: https://impeccable.style 🔗 The app, live: https://lnkd.in/eVagVky9 Evan Boyle, Mario Rodriguez, Burke Holland, Christoffer Noring, Lee Stott, Daniel Meppiel, Chris Romp, Joel Norman, Francesco Manni, Jaime De Mora, Morten Stange Bye, Haakon Hasli, Kristine Husabø Hansen, Else Tefre, Christian Tryti, Majid Hajian, Muhammad Daniyal (Dani), Mario Kedziora, Osman Cokakoglu, Ömür Sert, Jonas Helin, Adil I., Steph Rogers, Sebastien Le Calvez #GitHubCopilot #AgenticAI #AINativeDev #Impeccable #GitHubCopilotAppDesign, but with tests: an engineering approach to UX with ImpeccableDesign, but with tests: an engineering approach to UX with Impeccable
-
Lee Stott reposted thisLee Stott reposted this#AIEngineer World Fair 2026 is in full swing this week in San Francisco! Love the fact they stream many of the tracks. I'm on staycation this week meaning I am watching sessions from a beanbag on my projector screen at home: 🍹✨ https://lnkd.in/g_9K3-v6 Microsoft is a signature sponsor - and we have at ton of really good talks and workshops lined up. Just look at the "Track M" schedule. Scan the slides below for key sessions each day. https://lnkd.in/gN8bdHW8 Huge thanks to my colleague Pamela Fox who helped deliver a workshop that we had created for #MicrosoftBuild that focuses on the use of Agentic Skills to drive the agent optimization loop for Hosted Agents on Microsoft Foundry. This workshop is being maintained as a living resource for exploring the various Microsoft Foundry observability features through the lens of a single scenario (multi-agent travel concierge). Use it to get familiar with built-in observability features then move to code-first observability (with skills) and continuous observability (with agent optimizer) to build your intuition for agentic devops. Workshop Title: Observe. Optimize and Protect Your Hosted Agents in Microsoft Foundry Repo: https://lnkd.in/g-WJ_c8B #MicrosoftFoundry #Observabilty #AgentDevOps #Evaluations #Tracing #Monitoring
-
Lee Stott reposted thisLee Stott reposted thisThe #CopilotCLI has a new terminal interface and it sickkkkk!! Check out these interactive (clickable) tabs as I work through issues, PRs, etc. GitHub
-
Lee Stott reposted thisLee Stott reposted thisClaude Sonnet 5 is now generally available in Microsoft Foundry. This marks another step in bringing the industry's leading frontier models into enterprise production environments. With Claude Sonnet 5 in Microsoft Foundry, organizations can build and scale coding agents, research assistants, and business process agents while using the Azure-native security, governance, networking, and identity controls they already rely on. Developers can use Claude Sonnet 5 with Foundry Agent Service and Microsoft IQ to build, ground, evaluate, and operate agentic applications, with support for capabilities including US Data Zones, Zero Data Retention options, and consolidated Azure billing. Together, Microsoft Foundry and Claude Sonnet 5 provide a trusted path for bringing frontier AI into enterprise production environments. Read more here: https://lnkd.in/ebXNwXmp #MicrosoftFoundry #AI #EnterpriseAI #AgenticAI #AzureAI #Anthropic
-
Lee Stott liked thisLee Stott liked thisAfter 30.9 years at Microsoft, much of it in Xbox, it's Game Over for me: time to retire and take it easy. Joining Microsoft and moving across the Atlantic was a life changing experience, but it's time to VRP out. I met so many smart people during my time, I will miss you all, and the many teams I was a part of: Visual Studio Debugger, C#, HD DVD, KIN, Windows Phone, Xbox (Durango), PowerApps, and Xbox (Scarlett). Good luck to Project Helix - I'll be first in line for one when the time comes.
-
Lee Stott liked thisLee Stott liked thisToday was my 𝗹𝗮𝘀𝘁 𝗱𝗮𝘆 at Chitkara University as a 𝘀𝘁𝘂𝗱𝗲𝗻𝘁 💙 It's hard to believe that the journey which began in 𝟮𝟬𝟮𝟭 has finally reached its last day on campus. From 𝗕𝗖𝗔 𝘁𝗼 𝗠𝗖𝗔, these years gave me much more than a degree ever could 𝗼𝗽𝗽𝗼𝗿𝘁𝘂𝗻𝗶𝘁𝗶𝗲𝘀, 𝗹𝗲𝘀𝘀𝗼𝗻𝘀, 𝗺𝗲𝗺𝗼𝗿𝗶𝗲𝘀, 𝗹𝗮𝘁𝗲-𝗻𝗶𝗴𝗵𝘁 𝗮𝘀𝘀𝗶𝗴𝗻𝗺𝗲𝗻𝘁𝘀, 𝗲𝘃𝗲𝗻𝘁 𝗽𝗿𝗲𝗽𝗮𝗿𝗮𝘁𝗶𝗼𝗻𝘀, 𝗮𝗰𝗵𝗶𝗲𝘃𝗲𝗺𝗲𝗻𝘁𝘀, 𝗳𝗮𝗶𝗹𝘂𝗿𝗲s, and countless moments that shaped me. Walking through the campus today felt different. The same buildings, classrooms, and corridors suddenly carried years of memories. Thank you, Chitkara University, for being home for the last five years. 𝗔 𝗰𝗵𝗮𝗽𝘁𝗲𝗿 𝗰𝗹𝗼𝘀𝗲𝘀 𝘁𝗼𝗱𝗮𝘆, 𝗯𝘂𝘁 𝘁𝗵𝗲 𝗺𝗲𝗺𝗼𝗿𝗶𝗲𝘀 𝘄𝗶𝗹𝗹 𝘀𝘁𝗮𝘆 𝗳𝗼𝗿𝗲𝘃𝗲𝗿 ✨🧿 #ChitkaraUniversity #StudentLife #BCA #MCA #Memories #LastDay #NewBeginnings Dr. Madhu Chitkara
-
Lee Stott liked thisLee Stott liked thisLast Friday, 400+ people gathered at the Guildhall in London to mark the same milestone. They had each completed a data and AI apprenticeship with Cambridge Spark. And almost everyone in that room had learned these skills alongside a full-time job, at organisations like Financial Conduct Authority, Eurostar, HSBC, the NHS and Virgin Atlantic. Each of them now carries their newly learned capabilities back into their organisation, allowing for more growth for themselves, the company and their colleagues whom they will inspire. Congratulations to everyone who graduated on Friday. It was a privilege to celebrate with you!
-
Lee Stott liked thisLee Stott liked thisAfter 21 years, I will be retiring from Microsoft, and July 01 will be my last day. I’ve elected to take the Voluntary Retirement Package (VRP). I’m incredibly grateful for the opportunities Microsoft provided, the deep technical challenges I got to solve, and the lifelong friendships and partnerships created along the way. I will cherish those memories long after July 01. Above all, I’m grateful to be able to fully retire at 53. To my Microsoft colleagues, past and present: it’s been an absolute honor. Thank you. Dan
-
Lee Stott liked thisLee Stott liked thisWhat a great experience at the European Power Platform Conference. This one was a big relief too, since I had been prepping for this session for a while. Big thanks to Michael Heath and David Lorenzo López for all the support along the way. Our session was about #Agent365, and I got really excited when the audience told me they had not really worked with it yet. What really makes these trips though is the people. Sean Astrakhan - I could sit and listen to you all day. Thanks for the great session. Andreas Adner, we finally met face to face, my friend. You folks are the best, not just as friends but as the people cheering the loudest. Elaiza Benitez, so good to see you again. Parvez Ghumra, Bülent Altinsoy, Samir M., Agnius Bartninkas, Craig White, Keith Atherton, Daniel Laskewitz, Charles Sexton, Julie Koesmarno, Jakub Bajla, and plenty more, never a dull moment with you. Emma-Claire Shaw, thank you for the picture, and sorry we missed each other this time. We will fix that next round. And the Power CAT team is not someone you bump into every day. Great meeting you Henry Jammes, Giorgio Ughini, Adi Leibowitz, and good to see you again Lewis Baybutt 😊 And Bridgette Robertson, thank you for organizing and running such a smooth event. Can’t wait to see you guys at #PPCC26! #eppc26 #copilotstudio #agent365 #microsoftfoundry
-
Lee Stott liked thisLee Stott liked thisCOZY AI CLOSING: Tomorrow the last episode of Mr. Maeda’s Cozy AI Kitchen (CAIK 🎂) goes live! The special AI CHEF apron has had a wonderfully long run as a truly non sequitur partnership with 🧑🍳 Ross Heise as a way to make AI a little more … cozy. What is in the last episode of the Cozy AI Kitchen? There is an epic ending drone shot after me and Ross hang up our aprons that you don’t want to miss! And we bring back some SFX for those who remember the first episodes of the series. — Watch all 50+ 🍰 episodes https://lnkd.in/g6upvbGX that trace the history of AI’s evolution from its kickoff moment with gpt-3.5T 🤓
Experience & Education
-
Microsoft
View Lee’s full experience
By clicking Continue to join or sign in, you agree to LinkedIn’s User Agreement, Privacy Policy, and Cookie Policy.
Licenses & Certifications
Patents
-
Program execution monitoring using deep memory tracing
Issued 11782819
See patentA user-annotated reference implementation identifies variable values generated by the reference implementation during its execution. A software implementation under analysis is executed. Variable values in the running memory of the program code of the software implementation under analysis, during its execution, are identified and copied. The variable values traced from the running memory of the program code are compared against the annotated variable values generated by the reference…
A user-annotated reference implementation identifies variable values generated by the reference implementation during its execution. A software implementation under analysis is executed. Variable values in the running memory of the program code of the software implementation under analysis, during its execution, are identified and copied. The variable values traced from the running memory of the program code are compared against the annotated variable values generated by the reference implementation, to determine a similarity between the program code under analysis, and the reference implementation. An output is generated that is indicative of whether the traced variables from the program code under analysis are the same as the annotated variable values generated by the reference implementation
-
Brokerage tool for accessing cloud-based services
Issued 10908933
See patentThe claimed subject matter includes techniques for providing access to a cloud-based service. An example method includes generating a user interface form compatible with an application program, the user interface form being displayable via a user interface of the application program. The method also includes receiving a user request at a brokerage engine that runs as an extension of the application program through the user interface form. The method further includes identifying a cloud-based…
The claimed subject matter includes techniques for providing access to a cloud-based service. An example method includes generating a user interface form compatible with an application program, the user interface form being displayable via a user interface of the application program. The method also includes receiving a user request at a brokerage engine that runs as an extension of the application program through the user interface form. The method further includes identifying a cloud-based service to perform the request, and performing the request using the cloud-based service by translating the request to be compatible with the identified cloud-based service. The method also includes generating a new user interface form integrated with the user interface of application program, the new user interface form populated with data from the cloud-based service.
-
Program Execution Monitoring Using Deep Memory Tracing
Filed US MS 408804-US-NP
-
Brokerage Tool for Accessing Cloud-Based Services
Filed US MS# 407087-US-NP
Recommendations received
5 people have recommended Lee
Join now to viewOther similar profiles
Explore more posts
-
Steve Hare
Sage is supporting the Department for Science, Innovation and Technology’s AI Skills Boost, which aims to expand free AI training and upskill 10 million UK workers. Our AI Fundamentals course is one example of how we’re already helping customers adopt AI in practice. Empowering people - particularly small businesses - with the tools and skills to unlock its potential and drive productivity across the economy remains a big priority. https://lnkd.in/eWJ8UvE9
3 Comments
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content