Smith Horn Group delivers product strategy, from discovery to rapid prototyping, and user acceptance testing, with code. AI Workflow Transformation is delivered from decades of experience building software, AI technologies, and digital transformation.
External link for Smith Horn Group
Vancouver, British Columbia V6G 2L2, CA
Smith Horn Group reposted this
Building with #ClaudeCode, great. Making Kasey Fu laugh, priceless. On the #PMHIVE pod this week, we covered: - adventures in building and the delight of teaching others after many bruises along the way - Anti-Claw, and building with your own tools - What motivated Skillsmith YouTube and Spotify links below 👇
Smith Horn Group reposted this
By the numbers: this is the end of month 2 building Skillsmith, and only 10 months since I switched from using Lovable and Cursor over to Claude Code last May, 2025. If you have not set up a workflow where you are massively productive with agents, I invite you to call me or David Gratton so that we can get you on the agentic train. This exponential technology is accelerating in ways that the human brain can't quite come to terms with, yet. In sixty (60) days, I've written/orchestrated agents to write, 1.8 times the ENTIRE Harry Potter series in documentation, and a similar additional volume of code. That should not be possible. It was also one of 4 projects I've been involved with during this period. Skillsmith uses the workflow that we teach in the Northeastern University Vancouver workshop. There's a slide we present on 3-6 months to go from Novice to Practitioner. We're seeing our alumni pop at the 2-3 month point, and from a software perspective, walk on water and know kung-fu around the 6 month point. I'm no expert, yet. I see them in the Agentics Foundation calls, Rueven Cohen introducing the Prime Radiant, and deep discussion on WhatsApp and Discord. I'm no Steve Yegge or Boris Cherney. But I gotta tell you, that level 8 agent systems orchestrator doesn't look unachievable anymore in large part because of exponential learning and community. This is all learnable, just faster than ever before. The people at expert level have 10-30 years of software engineering behind them. It took them 10k hours, and another 10k hours, and another 10k hours to get there. I don't think the next generation of experts will need that many hours, or years. The number is closer to a 6 month sprint for the prior equivalent of 10k hours to gain mastery. It's about 960 hours, or 40 hours per week x 4 weeks per month x 6 months because the learning cycles are accelerated. It used to take days, weeks, months, quarters to see if your thinking produced results, or not, and then get feedback which may give false positives and negatives, ... and hopefully you have good mentors who can help you aside from DIY. With an agentic framework, you can construct solid high-signal feedback loops, and build with others for cohort-based learning which further accelerates the learning cycles. Skillsmith is maintainable code. It's production code. It's code that has already gone through intensive security testing. I can see hundreds of people have already installed the MCP and it only went live a few weeks ago with zero marketing. It's code that includes a website that has already been optimized, has dialled SEO/AEO, and is now running it's own weekly automated A/B tests. In the comments is a new blog on "How I AI". Some of this was presented at last week's Agentics Foundation Vancouver meetup. I took some of the questions and jammed into this blog. It's dense. I'll rewrite it at some point. I hope you enjoy and it provokes your learning. Feedback welcome, as always.
Smith Horn Group reposted this
Superpowers with Claude Code? 🦾 That's what it feels like. Read all the books in the library on a topic with near-perfect retention. Use that knowledge and understanding of patterns for a specific use case. Research, refine, and detail the problem at a level that would take a whole team ... months. Then build and prototype with extreme speed to test it over hours and days. TLDR 🔥 --> December 12 - if you're in Vancouver and have 3 hours to gift yourself superpowers, please join us in Yaletown to get you up and running, from Zero to One. A strong understanding of software development is a prerequisite. You'll need your laptop with a $20/month Claude plan, Docker, Linear, and VS Code (free plans). **Longer Story** - David Gratton and I ran our 10th workshop in 6 months last week - I had been using Google Docs to maintain instructor and learner content - We've been duplicating this doc and updating it each time, for each cohort, refining how to run a first-time onboarding to agentic engineering with technologies that are changing every few days! - I wrote out my frustrations as a user managing content that is always changing. There has to be a better way. Surely other people have figured out dynamic content management. - I set up a project in Claude, which facilitated a number of threads of research - that rolled into a concept brief - the concept brief rolled into a product requirement document (in the same Claude project) - that rolled into a design brief where I could see potential solutions, user stories, interactions ... - and took these docs into VS Code with Claude Code - From Friday night to Sunday morning, while getting a tree with the family, swimming lessons, biking to the park, laundry, ... etc., Claude build out a solution from my documentation - I'm reviewing a custom content management platform TODAY that I fully own, can maintain, and is core IP to Zero to One Events Ltd. as we figure out how to scale Agentic AI training and development for individuals and teams - this workflow, from concept to burning down a backlog of tasks in Linear, is what we will cover on December 12 in three (3) hours - while it's not an instant, magic wand feel, like vibe-coding, it is a big step towards "software-on-demand", which is a significant threat to many (but not all) SaaS business models --> luma link below in comments
Re-anchor the customer journey. Whatever product you're building, as the interface and expectations change rapidly, revisiting what primal story a person wants needs to be front and center as you explore what's next.
"Anything optimized for web or mobile rather than AI will start to feel outdated. How product teams react to this shift is crucial. Too many established companies will fall short because they only augment their offerings instead of fully reimagining them. So how should product leaders proceed, based on these implications?" - Scott Belsky I'm in this cognitive dissonance with apps, tools, and interfaces daily now. My comfort is getting normalized to VS Code, Claude-Flow, and the terminal to trigger Supabase, Render, and Vercel from the CLI. Awkwardly, many of these tools require manual log in from the terminal to get running, then the agent can take over and I can describe the destination.
David Gratton and I have workshops underway for this kind of role. The next public session is Thursday August 21 in Vancouver. Corporate sessions are available on request. https://021.events/
🔥 Apple is hiring an AI PM. Even Netflix is paying $900k/yr for the AI PM roles. Here are the good resources to become an AI PM: Before we dive into becoming an AI PM resources, the most important thing you need to master is building AI Product Strategy. Because your goal as an AI PM isn’t just to bake AI into your products... It’s to build AI products that give you a real competitive advantage. If you want to learn how in a 9-week cohort taught by OpenAI’s Product Leader Miqdad Jaffer, the first 100 people will also get a personal written review of their AI Product Strategy + $550 off: https://lnkd.in/dAHYq8nb Let's dive into the resources: 𝟏. 𝐁𝐚𝐬𝐢𝐜 𝐂𝐨𝐧𝐜𝐞𝐩𝐭𝐬 Start with what an AI PM is: https://bit.ly/whatisaipm Next, for most PMs, it makes no sense to dive deep into statistics, Python, or loss functions. Instead, read about Transformers, and LLMs: https://bit.ly/3EZtCLs 𝟐. 𝐏𝐫𝐨𝐦𝐩𝐭 𝐄𝐧𝐠𝐢𝐧𝐞𝐞𝐫𝐢𝐧𝐠 Free resources: 🔗 GPT-4.1 Prompting Guide 🔗 Anthropic Prompt Engineering 🔗 Prompt Engineering by Google 🔗 System Prompt Analysis for Claude 4 🔗 Anthropic Prompt Generator 🔗 Anthropic Prompt Library 🔗 Prompt Engineering Course By Anthropic All links: https://bit.ly/pcprompts 𝟑. 𝐅𝐢𝐧𝐞-𝐓𝐮𝐧𝐢𝐧𝐠 Learn by doing. No coding: 🔗 OpenAI Platform (start here) 🔗 Hugging Face AutoTrain (best for other models) 🔗 LaMA-Factory (fine-tune open-source LLMs) All links: https://bit.ly/pcfinetune 𝟒. 𝐑𝐀𝐆 (𝐑𝐞𝐭𝐫𝐢𝐞𝐯𝐚𝐥-𝐀𝐮𝐠𝐦𝐞𝐧𝐭𝐞𝐝 𝐆𝐞𝐧𝐞𝐫𝐚𝐭𝐢𝐨𝐧) RAG, by definition, requires a data source + LLM. But there are dozens of possible architectures. 🔗 A free, interactive RAG simulator: https://lnkd.in/dsi6C5b9 I also recommend a simple step-by-step exercise to build a RAG chatbot in practice. No coding: https://lnkd.in/dew--RqD 𝟓. 𝐀𝐈 𝐀𝐠𝐞𝐧𝐭𝐬 & 𝐀𝐠𝐞𝐧𝐭𝐢𝐜 𝐖𝐨𝐫𝐤𝐟𝐥𝐨𝐰𝐬 My favorite tool, by far, is n8n. You can host a free version locally or in the cloud. Start with those guides: 🔗 MCP for PMs 🔗 Automate Anything with n8n 🔗 AI Agent Architectures My favorite free resources: 🔗 Google Agent Companion 🔗 Anthropic Building Effective Agents 🔗 IBM Agentic Process Automation All links: https://bit.ly/pcaiagents 𝟔. 𝐀𝐈 𝐏𝐫𝐨𝐭𝐨𝐭𝐲𝐩𝐢𝐧𝐠 & 𝐀𝐈 𝐁𝐮𝐢𝐥𝐝𝐢𝐧𝐠 My default no-code tech stack: Lovable, Supabase, GitHub, Netlify, n8n, Stripe. Four practical tutorials: 🔗 AI Prototyping 🔗 How to Quickly Build SaaS Products With AI 🔗 How to Build a Full-Stack App with Lovable 🔗 No-Code B2C SaaS Template With Stripe Payments All links: https://lnkd.in/du5TSFfd 𝟕. 𝐅𝐨𝐮𝐧𝐝𝐚𝐭𝐢𝐨𝐧𝐚𝐥 𝐌𝐨𝐝𝐞𝐥𝐬 My favorite models (July 17, 2025): Claude Opus 4 for coding Gemini 2.5 Pro for everything else 𝟗. 𝐎𝐭𝐡𝐞𝐫 𝐑𝐞𝐬𝐨𝐮𝐫𝐜𝐞𝐬 All links: https://lnkd.in/dSA3JgFw — Shout out to Paweł Huryn for curating and writing these resources!