As a Product Manager with an Electrical Engineering degree and over 20 years in technology, I never lost my love for building things. But somewhere between talking to users, enterprise architecture discussions, stakeholder meetings, and system integration projects, the joy of hands-on coding had gotten buried under layers of setup overhead, configuration complexity, and infrastructure friction.
I built Unisys COBOL85 Mainframe applications, Windows DLLs, and web drag-and-drop features (long before HTML5) earlier in my career. I've been hands-on with multiple tech stacks, databases, and operating systems throughout my journey from engineer to Product Manager. But when you're juggling cross-functional teams and navigating enterprise software development cycles, the barrier to spinning up personal projects feels insurmountable.
Then AI changed everything. Not by teaching me to code, but by removing all the friction that had accumulated over 15+ years of enterprise development.
The Breakthrough: When Engineering Joy Returned
My first real revelation didn't come from learning to code for the first time. It came from remembering why I fell in love with engineering in the first place. That started in November of 2024. I wanted my son to safely explore large language models without the privacy concerns of public AI services. When I discovered LM Studio, which allows you to run AI models locally, I knew I had the technical foundation to build something. But in the past, a project like this would have meant weeks of environment setup, dependency management, and infrastructure configuration before writing a single line of meaningful code.
Instead of facing that familiar wall of overhead, I decided to try something different. The LM Studio Web Client became my breakthrough project, entirely prompt-engineered over a week. Five iterations with Claude.ai as my coding partner, and suddenly I had shipped a complete full-stack application with user authentication, real-time streaming chat, role-based permissions, and enterprise-grade security features.
Here's what struck me: for the first time in years, I spent 80% of my time on engineering decisions and 20% on setup overhead. Questions like "How do we architect user authentication that's both secure and family-friendly?" or "What's the optimal way to handle real-time streaming with proper error handling?" These weren't new concepts for me, but AI made implementing them feel effortless again.
That success opened floodgates I didn't even know existed.
From Family Tool to Community Impact
When a co-resident, Tom, started exploring how our community can learn about and benefit from using AI (in February of 2025), I suggested we tackle something that had frustrated me for years: our town's byzantine bylaw system. Thank you, Tom, for starting that exploration thread, and thank you to Mayor Iain Lovatt for believing in what a group of residents could accomplish.
Stouffville has an impressive digital archive of bylaws dating back to 1971, but they're written in legal language and scattered across thousands of PDF documents. Finding out whether you can park on Main Street requires checking multiple bylaws, understanding which ones have been amended, and decoding language written for legal experts, not residents.
Our volunteer team of 10-15 residents with diverse backgrounds (not all technical) processed over 9,000 municipal documents. We downloaded PDFs, converted them using Google Gemini's OCR capabilities, created embeddings for AI search, and built a natural language interface that lets residents ask questions and receive responses in plain English.
The real validation came during our weekend demo at the Leisure Centre. Despite less-than-ideal timing (schools winding down, Saturday Market happening across town), Tom and I set up our booth with homemade posters and QR code pamphlets. Out of hundreds who walked past, about fifteen stopped to engage, and each conversation taught us something new.
The moment that convinced me we were onto something? When Codey, a local bylaw officer, stopped by after seeing our demo and said, "This would really help the whole community." When the people who enforce bylaws for a living think your tool is useful, you're probably on the right track.
But the biggest surprise was people's genuine amazement when they realised residents had built this ourselves. "Wait, you're not from the government?" "Nope. Just residents who got tired of not understanding the rules."
The feedback was overwhelmingly positive. We achieved a 4.88/5 average correctness score from residents, with answers delivered in about 12 seconds on average. We showcased multilingual support, allowing people to ask questions in any language and receive answers in text. Given our multicultural community, this accessibility feature received real appreciation.
You can read more about it1.
The Professional Validation: Subject Matter Expert Evaluation
What started as a community experiment is now undergoing formal validation. The town has initiated a 6-8 week Subject Matter Expert (Bylaw officers) evaluation process, with bylaw officers and summer interns systematically testing our system. I built a sophisticated web-based evaluation system that allows multiple officers to test and rate responses simultaneously, ensuring our tool provides accurate, reliable information to residents It didn’t pan out well! That’s whole idea about rapid prototype and testing. We quickly reverted to our ever-trusted spreadsheet-based evaluation.
This collaboration with town officials has already yielded improvements. The experts caught an important issue where our system referenced outdated fee information from 2002, highlighting the value of professional review in refining AI systems.
But perhaps more importantly, it demonstrated something I never expected: that Product Managers can build systems sophisticated enough to undergo formal municipal evaluation. This isn't just a weekend hobby anymore. It's a tool that could fundamentally change how residents interact with local government.
The Portfolio Keeps Growing
The Stouffville bylaw project wasn't an isolated success. Each project built on the last, reinforcing the same truth: the best Product Managers are the ones who never stop being builders.
RMTFinder showcases how AI can transform professional service discovery. Built for Ontario's $2.1 billion massage therapy industry, it discovers Registered Massage Therapists from the official CMTO database and analyses their online Google reviews using AI-powered sentiment analysis. The system includes sophisticated leaderboard capabilities, demonstrating how Product Managers can tackle complex reputation analysis challenges that serve real market needs.
Stouffville Activities RAG App helps residents discover and register for town programs through intelligent conversation. It combines retrieval-augmented generation with semantic search, showing how AI can make government services more accessible.
Or building numerous instantaneous Edge/Chrome extensions (one of the easiest things that can be done today) for my colleagues at work that wiped out enormous frictions caused by manual, repetitive actions.
Power Automate workflows eliminate mundane tasks for colleagues and teachers, while Raspberry Pi projects power everything from home automation to music servers. Each solution addresses real problems for real people.
The AI Coding Partnership Advantage: Removing Enterprise Friction
Emphasising what millions of articles say about AI-assisted development for experienced engineers: it's not about learning to code. It's about removing the accumulated friction that years of enterprise development create. When you've built drag-and-drop features before HTML5 existed, designed RESTful APIs for legal platforms, and worked with multiple tech stacks, you don't need to learn programming concepts. You need to escape the overhead.
The AI becomes your frictionless development environment. No more spending hours on environment setup, dependency conflicts, or boilerplate code. No more context switching between architecture documents and implementation details. Instead, you have conversations about the engineering challenges that matter:
- "How do we implement secure session management for family use?"
- "What's the best approach for real-time data streaming with proper error handling?"
- "How do we design a vector database that scales with our document corpus?"
Traditional personal projects meant facing the same setup overhead that makes enterprise development slow. Virtual environments, database configurations, deployment pipelines, testing frameworks. By the time you've solved the infrastructure problems, the initial engineering excitement has died.
With AI assistance, I could focus on the architectural decisions and engineering problems I actually wanted to solve. The result? I rediscovered the joy of building things that had originally drawn me to engineering over 20 years ago.
The Professional Transformation
This enhanced building capability doesn't change the fundamental PM discipline of starting with questions to understand jobs-to-be-done. You still begin conversations seeking to understand the real problem, user needs, and context before considering any solutions.
What changes is your ability to rapidly validate solutions after proper discovery. Instead of writing requirements and waiting for engineering estimates, you can quickly prototype and test potential solutions yourself during the solution validation phase.
After building a complete chat application with database management, user authentication, and real-time messaging, suddenly those "impossible" feature requests at work become engineering challenges you can evaluate directly rather than mysterious technical debt.
You still say, "Help me understand what users are really trying to accomplish..." and "What job are they hiring this product to do?" But after proper discovery, you can also say, "I built a prototype, tested it with real residents at our weekend demo, discovered multilingual accessibility issues through systematic testing, and here's what we learned from the 4.88/5 user feedback."
You still ask, "What are the real user needs and constraints?" But you can also say, "I implemented a proof of concept to test our hypothesis, and here are the actual technical constraints and tradeoffs."
You still follow rigorous discovery processes, but you can ship learning iterations during solution validation instead of waiting for full development cycles.
The Shifting Bottleneck: When PMs Become the Constraint
One of the most unexpected consequences of this friction removal has been a fundamental shift in team dynamics. With AI assistance, engineering velocity has increased dramatically, but product management work—understanding user needs, making feature decisions, gathering feedback—hasn't accelerated at the same pace.
Andrew Ng recently shared a striking observation at Y Combinator that perfectly captures this shift. He noted that traditional Silicon Valley ratios of "1 PM to four engineers" are changing rapidly. As he put it: "literally yesterday one of my teams came to me and for the first time when we're planning headcount for a project this team proposed to me not 1 PM to four engineers but to have 1 PM to 0.5 engineers." For the first time in his career, a team proposed having twice as many Product Managers as engineers.
This dramatic shift makes sense when you consider what AI has and hasn't automated. AI can generate code, handle boilerplate, manage dependencies, and even architect solutions. But AI can't sit in a coffee shop and talk to strangers about their problems. It can't conduct empathy sessions or make nuanced product decisions based on deep market understanding.
The result? Engineering is no longer the bottleneck. Product discovery and validation have become the constraint.
For engineer-turned-PMs, this creates a unique advantage. You can both conduct proper discovery AND rapidly validate solutions during the prototype phase. You're not waiting for engineering estimates or development cycles to test your hypotheses.
Breaking the Monotony
The most rewarding part? These side projects break the monotony of traditional product management. Instead of endless stakeholder meetings about theoretical features, I'm shipping actual solutions, gathering real user feedback, and iterating based on genuine usage patterns.
Each project teaches something new. Understanding vector databases for the bylaw search system. Implementing real-time streaming for chat applications. Designing user interfaces that work across different age groups and languages. The AI handles the syntax and boilerplate; I focus on architecture, user experience, and product strategy.
The learning curve never flattens, and that's exactly what keeps this profession exciting even after fifteen years in.
The Community Impact Reality
These aren't just personal learning exercises. The bylaw AI system is helping residents understand complex municipal regulations while undergoing formal evaluation by town officials. Our weekend demo day drew curious residents who were amazed that community members could make municipal law accessible, with multilingual support that helped people ask questions in their native language.
The family chat application provides a secure space for children to explore AI without privacy concerns. RMTFinder demonstrates how AI can transform professional service discovery in a $2.1 billion industry. The automation scripts eliminate repetitive work for colleagues and teachers.
Each project solves real problems for real people, proving that Product Managers with technical curiosity can build solutions that matter, not just spec them.
The Reality Check: When AI Hits Its Limits
Before we get too carried away with AI-assisted development euphoria, recent research provides important context about where these tools excel and where they struggle. A rigorous study by METR in early 2025 offers a sobering counterpoint to the productivity claims.
METR conducted a randomised controlled trial with 16 experienced developers working on their own mature open-source repositories (averaging 22k+ stars and over 1 million lines of code). The surprising result? When developers used AI tools, they took 19% longer to complete tasks than without AI assistance.
Even more striking was the perception gap. Developers expected AI to speed them up by 24%, and even after experiencing the slowdown, they still believed AI had sped them up by 20%. This disconnect between feeling productive and actually being productive is a crucial warning for anyone evaluating AI's impact.
But here's the key insight that validates my experience: the study authors themselves note that "our results are consistent with small greenfield projects or development in unfamiliar codebases seeing substantial speedup from AI assistance".
This explains perfectly why my projects succeeded while the METR study showed slowdowns. The LM Studio Web Client, bylaw AI system, RMTFinder, and Stouffville Activities chatbot were all greenfield projects. I wasn't navigating years of accumulated technical debt, complex legacy architectures, or strict style guidelines that AI tools don't understand.
The METR study focused on developers working in codebases they'd contributed to for an average of 5 years, with established patterns, intricate dependencies, and institutional knowledge that current AI tools simply can't grasp. The average project was over 10 years old with over 1 million lines of code, exactly the opposite of my bootstrapping scenarios.
Where AI Shines vs. Where It Struggles
This research helps clarify the context dependency of AI-assisted development success:
AI excels in:
- New projects without legacy constraints
- Prototyping and proof-of-concept development
- Unfamiliar technologies where the learning curve applies to both humans and AI
- Projects where you can define architecture from scratch
- Situations where "quick and dirty" is acceptable initially
AI struggles with:
- Mature codebases with years of accumulated context
- Projects with strict style guidelines and established patterns
- Complex architectural decisions requiring deep system understanding
- Legacy integration challenges
- Code that requires an understanding of business logic built up over the years
This nuance is crucial for Product Managers considering AI-assisted development. The tool that enables rapid prototyping of the bylaw AI system might actually slow you down when integrating with your company's 10-year-old monolithic backend.
For my side projects, this context advantage was significant. I could make architectural decisions cleanly, choose modern tech stacks, and iterate rapidly because I wasn't constrained by existing systems. The AI and I were both starting from the same blank slate.
But all of this could slowly go away: Agentic, or domain-specific models, or advanced-reasoning models.
The Learning Curve Reality
The METR study also revealed that 56% of developers had never used Cursor before the study, highlighting the steep learning curve for AI coding tools. This suggests that my success came partly from treating AI assistance as a skill requiring practice, not just a magic productivity boost.
The developers who performed best in the study were those with the most AI tool experience, indicating that there may be a high skill ceiling for using these tools effectively. This reinforces that AI-assisted development is a capability to develop over time, not an instant solution.
The meta-innovation
One pattern that emerged repeatedly in my successful projects: using AI to create tools that helped us work with AI. When we needed to combine JSON files, validate data, or create test harnesses for the bylaw system, we'd prompt an AI to generate the code. This meta-approach accelerated development dramatically.
Tom used Cursor (VS Code with AI integration) for programming. I used Claude and Google's AI Studio to create HTML viewers for JSON files and diff tools for comparing versions. These weren't polished products. They were practical tools that solved immediate problems and kept us moving forward.
This isn't about replacing developers. It's about becoming a more technically informed Product Manager who can contribute at every level of the product development process. But it works best when you're building new things rather than maintaining old ones.
These side projects break the monotony of traditional product management while solving real problems for real people. They reconnect you with the engineering joy that originally drew you to technology, especially when you're working in that sweet spot where AI assistance actually accelerates rather than hinders development.
Looking Forward: The New Product Manager Advantage
Product Management has always been about bridging the gap between what users need and what technology can deliver. AI-assisted development doesn't change that fundamental role. It amplifies it while shifting where the real value lies.
Andrew Ng's observation about changing team ratios points to a larger trend: as engineering becomes less of a constraint, the premium on excellent product judgment increases. Teams that can rapidly understand user problems, make good product decisions, and validate solutions quickly will have enormous advantages.
When you can rapidly prototype your own solutions after proper discovery, you become a more effective bridge between business and engineering teams. You can evaluate technical feasibility firsthand rather than relying entirely on estimates. You can test ideas with real users before committing to full development cycles.
For engineer-turned-PMs, this creates a unique competitive advantage. You have the technical depth to build rapidly AND the product discipline to ensure you're building the right things. You can both conduct proper discovery AND validate solutions during the prototype phase.
The future belongs to Product Managers who can think strategically, discover thoroughly, and execute technically. AI just levelled the playing field on execution. The question is whether you're ready to step up to it.
The Always-Learning Mindset
These projects maintain that startup energy and a learning mindset make this profession exciting. They remind me why I fell in love with building products in the first place. When you're processing 9,000 municipal documents or building AI-powered reputation analysis systems, you remember that product management is fundamentally about solving problems that matter.
The bylaw system alone has logged nearly 400 queries from users seeking clarity on everything from parking regulations to pet bylaws. Our analysis revealed fascinating patterns: parking and traffic questions top the list, followed by animal control and zoning inquiries. Most popular topics show us where residents need the most help navigating municipal rules.
The Call to Action
To my fellow engineer-turned-Product Managers: stop letting enterprise overhead kill your building passion. Those technical skills you developed over decades aren't gone; they're just buried under layers of process and friction that AI can remove, especially for new projects.
Pick a problem that bothers you, find an AI coding assistant, and start building something fresh. You'll be amazed how quickly you remember why you loved engineering in the first place. But be realistic about context: AI shines for greenfield development but may slow you down in mature, complex codebases until you develop real expertise with the tools.
Start small. Build the LM Studio wrapper for your family. Automate that annoying process at work. Create a tool that helps your community. Each project removes more friction and builds more confidence for the next. Focus on new solutions rather than legacy integration initially.
For those coming from pure PM backgrounds, the barrier to entry has never been lower, but understand that AI assistance works best for new projects without accumulated technical debt. For those of us with engineering foundations, AI isn't teaching us new skills—it's removing the obstacles that made personal projects feel impossible, particularly for fresh starts.
The conventional boundaries between Product Management and engineering are dissolving, but context matters enormously. AI has made it possible for technical Product Managers to be true product builders again for new initiatives, not just requirements gatherers. The question isn't whether this trend will continue. It's whether you'll embrace it where it works best.
The future of Product Management belongs to those who embrace this new reality, where it makes sense—building new solutions that solve real problems for real people, while understanding the very real limitations when dealing with complex, mature systems.
What problems are you itching to solve with a fresh start? What's stopping you from building the solution yourself?
The AI is waiting. The problems are real. The context matters enormously. The only question is: are you ready to break the monotony and start building where AI actually helps?
This represents my personal journey with AI-assisted development and doesn't reflect any professional affiliations. What community problems are you passionate about solving? I'd love to hear about your own journeys from idea to implementation.
More about our Bylaw project:
Part 1: "Teaching AI to Understand Bylaws: Stouffville’s Civic Tech Experiment"
Part 2: "Building in Public: The Technical Journey of Teaching AI to Read 50 Years of Bylaws"
Part 3: "When AI Builds AI: How We Used Robots to Teach Robots About Bylaws"