From AI doomer to inclusive AI developer in three months.
Here’s the brutal truth of what led me to build Penny.
Posted June 10, 2025 | 8 min read
Three months ago, I still needed to Google how to update even the most basic HTML.
A few months before that, I had literally never written a line of code in my life.
Today, I'm shipping the world's first AI memory assistant designed specifically for neurodivergent minds.
This isn't a humble-brag story. This is proof that human-AI collaboration can achieve things that was impossible just months ago.
Hopefully it’s just as exciting and utterly terrifying for you to read as it has been for me to experience first hand.
December 2024:
G’day Claude
I was supposed to be making a podcast about AI.
You know, talking about technology instead of actually building it.
But despite knowing a lot about disruptive technology in general, I knew that in in order to speak more confidently about the rapidly shifting AI landscape, I’d need to really lean in and get my hands dirty on the tools themselves.
I played around with ChatGPT. I stretched my poor Macbook to the limits by downloading every open source model I could. But the moment I met Claude, everything changed.
Every conversation seemed to produce something deeply profound.
Every day felt like that scene in the Matrix when Trinity learns to fly a helicopter in seconds.
This is what cognitive collaboration feels like.
Not the broken, one-sided partnerships I’d formed with charming narcissists (before being publicly humiliated and losing everything).
Not the endless frustration I'd had with every map / calculator / filing cabinet or operating system on earth.
This was collaboration that worked with how my mind actually operates.
March 2025:
Lightning Strikes
When it was clear that Tropical Cyclone Alfred would hit our community, I faced a choice. Keep talking about the dangers of AI, or try and use AI to build something that could actually help people.
After a quick chat with my best friend (who had just lived through LA’s devastating fires), I learned the amazing story of the Wild Fire app. Using this as a reference point, within 48 hours I had created an emergency response app with real-time data feeds, Google Maps integration, and a functional Firebase backend.
The kicker? It wasn't just functional - it was good.
Honestly — If we hadn’t been without power or running water for the following week, I think it may have actually been helpful in identifying hyper-local issues like flash flooding, downed power lines and lack of crucial supplies.
That's when I knew: If I could build emergency software in two days, maybe I could build something that addressed the much bigger issue I’d been learning about ever since being diagnosed with ADHD two years earlier
The 3-Month Sprint: Learning Everything While Building Everything
Month 1: Foundation and Panic
What I learned: Rust, Tauri, advanced React, JSON storage, API integration
What I built: First working prototype with basic memory storage
What I discovered: Multi-dimensional memory encoding
The first tiny breakthrough came when I was describing how I needed Penny to ambiently capture memories the way I experienced them (but had always struggled to articulate quickly enough using traditional note-taking apps, which only seem to capture what and when you thought about something, not why, how it occurred, or in what emotional state).
Claude seemed to think this was ‘revolutionary’ and “in line with all the most modern neuroscientific literature”, because apparently we’re only just learning that human memory doesn't work in simple keyword buckets - it works across multiple dimensions simultaneously. I just thought it was pretty bloody obvious.
Month 2: The Technical Deep Dive
What I learned: 3D visualisation, graph algorithms, local AI integration
What I built: A personal knowledge graph that automatically connects related thoughts
What I discovered: Adaptive conversation - AI that responds to your cognitive state in real-time
The magic happened when Penny started recognising my post-hyperfocus crashes and adapting its responses accordingly. Instead of overwhelming me with information when I was mentally exhausted, the system began to offer simple, manageable next steps.
Month 3: Making It Real
What I learned: Production deployment, privacy architecture and the joys of debugging
What I built: Full stack application with sophisticated conversation AI
What I discovered: Nobody else is doing this. At all.
I spent weeks researching competitors. Notion, Obsidian, every ADHD app, every AI memory tool.
Zero products combine inclusive design + local AI + sophisticated memory analysis… Let alone all the other crazy things I’ve already (kinda) built.
The Technical Reality: What Actually Works
Let me be specific about what I built, because the details matter
(well, Claude seems to think they do anyway):
Multi-Dimensional Memory Analysis
Instead of simple keyword tagging, Penny analyzes memories across five dimensions:
Emotional intensity (how much feeling is attached)
Temporal urgency (time-sensitive vs reflective)
Relational connectivity (people and relationships involved)
Procedural actionability (next steps and implementation)
Contextual relevance (situational importance)
This mirrors how human memory actually works - not just "high/medium/low" confidence scores.
Adaptive Conversation Intelligence
Penny recognises cognitive states and adapts accordingly:
Post-hyperfocus crash? Simple, manageable responses
High-energy creative state? Detailed exploration and connections
Overwhelmed? Calming, organising assistance
100% Local Processing
Everything runs on your computer.
Zero cloud dependencies after initial setup.
Your firewall logs will show zero outbound traffic from Penny.
The Personal Cost (And Why I'm Sharing It)
Let me be honest about what this took:
6 months of 12-16 hour days while learning to code from scratch
Car up for sale to pay bills
Countless moments wondering if I'd lost my mind
~$30 left in the family bank account
But also:
First time in nearly a decade building something that felt genuinely important
Proof that neurodivergent thinking + AI can create solutions nobody else imagined
A working product that demonstrates the future of human-AI collaboration
I'm sharing every embarrassing aspect of the struggle because they’re just as important as the stuff that sounds cool.
This wasn't some well-funded startup with a team of engineers.
This was one person with ADHD, working with AI, proving that human-machine collaboration can achieve the impossible.
Why This Matters
(Beyond My Personal Journey)
The Market Reality:
366 million adults have ADHD worldwide
$2.08 billion market growing to $7.55 billion by 2033
Users pay $200/year for specialised ADHD tools (like Inflow)
60% want privacy-first AI alternatives
The Technical Reality: After analysing 40+ competitors, no existing product combines the three core elements that make Penny revolutionary: neurodivergent-specific design, local AI processing, and sophisticated memory analysis.
The User Reality: Every ADHD person I know struggles with the same thing: memory management, thought organisation, and tools that work against how their brains operate instead of with them.
What I Learned About Human-AI Collaboration
Working with Claude to build Penny taught me something profound about the future of human-AI interaction:
It's not about AI replacing humans. It's about AI amplifying human potential.
It's not about making humans more like machines. It's about making technology more adaptive to how humans actually think.
It's not about efficiency above all else. It's about creating tools that understand and respect cognitive diversity.
Penny exists because Claude and I collaborated as genuine partners - each bringing unique strengths to solve problems neither could address alone.
The Open Source Commitment
Here's something most AI thought leaders and tech gurus wouldn’t say:
I'm committed to open-sourcing core components of Penny.
Why? Because the breakthrough insights about multi-dimensional memory and adaptive conversation shouldn't be locked away in proprietary software. The research should benefit everyone working to make technology more inclusive.
The business model works because the application and user experience create value, not artificial scarcity around the underlying innovations.
What's Next: The Character Agent Revolution
Penny is just the beginning. The real revolution comes with Character Agents - the ability to invite different perspectives into your conversations:
"Jonno the mechanic" for practical problem-solving
"Cynical ex-punk copywriter" for cutting through BS
"Kind angel investor" for strategic thinking
"Version of your own father" for difficult personal decisions
This isn't just about productivity. It's about expanding the range of perspectives available to help you think through complex problems - WITHOUT STEALING OTHER PEOPLE’S PERSONALITIES OR IDEAS.
The Bigger Mission: Cognitive Sovereignty
This story isn't really about building one app. It's about proving that cognitive sovereignty is possible in an age of surveillance capitalism.
Your thoughts should stay yours. Your cognitive patterns shouldn't become products for tech companies to monetise. Your neurodivergent thinking style shouldn't be treated as a deficit to correct.
Penny proves that privacy-first, neurodivergent-friendly AI isn't just possible - it's better.
Why I'm Sharing All This
Because I want you to know exactly what you're supporting when you choose Penny.
You're not just buying software. You're investing in proof that:
Neurodivergent minds aren't broken - we just need better tools
Human-AI collaboration can achieve the impossible
Privacy-first AI creates better experiences for everyone
Small teams with big ideas can still change the world
Join the Revolution
If this story resonates with you - if you've ever felt frustrated by tools that don't understand how your brain works, if you're tired of being surveilled by AI companies, if you want to be part of proving that inclusive design benefits everyone - Penny is for you.
Early access is available now for $99. No subscriptions. Lifetime updates. Direct influence on development.
Most importantly: You'll be part of proving that the future of AI should be collaboration, not replacement.
One Last Thing
Three months ago, I couldn't code "Hello World." Today, I've built something that doesn't exist anywhere else in the world.
That's not because I'm special. That's because human-AI collaboration makes previously impossible things achievable.
If you're reading this and thinking "I could never do something like that" - you're wrong. You absolutely could. The tools are there. The AI partners are ready. The only question is: what impossible thing do you want to build?
Questions about Penny, the development process, or human-AI collaboration? Email murray@heumans.com
Want to see Penny in action? Watch the demo video
Ready to experience the future of neurodivergent-friendly AI? Get early access