Tony Cooper
Principal, Cooper Agency
Last week I promised to tell you why SQLite is probably enough for your business.
Short answer: it handles 304 database tables, complete financial tracking, and a boutique client roster without breaking a sweat.
You’d think running a business platform would require PostgreSQL, or MySQL at minimum. Enterprise infrastructure for professional operations. But SQLite - a single-file database - powers my entire SEO intelligence platform. Zero configuration. Zero maintenance. Zero performance issues.
For most businesses, it’s genuinely enough.
Database choice isn’t the interesting question anymore. Everyone’s solved that problem. Boring technology works.
The interesting question is this: What can’t AI do?
Because while everyone’s selling AI as magic, I’ve spent six months working with Claude Code every single day, building an entire business platform at 40-80x typical velocity. And I can tell you exactly where AI stops and where human expertise becomes the thing that actually matters.
This is the honest list.
SQLite is Probably Enough (The Quick Answer)
Let me get the promised SQLite discussion out of the way.
My platform runs on SQLite locally. Here’s what it actually handles:
- 304 database tables across the entire system
- Complete financial tracking with years of invoices
- A boutique client roster with complete SEO intelligence
- Audit runs with thousands of data points
- Financial tracking across multiple accounts
- Client success documents and work I’ve delivered
- Marketing campaigns and email logs
- Task management and time entries
Performance issues: None.
Maintenance overhead: Zero.
Configuration complexity: Literally none. It’s a single file.
When you actually need PostgreSQL:
- Multiple users writing simultaneously (I’m the only operator)
- Distributed systems requiring replication (a single VPS works fine for me)
- Data volumes in hundreds of gigabytes (I’m nowhere close)
- Complex geospatial queries (I don’t need them)
- Full-text search at massive scale (SQLite’s FTS works for my needs)
The reality: SQLite handles “enterprise” workloads that most businesses never approach. The cargo cult around “needing PostgreSQL” is developers solving problems they don’t have.
That’s the SQLite answer. Simple databases work.
But here’s what I actually want to talk about: What AI can’t do.
What AI Actually Can’t Do (The Honest List)
I work with Claude Code every single day. Six months of intensive context-driven development. I’ve been building features in hours instead of weeks. Genuine 40-80x velocity on technical implementation.
And here’s what it absolutely cannot do.
It can’t make strategic business decisions.
Claude Code can build a feature in 45 minutes. It can spin up a complete ecommerce website in an afternoon. It cannot tell me if that feature is worth building.
When I say “should I add client health scoring to the dashboard?” - the system will enthusiastically start implementing. It won’t ask whether this wins clients or serves existing ones. It won’t calculate the ROI vs spending that time on proposals. It won’t question if this is the highest-leverage activity I could be doing right now.
That strategic pause requires human judgment.
In November, I discovered I’d been building features when I should have been growing the client roster. The system built everything I asked for brilliantly. It never once questioned whether I was building the right things. The standard I keep coming back to is one Michelin star forever — not chasing scale, but making everything you serve worth coming back for.
It can’t maintain client relationships.
Claude Code can draft professional emails. It cannot build trust over a coffee conversation.
Real client relationships require reading body language during discovery calls. I need to understand unstated concerns. I need to know when to push back on bad ideas. I need to recognise when “yes” actually means “I’m worried.” I need to build rapport that survives mistakes.
I can use the system to prepare for client meetings. I cannot use it to conduct them with genuine human connection.
26 years of client relationship experience beats any training data.
It can’t know when the pattern is wrong.
This is the critical limitation.
The model works through pattern recognition. It’s seen Django implementations thousands of times. When patterns match proven examples, velocity is extraordinary.
But what happens when the proven pattern shouldn’t apply?
The model will suggest patterns it’s seen work many times before. Standard approaches. Proven implementations. And sometimes those patterns are completely wrong for your specific use case - but the model can’t recognise that the familiar pattern is the wrong solution.
That recognition requires understanding your specific constraints, your specific business context, your specific requirements in ways that pattern matching simply can’t capture.
AI gives you speed. Expertise gives you direction. You need both — and the bridge between them is learning to situate your AI so it works within your constraints, not against them.
It can’t recognise when to question vs execute.
I’ve developed a concept I call “strategic pause triggers” - moments when you should stop and think rather than immediately build.
Client meeting preparation. What’s the objective beyond the meeting itself? New feature requests. Does this win or serve clients, or am I avoiding harder work? Time allocation decisions. Is this the highest-leverage activity? Technical architecture choices. Does boring technology actually apply here?
Claude Code will execute brilliantly on any request. It won’t pause and ask if I should be requesting something different.
That pause - the “wait, is this the right problem?” moment - requires human judgment that I developed over years of expensive mistakes.
It can’t apply 26 years of hard-won expertise.
The model has 20 years of Django training data. I have 26 years of internet business experience.
Those are different things.
Training data shows what worked in documented scenarios. Experience shows what failed in ways nobody documented. Why clients say yes when they mean no. Which shortcuts create long-term problems. When breaking best practices is actually correct. How business constraints override technical perfection.
That pattern recognition isn’t in AI training data. It’s developed through years of expensive lessons that taught me when “technically possible” doesn’t mean “strategically wise.”
The Dr. Ford Reality
In my collaboration with Claude Code, I’ve developed two distinct modes.
Tactical Execution Mode: Build this feature, fix this bug, implement this integration. The system works at 40-80x typical velocity. Extraordinary.
Strategic Advisory Mode (Dr. Ford): Apply multi-lens analysis, question assumptions, challenge direction, identify blind spots. The system provides frameworks, but human judgment makes the call.
Here’s the pattern I discovered: When I skip strategic thinking and jump straight to tactical execution, I build the wrong things efficiently.
Six months of git commits showed this evolution clearly. Early on: “Build X” then Claude builds X brilliantly, then I realise X wasn’t the right solution.
Now: Strategic pause, then multi-lens analysis, then I validate direction, and only THEN do I build with AI velocity.
The velocity only matters when you’re building the right thing.
When Boring Technology is Actually Wrong
Everyone’s excited about my “boring technology wins” philosophy. Django, SQLite, Astro, Tailwind - proven patterns that AI knows deeply.
But here’s the question worth asking: When is boring technology the wrong choice?
Real-time collaborative editing. SQLite’s single-writer limitation actually matters there. Actual millions of users - not the imagined millions everyone claims they’ll get, but the real ones. Complex geospatial analysis where specialised databases exist for good reasons. High-frequency trading systems where microseconds matter. Distributed teams editing simultaneously where PostgreSQL’s concurrency model wins.
The judgment required: I need to know when my constraints actually violate boring technology’s strengths.
AI can tell you the tradeoffs. It cannot tell you which tradeoffs matter for your specific situation with your specific constraints with your specific goals.
That requires human expertise applied to business reality.
The Velocity Multiplier Works Both Ways
Here’s the uncomfortable truth about AI assistance.
If you have excellent judgment: AI multiplies that into extraordinary results.
If you have poor judgment: AI helps you build the wrong things very efficiently.
40x velocity on correct strategic direction = competitive advantage.
40x velocity on poor strategic direction = expensive mistakes delivered quickly.
The multiplier is neutral. Your judgment determines the outcome.
What This Actually Means
Everyone’s focused on AI capabilities. Asking “what can AI do?”
The critical question is: “What must humans still do?”
Based on six months of intensive context-driven development, here’s what I’ve learned.
AI excels at pattern recognition and application. It implements proven solutions quickly. It executes technically at 40-80x speed. It generates code following established patterns. It researches and synthesises documentation.
Humans are required for strategic business judgment. I’m required for client relationship building. I’m required for knowing when patterns don’t apply. I’m required for recognising “wrong problem, right solution” scenarios. I’m required for applying hard-won experience to novel situations. I’m required for making the strategic pause before building.
The winning combination: Human strategic judgment + AI tactical execution.
Not “AI replaces expertise.” Not “AI is just autocomplete.”
AI amplifies expertise. But only if expertise exists.
The 26-Year Advantage
Why does my context-driven development work so well?
Not because I’m an exceptional coder. I can read code, I can modify intelligently, I can understand what’s happening. But I’m not writing complex systems from scratch.
It works because I’ve made every expensive business mistake over 26 years.
I know what clients actually want vs what they ask for. I recognise scope creep before it starts. I understand which shortcuts create technical debt and which are pragmatic. I can tell when “add this feature” means “I’m avoiding harder work.”
That pattern recognition isn’t in AI training data. It’s developed through years of expensive lessons.
AI gives me 40x execution speed. Experience gives me strategic direction. Together: sustainable competitive advantage.
Separately: expensive mistakes delivered efficiently, or good judgment executed slowly.
The Honest Assessment
If someone asks “should I use AI for my business?” - here’s my honest answer.
Yes, if you have strong strategic judgment to direct it. If you can recognise when to pause vs execute. If you know when proven patterns don’t apply. If you have expertise worth amplifying.
No, if you’re hoping AI will provide business judgment. If you expect it to replace client relationships. If you think pattern recognition equals strategic thinking. If you want AI to tell you what to build.
AI is the most powerful tool I’ve ever used. It’s not a replacement for expertise. It’s an amplifier.
And amplifiers make everything louder - including mistakes.
Try This Instead
Next time you’re about to ask AI to build something, pause.
Ask yourself: Is this the right problem to solve? What’s the business objective beyond task completion? Does this win or serve clients, or am I avoiding harder work? Am I building, or am I displacement-activity building? What would 26 years of experience say about this?
Only after strategic validation - then I apply AI velocity.
You’ll know it’s working when you build less but achieve more.
Because the right thing built in hours beats the wrong thing built in minutes.
P.S. - Next Week: The language problem nobody told you about. How metaphors solve AI’s strategic drift problem - why “The Palantir” and “The Monday Service” aren’t whimsy, they’re infrastructure for thought.
P.P.S. - The Strategic Framework: Want the multi-lens decision framework I use before AI builds anything? The checklist that determines if you’re solving the right problem? Reply with “STRATEGIC” and I’ll send you the systematic pause protocol that prevents building wrong things efficiently.
Tony Cooper We Build Stores - Where 26 Years of Experience Delivers in One Hour What 26 Hours of Not Knowing Cannot
tony.cooper@webuildstores.co.uk 07963 242210
This Week: SQLite is probably enough. AI definitely isn’t. Human judgment + AI velocity = competitive advantage. Either alone is insufficient.
Questions or feedback on this issue? Write to us
More from the Cooper Agency newsletter
The Pretty Frontend, The Ugly Backend
Django has 20 years of training data. SQLite is simple enough that AI understands it perfectly. Astro components are predictable. Tailwind utilities compose naturally. That's why boring technology wins in the AI era.
When AI Isn't Enough: What Still Requires Human Judgment
SQLite handles 304 tables and complete financial tracking without complaints. But AI can't make strategic decisions, maintain client relationships, or know when boring technology is wrong. Here's what actually requires human expertise.
Why AI Needs Metaphors
Last week I explained what AI can't do. This week: how metaphors solve the strategic drift problem. The Palantír, The Conductor, The Monday Service - this is infrastructure for thought.