Move Fast and Brace Things
Make it fast. Make it cheap. Make it good. Pick three.
Mark Zuckerberg famously implored his developers to "Move fast and break things."
Many early adopters approach AI-assisted development with that same mentality. But what works for a Silicon Valley startup building a college dating app does not translate to the business world. My clients don't want to see the words "break things" anywhere near their systems–and rightfully so!
What if, instead of breaking things, we added braces to make them stronger?
The Software Reliability Paradox
I've used AI extensively to help me read, debug, and write code.
It's difficult to overstate how much it can boost your productivity when used effectively. It's also getting more reliable every day. But that increase in reliability carries with it a paradoxical increase in risk. It's my software reliability paradox on steroids:
As software reliability increases, so too does the degree of harm it has the potential to create.
One of the great risks with AI is that, as it becomes more reliable, we will trust it more. And as we trust it more, we will give it increasing responsibilities. But the more responsibilities we give it, the more we open ourselves up to catastrophe. You read new horror stories every day, like the AI agent that accidentally deleted a company's production data and all its backups.
The understandable reaction from many developers to stories like that is to forswear AI entirely. The biggest problem with that approach?
AI today is the worst it will ever be.
You can't simply ignore it.
It's only going to get faster. It's only going to get cheaper. It's only going to get more capable. It's only going to get more reliable.
The Monster and The Cage
Here's an extended excerpt from my Software Reliability Paradox article:
The Cage represents the reliability of the Software. The Monster is the harm the Software has the potential to create.
Unreliable Software has a small Cage. The Monster need not be very big to break out of the small Cage. In fact, the Cage is so small that as the Monster grows, he can't help but break out of the Cage. If he does break out, though, he's not big enough to do much damage.
Reliable Software has a big Cage. A small Monster will have trouble breaking out of the big Cage. But the big Cage offers plenty of room for the Monster to grow. By the time the Monster is big enough to break out of the big Cage, it's going to be one big Monster.
The AI monster will keep growing.
It's our job to build it a stronger cage.
Mike Wolfe
Paradigm Shift
The key to understanding and making the most of this moment is to realize that this is not the typical evolution of software development.
AI-assisted development is not simply the next higher level of abstraction in programming languages. It unlocks brand new possibilities and challenges. You need to embrace both.
Move fast and brace things.
That's the new tagline for this website. "Move fast" is all about realizing the new possibilities AI unlocks, while "brace things" is my multi-faceted answer to the unique challenges AI development brings.
Key Principles
At my company, we've been leaning hard into AI-assisted development for the past 18 months. In the past 6 months, I've personally written almost no code, while simultaneously shipping more features at higher quality and faster than ever before.
I did that by following the principles below.
I'll write about each principle in more detail in future articles.
Move Fast
- Experiment with low-stakes projects
- Documentation-first development
- Tight feedback loops
- Agent-verifiable outcomes
- Give agent direct access to development data
- Maintain architectural decision records
- Pay for top-tier models (they're worth it)
Brace Things
- Version control everything
- Isolate the development environment
- Test-driven development
- Functional core, imperative shell pattern
- Use the B-BEAR technique
- Write-only automated off-site backups of production data
- Automate deployments AND rollbacks
- Multiple algorithms for critical logic
- Structure over convention for security
NOTE: I wrote the above lists from memory, so I'm sure I left out a few important principles. As with AI itself, I expect these lists to evolve.
