When we designed Source Code Accelerator, the obvious move was to open with the tools. Show people agents, show them code generation, get them building in week one. We put problem framing and systems thinking in week three instead, before a single model touches a single line of code. Here’s why.
Fast answers to the wrong question
AI has made producing a solution almost free. Ask for an app, a script or a data pipeline and you’ll get one in minutes. What it hasn’t made cheaper is deciding what the solution should be.
That gap is where most failed projects live. A team builds the thing it was asked for, ships it quickly, and discovers it solved a symptom. With AI, you reach that moment faster and with more confidence, because the output looks finished.
So week three starts with one habit: frame the problem before you reach for a solution. Who has it? What do they do today? What would have to be true for it to go away? What does “solved” look like, and how would you know?
Complicated is not the same as complex
The second idea we teach is the difference between complicated and complex problems.
AI tools are very good at complicated problems. They are much weaker on complex ones, because what matters isn’t in the prompt. It’s in how people, incentives and systems respond once the thing exists. Developers who can tell the two apart know when to trust a quick build and when to slow down.
Second- and third-order consequences
The third idea is looking past the first effect of a change. Take an app that matches graduates to internships faster:
- First orderGraduates are matched to internships faster.
- Second orderEmployers screen more applicants and lean harder on automated filters.
- Third orderCandidates without polished CVs, often the ones we most want to reach, are filtered out before a person ever sees them.
None of that shows up in a code review. All of it shows up in people’s lives. In week three, participants map these chains for their own project ideas before they build anything.
Why this comes before the AI weeks
The second half of Source Code Accelerator covers agentic development, code review and security, and building and evaluating with models. Those weeks go better when people arrive knowing what they are trying to change and what could go wrong.
It also changes how they use the tools. A developer who has framed the problem writes clearer prompts and specs. They review generated code against a purpose, not just against whether it runs. And they can explain their choices to a manager, a client or a community who will live with the result.
The most durable skill was never the syntax. It was the thinking that came before it.
We spent ten years teaching people to code. What we’ve learned is that the most durable skill was never the syntax. It was the thinking that came before it. AI hasn’t changed that. It has made it more important.
