Most people get stuck because nobody defined what “solved” actually looks like before the work started. Here’s a five-step method that works whether you’re going to Mars or growing a newsletter.
1. Work backwards
Before you touch the problem, write the press release. This is Amazon’s PR/FAQ method - describe the finished result as if it already shipped and worked. Headline, the problem in one sentence, the solution, a quote from a happy customer.
If you can’t write a headline a stranger would care about, you haven’t found the problem. You’ve found a feature. “I want more subscribers” isn’t a destination. “This newsletter helps managers avoid the mistakes that get people fired” is. This is also where you discover your why and if the problem is worth solving.
2. Define first principles and assumptions
Separate what’s physically true from what you just believe. Strip the problem to what actually causes the outcome, not what everyone else conventionally does.
Constraints are limits you can’t argue with. Assumptions are beliefs you haven’t tested, and depending on your experience, some or many of them are wrong. Break the problem into pieces that don’t overlap and don’t leave gaps, so each piece is a smaller, attackable problem on its own.
For each assumption, find who actually owns the ground truth, not who’s convenient to ask. Someone who’s made 500 bad hires knows more about hiring than a founder guessing alongside you.
3. Define multiple solutions and their tradeoffs
Don’t be attached to your first idea. Generate at least three real options before choosing. For each option, write down the pros and cons of each.
Lay the options side by side. The highest-leverage one is rarely the flashiest one.
4. Define the open questions and the smallest next step to start executing
You won’t have full information. List what you don’t know, then find the cheapest, fastest test for each unknown, not the most thorough one.
The best way to know what you don’t know is to start doing. Find the smallest next step to start building your momentum.
5. Measure direction without over-correcting
Borrow from control systems here. A well-tuned system reacts to the trend, not the last data point. Set the reaction too sensitive and it oscillates, one bad week and you scrap the strategy, one good post and you overinvest in a fluke.
This is also where the struggle does real work. Research on “productive failure,” from learning scientist Manu Kapur at ETH Zurich, found that people who wrestle with a problem before getting the answer build deeper, more transferable understanding than people who get taught the solution up front.
The discomfort of not knowing yet isn’t a sign you’re doing it wrong. It’s the part of the process building the judgment you’ll use on the next problem.
The through-line: define the destination before the plan, separate what’s true from what you assume, generate real options, test cheaply, and trust the struggle to do its job.
Appendix
Prompt You Can Use
# Copy-paste prompt: apply this framework to your own problem
Fill in the brackets and paste this into Claude (or any AI assistant) to run your problem through the 5-step framework from this article.
---
I want to work through a problem using a 5-step systematic framework. Here's my problem:
**What I'm trying to solve:** [describe your problem in 1-2 sentences]
**Why it matters / what happens if I don't solve it:** [one sentence]
**What I already know or have tried:** [list anything relevant, or write "nothing yet"]
Walk me through all 5 steps, one at a time, and ask me questions where you need more information instead of guessing:
1. **Work backwards.** Help me write a one-sentence "press release" headline for what success looks like, as if it already happened. Then draft an external FAQ (questions a real person affected by this would ask) and an internal FAQ (the hardest questions I'd get from someone skeptical of my plan, including why this hasn't been solved already and what could kill it).
2. **First principles and assumptions.** Separate what's actually a fixed constraint (physically or structurally true, I can't change it) from what's just an assumption I'm carrying (a belief I haven't tested). Push me to name at least one assumption that, if wrong, would change my whole approach.
3. **Multiple solutions and tradeoffs.** Generate at least 3 realistic solutions, each with a clear pro and con, written as short paragraphs (not a table). Then pick the one you'd recommend and explain why in a few sentences, including why you ruled out the others.
4. **Open questions and the smallest step.** List what I genuinely don't know yet. For the most important unknown, tell me the cheapest, fastest way to test it, not the most thorough way. I want a real next action I could take this week.
5. **Measure direction without over-correcting.** Tell me what to track, and over what time window, so I don't overreact to a single good or bad data point. Remind me that the discomfort of not knowing yet is part of the process, not a sign I'm doing it wrong.
Keep each step concise. Ask me clarifying questions before step 1 if my problem statement is too vague to work with.