78 Solutions is a mindset for deliberate practice in software — choosing the right challenges, solving them well, and building momentum without burning out.
About 78 SolutionsSome problems aren’t worth solving. This method helps you spot the ones that matter — the ones with long-term payoff and deep learning.
Write, test, improve. Then do it again. You’ll build skill through repetition and code reuse, turning your codebase — and your career — into a perpetual training ground.
Every solution is a seed. Let your past solutions guide (or help solve) your future problems, and let your systems evolve with you over time.
Each week, you choose 1.5 problems to solve. Perhaps you will choose one problem worth solving and one small improvement to your toolkit. Or perhaps you will solve three problems over two weeks. The goal isn’t just to ship — it’s to learn while you build, and to build things worth keeping.
Define a problem clearly and tie it to a bigger purpose. Even small problems can unlock big growth if chosen wisely.
Work with focus, using habits that sharpen your thinking — like writing tests, reflecting on what’s hard, or reusing past solutions.
Document your progress. Save what works. Each solution becomes a tool in your belt — for next week, next project, or your future self.
Recent articles and submissions from contributors — or technically, just one (for now). Want to publish something on 78solutions.com? Visit the contributors page to learn how.
Reusable software is one of my personal principles because it delivers better software experiences to customers and provides better business value for me. The FitnessTracker project created the opportunity to lift Rust modules out of FitnessTracker, drop them into a new project, write custom React for the front end and add some new modules to create a totally new product. This article tells that story.
For years I used to believe that every problem was worth experiencing so you could learn its lessons. And I was wrong - some problems absolutely suck and are worth doing everything in your power to avoid. Heart problems fall into the latter.
While generative AI has both good and bad traits, its potential as a security tool is perhaps its best trait. In this case, Claude Code not only helps an ML engineer uncover a vulnerability but also helps him alert the right people. This article is an interesting counterpoint to the constant news about open source projects getting inundated with low quality AI contributions.
Sometimes when an open source project goes viral, domain squatters will jump in and buy up domains around that project. It is potentially lucrative and especially in the age of generative AI, it takes virtually no time or effort. An AI assistant called Nanoclaw is caught up in this right now. And while domain squatting is not usually that interesting, in this case the sheer amount of power AI assistants can get makes this a potentially interesting attack surface.
OpenAI has released two permissively licensed models, gpt-oss-120b and gpt-oss-20b, designed for agentic workflows and full transparency. Available on Hugging Face and GitHub, they offer a rare glimpse into OpenAI’s approach to open-source development.