Most personal projects die quietly, in a private repo, the moment the first approach turns out to be wrong. Let's Fail Project is my attempt to build differently: a living, public GitHub ecosystem where I ship experimental software in the open, document what breaks, and let the failures be as visible as the wins. The public roadmap is the record of that — not a highlight reel, a working log.
The name is not self-deprecating. It's the point. If failure is going to happen anyway — and in any project pushing past what you already know how to build, it will — the only real engineering decision is how it happens: how frequently we iterate, how much it costs, and how quickly we learn.
Fail Often, Fail Cheap, Fail Fast
These three constraints aren't independent slogans; they compound.
Fail often is about execution frequency. The more iterations you run, the faster you build intuition and expose hidden assumptions. This maps directly to the Build-Measure-Learn feedback loop, where the number of cycles is the key engine of progress. By keeping the scope of each experiment minimal, you can afford to pivot frequently without losing momentum.
Fail cheap means the cost of being wrong should be small relative to what you'd spend proving yourself right the slow way. This is the core argument of Eric Ries's The Lean Startup, which formalized the Build-Measure-Learn loop specifically to limit how much time and money go into an idea before it's tested against reality (TechTarget; Wikipedia). Ries built on Steve Blank's customer development work, which pushed the same logic further upstream: don't perfect a solution before you've confirmed the problem is real. The underlying goal in both is avoiding the sunk-cost trap — the tendency to keep funding a bad direction just because you've already funded it.
Fail fast is what makes the first two matter at all. A cheap, frequent failure that takes six months to surface has already stopped being cheap. John Boyd's OODA loop — Observe, Orient, Decide, Act, developed for fighter pilots who needed to out-decide an opponent in real time — makes the case that the speed of the iteration, not the accuracy of any single decision, is the actual competitive advantage. W. Edwards Deming's PDCA cycle (Plan-Do-Check-Act), which rebuilt Japanese manufacturing quality after WWII, is the slower, more methodical sibling of the same idea: short, disciplined loops beat long, confident ones.
Why This Requires Psychological Safety, Not Just Process
None of the above works if failing visibly feels dangerous. Google's Project Aristotle, its multi-year internal study of what actually makes teams effective, found psychological safety to be the single strongest predictor of team performance — stronger than talent, clearer goals, or better leadership. Teams that could acknowledge failure without it becoming personal were the ones that actually learned from it and improved.
That finding scales down to a team of one. Building in public — publishing a roadmap where abandoned experiments sit next to shipped ones — is a deliberate attempt to manufacture that same safety externally: if the failure is already public and already framed as expected, there's nothing left to hide, and nothing left to lose by trying the next thing.
What This Looks Like in Practice
Let's Fail Project isn't a philosophy in the abstract — it's the operating model behind everything currently in the ecosystem:
- Pinky and the Brain — a cloud-native Deep Agent built on LangGraph.js, grounded by a custom RAG pipeline, deployed to AWS via Terraform, and reachable through an IDE protocol, REST/SSE, and MCP. Each of those integration surfaces was its own small, cheap, fast bet on whether a given protocol was worth committing to — not a single monolithic architecture decision made once and defended forever.
- patb-cli — a terminal REPL and Zed IDE bridge for that same agent, built to test a different interaction model without touching the agent's core.
- Monday Money — a lightweight, privacy-focused desktop app for personal finance, small enough in scope that a wrong turn costs a weekend, not a quarter.
Each of these is scoped so a bad bet is recoverable in days, not months — and documented so the bad bets are as visible as the good ones. That's the actual mechanism behind "ship-first": not recklessness, but failure frequent enough, cheap enough, and fast enough that shipping the next attempt is always the easier option.
Sources
- Eric Ries, The Lean Startup (2011); Steve Blank, The Four Steps to the Epiphany — TechTarget: What is fail fast?, Wikipedia: Fail fast (business)
- John Boyd's OODA loop — Wikipedia: OODA loop
- W. Edwards Deming's PDCA / Shewhart cycle — OODA Loop vs PDCA Cycle
- Google's Project Aristotle on psychological safety — Psychological Safety in Software Workplaces: A Systematic Literature Review (arXiv)