Smorster

7 Programming myths that waste your time

Fireship · 6m 17s · Watch on YouTube · 28 sources

Decision Card

Effort: Zero-setup — a 6-minute watch; the only real work it implies is a 30–60 minute audit of one of your own projects for coverage-number targets, premature abstractions, and over-provisioned infrastructure.

Honest take: The video’s own numbers don’t hold up to its debunking standard — the title promises 7 myths, the narration announces “nine smart ideas” (00:33), and it closes with a sponsor read claiming interactive lessons are “six times more effective than watching video lectures,” a figure that overstates the underlying active-learning research (which found roughly 1.5× lower failure rates, not 6× effectiveness). There’s also inherent tension in an anti-hype message delivered by a channel whose stated goal is covering the latest tech.

Concrete next steps:

  • Boring-tech default — adopt. Before betting on a proprietary or VC-funded platform, check its exit path; read the Fauna End of Life FAQ as a 15-minute case study in early-adopter risk.
  • RUG over strict DRY — adopt. Tolerate duplication until the third repetition, then abstract; zero setup, just a habit change on your next PR.
  • Drop 100%-coverage targets — adopt. Read Martin Fowler’s Test Coverage bliki (10 min) and replace any hard CI coverage gate with a “find untested areas” report.
  • Optimize-on-evidence — adopt. Ship correct code first; profile only when production shows real pain (ongoing practice, no upfront cost).
  • Single VPS before microservices — try. Cost out your current cloud setup against one small VPS for an afternoon; downsize if user counts don’t justify the complexity.
  • Paradigm purism (pure-OOP or pure-FP) — skip. Multi-paradigm mixing is already the mainstream position; no action needed beyond not joining a cult.
  • “AI replaces programmers soon” panic — skip. But do keep reviewing AI-generated diffs line-by-line, since verbosity and scope creep are documented failure modes.

TL;DR

Fireship debunks a set of programmer time-wasters: chasing bleeding-edge tech, paradigm dogma, over-applied Clean Code/DRY, 100% test coverage targets, premature performance and infrastructure optimization, and over-reliance on AI coding tools. The recurring prescription is to default to boring proven technology, abstract only after repetition hurts, and optimize only when production evidence demands it.

Key Points

  • The video frames each myth by its lure, why it’s a trap, and how to avoid it — drawn from the author’s own wasted-effort retrospective. 00:33
  • “Dinosaur” tech dominates real jobs: WordPress/PHP run most web applications, Java the enterprise, SQL most databases, C++ low-level systems — and CTOs won’t replace them soon. 00:54
  • Fauna is the cautionary tale for early adoption: a solid, ex-Twitter-engineered but proprietary VC-funded database whose business failed, stranding early adopters who’d have been fine with boring SQL. 01:39
  • Paradigm “cults” (OOP purists, FP extremists) each teach useful things, but dedicating yourself to one is a waste; his current code blends both. 02:10
  • Clean Code’s core advice is good, but over-applying DRY produces endless wrappers, interfaces, and pointless indirection — more refactoring than feature-building. 03:16
  • Proposed replacement acronym: RUG (“repeat until good”) — duplicate first, extract an abstraction only after repetition becomes painful. 03:28
  • Optimizing for 100% test coverage encourages line-touching tests that catch no real bugs, gives false security, and slows CI; quality of tests beats quantity. 03:52
  • Correctness first; optimize performance only when it’s “painfully obvious that your code sucks in production.” 04:31
  • Facebook-scale serverless/microservices/sharding architecture is unnecessary for small apps — “one small VPS is perfectly fine for my five users.” 04:48
  • AI coding tools are simultaneously the biggest productivity booster and, used improperly (e.g. rubber-stamping verbose Claude 3.7 Sonnet output), the biggest time waster; the antidote is a solid problem-solving foundation. 05:04

Notable Quotes

“Most CTOs are smart enough to know that if it ain’t broke, don’t fix it.” 01:33

“I think a better acronym would be rug, repeat until good.” 03:27

“AI programming tools are both the greatest productivity booster I’ve ever seen in my life, but when used improperly, they can also be the biggest time waster.” 05:21

Verified Claims

  1. PHP still runs most web applications. 00:54 — W3Techs: PHP usage, W3Techs: server-side languages overview. W3Techs (July 2026) puts PHP at ~71.8% of websites with a known server-side language, far ahead of the next contender. Verdict: Confirmed (among sites with a detectable server-side language — an important qualifier the video omits).

  2. Most databases are SQL-based. 01:01 — DB-Engines Ranking, DB-Engines ranking by category. Relational DBMS hold ~73% of total popularity; the top four systems (Oracle, MySQL, SQL Server, PostgreSQL) are all relational. Verdict: Confirmed.

  3. Critical banking systems still run on COBOL. 01:27 — IBM: What Is COBOL?, DXC: Why banks still rely on COBOL-driven mainframes. Widely cited figures: ~95% of ATM transactions and 40%+ of online banking systems touch COBOL, with ~$3T daily commerce flowing through it. Verdict: Confirmed (though the headline percentages trace to older industry surveys).

  4. Java powers 3 billion devices. 01:29 — Skeptics StackExchange: Do 3 billion devices run Java?, Hacker News discussion. The number is a Sun-era marketing figure, unchanged for well over a decade, driven largely by Java ME feature phones and Java Card smartcards. Verdict: Inconclusive — plausible in cumulative terms, but a stale marketing stat, not a current measurement.

  5. Fauna was built by Twitter engineers, was proprietary/VC-funded, and recently shut down its service. 01:39 — InfoQ: Fauna Shutting Down, SiliconANGLE: How former Twitter engineers designed Fauna. Founded by ex-Twitter engineers Evan Weaver and Matt Freels; the company announced in March 2025 it couldn’t raise capital and turned the service off May 30, 2025, though it pledged to open-source the core. Verdict: Confirmed.

  6. 100% test coverage is misleading; high coverage does not equal high quality. 03:40 — Martin Fowler: Test Coverage, Thoughtworks: Are Test Coverage Metrics Overrated?. Fowler: coverage finds untested code but says little about test quality, high numbers are cheap to game, and he’d “be suspicious of anything like 100%.” Verdict: Confirmed.

  7. Claude Sonnet 3.7 is good at code but notoriously verbose and over-engineers. 05:06 — Anthropic: Claude 3.7 Sonnet announcement, 16x: Claude 3.7 vs 3.5 Sonnet for coding. Anthropic reported state-of-the-art SWE-bench coding performance; independent testing documented full-file verbose output that prompting couldn’t suppress and scope creep beyond the original request. Verdict: Confirmed.

  8. (Sponsor claim) Interactive lessons are “six times more effective than watching video lectures.” 05:46 — E-Student: Brilliant.org review, ERIC: Assessing the Impact of Brilliant.org. The underlying active-learning research (a PNAS meta-analysis of 225 STEM studies) found lecture students ~1.5× more likely to fail and roughly six percentage-point exam gains — not a 6× effectiveness multiplier. Verdict: Disputed.

Tools, Papers & Standards Mentioned

Follow-up Questions

  1. Did Fauna’s promised open-source release of its core database and FQL actually ship after the May 2025 shutdown, and is it maintained well enough to matter — or does the cautionary tale end at “proprietary means dead”?
  2. If line coverage is a poor quality metric, how much better is mutation testing (e.g. mutation score) at predicting real defect-catching ability, and what does it cost in CI time at typical repo sizes?
  3. Have Anthropic’s newer models measurably reduced the verbosity/over-engineering behavior documented in Claude 3.7 Sonnet, and is there a benchmark that tracks “scope creep” rather than just task success?

Sources