Click any tag below to further narrow down your results
Links
Sean Goedecke argues that people who hate big tech jobs often dislike the organizational mess itself, which makes it hard to engage with. He says the real meaning comes from the scale of impact: small improvements to widely used enterprise software reach millions of people.
- Many engineers' unhappiness comes less from dysfunction blocking their work and more from a compulsive disgust with messiness, which makes even thinking about the dysfunction feel tainting.
- Pretending to care about corporate rituals like OKRs and hype can be alienating and can slowly erode your sense that anything matters.
- Improving enterprise software that millions use is valuable, since every bug fix or feature saves users cumulative months or years of time, and slowing software decay is worthwhile.
Sean Goedecke built AutoDeck, an app that generates an endless feed of spaced repetition flashcards on any topic with an LLM, adjusting difficulty from the user's responses. He uses it to explain why he skipped the chat interface and what building it with Codex was like.
- Generating flashcards is the hard part of spaced repetition for new topics, and LLMs make it possible to create decks on demand.
- Parallel generation produced duplicate cards, and waiting for one batch was slow, so he streamed cards in XML chunks as each was generated, since JSON can't be parsed mid-stream.
- Codex did most of the coding but made bad calls he had to fix, such as over-engineering a background job system and tracking cards by index instead of ID; he found Claude Code slower and needing more intervention.
- Paid inference costs pushed him to charge for the app, which he thinks made him build something more useful than a free side project would have been.
Sean Goedecke argues that database rows are effectively global variables, which clashes with the common advice to avoid globals. He tests five explanations for why databases get a pass and concludes that explicitness and disciplined usage, not anything inherent to databases, are what make them tolerable.
- Databases share the two core problems of global variables: any part of a program can read or write any record, and different parts become implicitly coupled through shared state.
- Many proposed reasons databases are better fail under scrutiny. ACID guarantees don't fix the core problem, and wrapping a global in a mutex doesn't make it acceptable.
- The two reasons he finds convincing are that database queries are explicit (they need a connection and a query call) and that teams use databases for a narrow set of purposes. Both are conventions, not inherent traits, so a codebase could apply them to globals too.
Sean Goedecke argues that many engineering clashes, like open-source purists versus big-company engineers, come from people doing fundamentally different kinds of work rather than one side being less skilled. Pure engineering is open-ended craft aimed at the best technical solution, while impure engineering is about shipping good-enough solutions to someone else's deadline.
- Pure engineering resembles art or research and is driven by aesthetics, while impure engineering resembles plumbing or construction and must compromise to meet business needs and schedules.
- Pure engineering had more room at large tech companies in the 2010s, when hype-driven hiring funded it as a kind of developer marketing; that funding has shrunk as companies now need to make money.
- Skilled pure engineers often struggle with impure work and vice versa, and the Casey Muratori versus Windows Terminal dispute and George Hotz's Twitter stint are used to show that impure work is harder than it looks.
This is an index page listing every post Sean Goedecke has tagged "explainers," covering topics like LLMs, DeepSeek, Model Context Protocol, blockchain, diffusion models, and interpretability. It's a jumping-off point rather than an article with a single argument.
- The index is dominated by AI topics: how LLMs and ChatGPT work, DeepSeek's cost and reasoning work, MCP, and interpretability.
- Blockchain shows up several times, including a Stripe "Tempo" FAQ and beginner-level primers.
- A few posts fall outside AI and blockchain, such as how to read code, code review mistakes, and writing technical explanations.
Sean Goedecke argues that LLMs don't have fixed goals and their internal workings can't yield the kind of motive-level understanding needed to guarantee safety, so "solving alignment" for them is incoherent. He concludes that mechanistic interpretability, though useful, can't deliver the confidence required to prevent rogue superintelligent behavior.
- LLMs are trained to imitate human text, so they lack a single fixed goal that could be formally specified, unlike chess engines or the classic "blind optimizer" paperclip model.
- Mechanistic interpretability can trace calculations but not motive, and it can only label behaviors we've already elicited, leaving many failure modes unobserved.
- Feature labels are guesses that can't be distinguished from alternatives (the "kindness unless 2030" problem), and the combinatorial complexity per token is too large for humans to check.
- Current mitigations like chain-of-thought monitoring and sandboxing are useful but judged insufficient against a superintelligence.
Sean Goedecke argues that very smart people often write inelegant code or solve problems by brute force, because raw working memory lets them skip the clever abstractions everyone else needs. He extends the idea to LLMs, suggesting their tolerance for long, messy proof chains could make human creativity more valuable.
- Von Neumann solved the fly-and-cyclists puzzle instantly by summing the infinite series, skipping the elegant shortcut most people look for.
- For software engineers, the key trait is working memory, and engineers with large working memory can tolerate sloppy codebases because they can hold the whole system in their heads.
- LLMs can parse hundreds of proof steps as easily as ten, which may make elegant abstractions less necessary for them, and that could weaken their ability to generate new conjectures.
Sean Goedecke argues that using AI well is a real, durable skill, not something you can pick up instantly once models improve. He lists the parts of that skill, from supplying context to reviewing AI-written code, and says engineers who avoid AI entirely are falling behind.
- Good AI users supply context and explain the reasons behind a goal, because agents guess badly when they're missing information.
- Knowing what models can and can't do is durable: new models improve most on tasks older ones already handled well, so that sense carries over.
- AI-generated code looks polished but makes different mistakes than humans do, such as adding unnecessary mutexes for trivial edge cases, so reviewers need a separate skill set.
- Avoiding AI completely means being left behind, though the author warns against going all-in, since LLMs reward expertise that comes from doing the work by hand.