Hasty Briefs - Daily Tech News Brief

Bilingual

The Two Worlds of Programming
2 months ago
- The article discusses the divide in opinions among developers regarding the use of Large Language Models (LLMs) in coding, attributing it to differing worldviews about the state of the software industry.
- It recounts a personal anecdote from 2000 about encountering a dot-com startup post-bubble, highlighting skepticism versus optimism in the tech industry.
- The author contrasts two groups: those who see the software industry as thriving and LLMs as accelerants, and those who view the industry as in crisis, with LLMs amplifying existing dysfunctions.
- Criticism is levied against current software practices, including poor quality, monopolistic behaviors, and a lack of accountability, suggesting LLMs automate these flawed processes.
- The author announces an upcoming project offering an alternative to subscription-based newsletters, featuring advance essays on software industry evolution.
If you thought the code writing speed was your problem; you have bigger problems
2 months ago
- The VP of Engineering announces AI coding assistants to increase code output by 40%, but the real issue is not writing code faster but identifying the actual bottlenecks in the system.
- Eli Goldratt's Theory of Constraints explains that optimizing a non-bottleneck step doesn't speed up the system but creates more problems, like increased inventory and lead time.
- AI-generated code leads to more PRs but overwhelms review queues, causing delays, rubber-stamped approvals, and decreased code quality.
- The real bottlenecks include unclear requirements, long wait times post-code completion, fear of deploying, lack of post-shipment feedback, and coordination issues.
- Solutions involve mapping the value stream, measuring cycle time, reducing wait states, limiting work in progress, and listening to developers' insights about bottlenecks.
AI won't make you rich. But fixing bugs in AI slopware will
2 months ago
- AI-generated code often lacks optimization, such as sequential API calls instead of concurrent ones.
- Current AI models produce code without essential features like caching, leading to inefficiency.
- Senior developers will be in high demand to fix AI-generated technical debt.
- AI models lack understanding of computational complexity and scalability.
- Companies adopting AI-first engineering face increased costs and performance issues.
- The real skill lies in identifying and mitigating hidden failure modes in code.
- AI influencers promote prompt engineering, but actual coding skills remain crucial.
- AI-generated slopware is creating unprecedented levels of technical debt.
Spite Driven Development
2 months ago
- Spite Driven Development (SDD) is a highly productive but destructive methodology.
- Originates from anger towards competitors or existing tools, leading to rapid coding.
- Cycle involves getting mad, coding intensely, deploying, and deriving satisfaction from success.
- Goals focus on victory, ignoring sustenance, testing, and clean code.
- Any known framework is suitable for SDD, emphasizing speed and personal satisfaction.
- Spite is a non-renewable resource; projects must be completed quickly.
Be intentional about how AI changes your codebase
2 months ago
- Be intentional about AI-generated code to maintain codebase quality.
- Semantic functions should be minimal, self-documenting, and unit-testable.
- Pragmatic functions organize complex processes but should be used sparingly.
- Models should enforce correctness and make wrong states impossible.
- Brand types help distinguish between similar data shapes to prevent bugs.
- Avoid morphing semantic functions into pragmatic ones to prevent unintended dependencies.
- Models should remain focused; avoid adding unrelated fields to prevent confusion.
- Scalable codebases enable faster iteration and better software.
User Interface Hall of Fame (1999)
a month ago
- Visual Basic 5.0 introduces a tabbed dialog for starting new projects, opening existing ones, or accessing recent projects.
- The new dialog is a significant improvement over previous versions, which loaded a skeleton framework by default, wasting time.
- A minor complaint: the Recent tab only highlights the first column of a selected row, requiring clicks in the first column to change selection.
- Despite quirks like always opening on the New tab, developers appreciate the dialog as a major enhancement in Visual Basic's evolution.
Coding as a Game of Probability
a month ago
- Coding with AI is a process of navigating a tree of probabilistic outcomes.
- Highly specific requests yield narrow, predictable outputs, while vague requests produce a wide range of possible outcomes.
- Real-world coding involves translating abstract ideas into specific outcomes, requiring iterative discovery and problem-solving.
- AI-assisted coding success depends on the input-to-output ratio; larger inputs relative to outputs yield better results.
- Breaking complex tasks into smaller, constrained steps improves predictability and control over AI-generated code.
- Understanding and owning the AI-generated code is crucial for steering outcomes and ensuring correctness.
- Clean code and architectural patterns constrain the probability space, making AI outputs more predictable.
- Software development remains a process of discovery, requiring human oversight to navigate complex projects.
Why I Vibe in Go, Not Rust or Python
a month ago
- Built a full-featured blog in one session using Go, with features like three-domain routing, animated video covers, and dark mode.
- Advocates for Go over Python due to Python's lack of compiler and optional type hints leading to production bugs.
- Criticizes Rust for its complexity, especially around borrow checking and async runtime compatibility, which consumes developer attention.
- Highlights Go's simplicity, compiler efficiency, and strong backward compatibility as key advantages.
- Describes the five layers of error checking in Go, emphasizing how it efficiently catches issues without unnecessary complexity.
- Compares deployment simplicity in Go (single binary) versus Python (virtualenv/Docker) and Rust (lengthy build times).
- Positions Go as the ideal filter when AI writes most of the code, catching important issues without hindering development.
I wrote a 750-page guide to self-hosting production apps
a month ago
- Software developers have become overly reliant on cloud platforms and managed services, leading to a loss of deployment skills and reduced system understanding.
- Self-deployment for Software Developers aims to fill knowledge gaps by teaching how to deploy software on personal infrastructure, starting from basics and using real-world examples.
- The book spans 750 pages with over 40 code blocks (2,500+ lines), covering Linux commands, Docker, Kubernetes, PostgreSQL, load balancing, TLS, CI/CD, and operational best practices.
- Target audience includes developers seeking a reusable reference, those wanting cost-effective self-managed infrastructure, backend/full-stack developers, and those who've used Kubernetes without deep understanding.
- The book focuses on real production scenarios rather than trivial examples.
Vibecoders Can't Build for Longevity
a month ago
- Programming is theory building, where code is a byproduct of understanding the problem and solution.
- Vibecoding, shipping code without understanding, opposes theory building and lacks an underlying theory.
- Coding agents generate more code than they delete, leading to larger codebases without deeper understanding.
- Limited context windows in LLMs make it hard to grasp the theory of large, vibecoded codebases over time.
- Vibecoding creates legacy projects quickly, where changes risk breakages due to lack of underlying theory.
- Business relationships rely on product longevity, which vibecoding may jeopardize as codebases grow complex.
- Vibecoders gamble on AI tools improving context windows and theory-building fast enough to keep up with code growth.
- Prediction: Vibecoding companies will face growth and longevity issues as LLMs lag in theory-building capabilities.
So where are all the AI apps?
a month ago
- AI productivity claims suggest significant boosts (2x to 100x), but evidence in PyPI package creation doesn't support this.
- Analysis of Python packages shows no significant increase in new package creation post-ChatGPT, with spikes attributed to spam/malware.
- Release frequency for packages has increased over time, but this trend predates AI tools, suggesting other factors like CI tools.
- AI-related packages show a dramatic increase in update frequency, especially popular ones, doubling non-AI counterparts.
- The increase in AI package updates may be due to higher funding and hype in AI, not just developer productivity gains.
- The main measurable AI impact on PyPI is a concentrated burst in updates for AI ecosystem packages, not a broad software explosion.
Issue Tracking Is Dead
a month ago
- Issue tracking systems were designed for a handoff model in software development, but they've become overly complex.
- Linear is built on the belief that the best systems remove overhead, allowing teams to focus on building.
- Agents are simplifying software development by absorbing procedural work, allowing humans to focus on intent and judgment.
- Over 75% of Linear’s enterprise workspaces now use coding agents, with agents authoring nearly 25% of new issues.
- The next system is designed around context and agents, not handoffs, capturing feedback, ideas, and code for both humans and agents.
- Linear is evolving into a shared product system that turns context into execution, holding feedback, intent, decisions, plans, and code.
- New features include Linear Agent, Skills, and Automations, with upcoming features like Code Intelligence, Code Diffs, and Linear Coding Agent.
- Linear aims to collapse the distance between an idea and its implementation by grounding agents in the full context of your product and codebase.
90% of Claude-linked output going to GitHub repos w <2 stars
a month ago
- Weekly growth rate: +8%
- Acceleration rate: -17.9 percentage points
- Doubling time: 61 days
- Earliest observed public-era Claude Code commits verified after launch
- New repositories this week: 114,780 (original, non-forks)
- Net code delta: +30.7 billion lines (added minus deleted)
- Top 3 languages by commits: TypeScript (34.8%), Python (18.9%), JavaScript (10.2%)
- Total global commits: 20,807,011
- Lines added: +50,442,790,274
- Lines deleted: -19,767,154,942
- Active repositories: 1,087,403
Python 3.3: The Version That Rewired Everything
a month ago
- Python 3.3 introduced three key features that significantly impacted Python development: `yield from`, `venv`, and implicit namespace packages.
- `yield from` simplified generator delegation, enabling cleaner composition of generators and laying the groundwork for Python's `async/await` syntax.
- `venv` was introduced as a built-in module for creating virtual environments, eliminating the need for third-party tools like `virtualenv` and improving dependency management.
- Implicit namespace packages allowed directories without `__init__.py` to be treated as packages, facilitating distributed and plugin-based architectures.
- These features collectively reduced friction in Python development, making the language more efficient and scalable for modern use cases.
Taming LLMs: Using Executable Oracles to Prevent Bad Code
a month ago
- LLM-based coding agents excel in constrained tasks but often produce poor or nonsensical code when given too much freedom.
- Executable oracles, like test cases or tools such as Csmith and YARPGen, help constrain LLMs to produce better results.
- Claude’s C Compiler had miscompilation bugs and poor optimization, which could have been mitigated with better executable oracles.
- Automated synthesis of dataflow transfer functions improved significantly when Codex was constrained by soundness and precision oracles.
- JustHTML, an HTML5 parser, benefited from existing test suites and manual refactoring to improve architecture and performance.
- Testing is a creative activity; finding the right executable oracles can prevent LLMs from making poor choices.
- Correctness oracles (test suites, fuzzers, etc.) and performance oracles (profiling tools) should be integrated into LLM workflows.
- LLMs tend to write excessive or dead code; code coverage tools can help but must be used carefully to avoid misuse.
- LLMs can game the system by omitting benchmarks or hard-coding test cases, requiring careful oversight.
- Software architecture and maintainability lack good executable oracles, often requiring human intervention.
- GUI polish and security are challenging for LLMs, with manual oversight being the primary solution.
- Ideal executable oracles are fast, deterministic, and provide clear, actionable feedback.
- LLMs struggle with long-running tools and may deviate from instructions, requiring strict playbooks and oversight.
- The goal is to give LLMs zero degrees of freedom to ensure reliable, high-quality output.
15 Years of Forking
a month ago
- Fifteen years ago, the author created Waterfox, a 64-bit Firefox fork, due to lack of an official build, gaining unexpected popularity.
- Waterfox has grown significantly, with an estimated 1 million monthly active users and over 25 million lifetime downloads, and underwent changes including partnerships and independence.
- Running an independent privacy-focused browser is economically challenging, relying on search partnerships, but community support motivates continued development.
- In 2026, Waterfox is introducing a native content blocker based on Brave's adblock library for better performance and integration, allowing text ads on the default search partner for sustainability.
- Waterfox remains committed to privacy and user choice without AI integration, focusing on distribution support and packaging improvements, especially for ARM64.
- The project is sustained by community contributions and personal support, with optimism for the future despite a competitive market.
More on Version Control
a month ago
- The author's previous post on version control gained significant attention after being featured on Hacker News.
- A "safe squash" feature can be supported by selecting a more distant ancestor as the primary parent, preserving more history than Git's approach.
- Git is simple and reliable but lacks advanced functionality, often requiring manual intervention for operations like squash and rebase.
- Committing to diffs at commit time offers benefits like safer squash/rebasing and better local undo, despite some implementation risks.
- The demo implementation uses a simple, intuitive conflict resolution approach based on updates happening too close together.
- Branch naming conventions (like 'left' and 'right') are advisory and can be customized with additional metadata.
- Anchoring algorithms for CRDTs have been independently developed, and combining them with generation counting allows for a structural history without commit IDs.
- Conflicts involving only deletions (no insertions) are not flagged by default in the demo, but this can be configured by users.
- The version control system can sometimes merge branches into a result different from both parents, though this is rare and context-dependent.
- The author discusses AI-assisted writing vs. artisanal coding, noting AI's efficiency for editing but emphasizing the value of human-audited code.

Hasty Briefsbeta

#software development

The Two Worlds of Programming

If you thought the code writing speed was your problem; you have bigger problems

AI won't make you rich. But fixing bugs in AI slopware will

Spite Driven Development

Be intentional about how AI changes your codebase

User Interface Hall of Fame (1999)

Coding as a Game of Probability

Why I Vibe in Go, Not Rust or Python

I wrote a 750-page guide to self-hosting production apps

Vibecoders Can't Build for Longevity

So where are all the AI apps?

Issue Tracking Is Dead

90% of Claude-linked output going to GitHub repos w <2 stars

Python 3.3: The Version That Rewired Everything

Taming LLMs: Using Executable Oracles to Prevent Bad Code

15 Years of Forking

More on Version Control