Posts

React

React 1) vanilla useEffect state tracking for "side effects" like APIs, event listeners, timers, and cleanup. Talk about the "ignore flag" cleanup pattern inside useEffect to drop stale promises when components unmount or dependencies change. Versus declarative, hook-based data-fetching library (like v4 TanStack Query   was React Query v1-3)  useMutation auto receives isPending, isError and isSuccess flags.      useEffect fires asynchronously after the browser paints the screen to keep the application responsive. useLayoutEffect fires synchronously after DOM mutations but before the browser paints. Using useLayoutEffect blocks the paint loop, causing slight performance penalties, but it prevents visual layout flickers when calculating element dimensions or positions dynamically.   2) export interface OrderRequest { ... fields ... } to keep DTO in synch. (Tools like TypeGen , NSwag , or Typewriter ). 3) Q: Problem with lots of re-renderi...
Post a Comment
Read more

AI Advantages and Disadvantages

  AI Advantages and Disadvantages AI is good at writing code but not engineering an application. AI is better with scaffolding. AI is bad a making changes, and are good are doing it one shot. AI training is one-shot training and not iterative development on your code base. All expectations mentioned up-front.   AI fails at large-scale applications where structure, architecture, and ease of  changing fails.   Human comprehension plan graded by AI.   AI fails at OOP since Python focused.   AI fails at larger projects to maintain clean code or come up with object  pattern architecture (instead it does flat classes in a project).   AI fails at being consistent with object patterns, even though doing similar  thing.   AI good with architecture of projects in a solution.   AI is fair with naming objects, but inconsistent over different sessions.     AI creates highly coupled objects and lots of visibility from one object to anot...
Post a Comment
Read more

SaaS Moats

SaaS Moats Supply Attacks     1) Your customer builds your app with AI. Only need to replace 3 features they use. 2) Competitor reaches parity quickly. 3) Open source gets good enough. 4) The platform/integration partner ships your feature for free. 5) Custom software at SaaS prices - generic $99, built for you $299. 6) Model absorbs your product. You just became a wrapper. Demand Attacks 1) Will your customer still exist 2) Are your seats (number of employee users) going away? 3) Does the step you manage still matter? 4) Is a human still the one using your software? 9 Reasons Customers Stayed 1) Unique features 2) Owned traffic channels 3) Clever code 4) UX familiarity 5) Integrations 6) High switching costs 7) Strong Brand - still in AI  8) Coordination between people - still in AI 9) System of record - still in AI 8 Moats 1) Buyers know and trust you. ex: Ecosystems - Quickbooks or Shopify partners. 2) System of record - daily operations are built around you. ex: ...
Post a Comment
Read more

New Computer Language

New Computer Language  We as programmers could have a new computer language. No heap. No pointers. So deterministic. Bounded while statements. Bounded loops. Active models. Go back to record structure (or plex). Extremely high if you have that. build for GPU vs. CPU are 2 completely different ways. CPUs are optimized for path finding and just preloading data. GPU is very deterministic - allocate a piece of memory, fill it, run some functions on it.   no heap or allocation. Single Instruction, Multiple Data (SIMD)     Do not do dynamic allocations anymore Discriminated unions - Rust Borrow checker - heap - don't use Imagine with no borrow checker. Start index. Now only have offsets for all of    the fields in your record. Hard Real-Time and Embedded Languages: MISRA C / SPARK Ada. Lustre synchronous programming language and Esterel. * No Recursion: The call stack depth must be completely predictable at compile time. * Fixed-Size Collections: Arrays cann...
Post a Comment
Read more

Top open-weight AI models for coding in late 2026

  Top open-weight AI models  for coding in late 2026 • GLM 5.3 (Z.ai / Zhipu AI): A 753B-parameter flagship model (with 8 active experts per token) featuring a 1,048,576-token context window, currently ranked among the top overall and long-horizon agentic coding models. • DeepSeek V4 Pro (DeepSeek): A 1.6T-parameter (49B active) MoE model with a 1M-token context window under an MIT license, delivering frontier-class code reasoning and repository-scale analysis at a fraction of closed-model costs. • Kimi K3 (Moonshot AI): A massive 2.8T-parameter (104B active) frontier open-weight model with a 1M+ token context window, excelling in frontend code generation, multi-agent workflows, and complex long-document tasks. • Qwen 3.8 Max (Alibaba): A 2.4T-parameter model (95B active) supporting long contexts up to ~1M tokens, widely praised for enterprise-grade coding and agentic execution. • gpt-oss (OpenAI): OpenAI's open-reasoning model available in 120B and 20B parameter variants unde...
Post a Comment
Read more