implementation cost is going to zero. so what's left? vision. taste. judgment. the cognitive demands didn't shrink. they concentrated.

about 30% of my dev time goes to meta-work now. custom commands, agents, feedback loops, memory systems. not shipping features. building the system that ships features. every piece compounds into the next. and eventually, i want other builders to be able to use it too. more on that soon.

the pragmatic engineer surveyed 900+ engineers. 56% now do 70%+ of their work with AI. claude code went from zero to the #1 AI coding tool in 8 months. this isn't early adoption anymore. also who are the other 44% and what's the 30%? am i weird for being at like 99%?

SaaS isn't dying. but it's no longer the default answer to every business problem. Looks like a growing number of businesses are hiring builders to create internal tools instead of buying another subscription. If I can be of help, reach out!

last week i shipped across 6 projects in ~35 hours. not because i worked harder. because AI handled most of the execution. the leverage is real. but i'm still the bottleneck for every prompt. writing about that tension in this week's newsletter. subscribe: jkudish.com/newsletter

built a /deep-review command: fans out code review to 4 AI models in parallel. if 2+ flag the same issue, severity gets bumped. auto-fixes critical and high findings before the PR exists. built on @aarondfrancis's counselors package.

what's the one thing in your AI coding setup that compounds the most? for me it's context. CLAUDE.md files, auto-memory, project-specific configs. every session leaves behind knowledge that the next session builds on. the AI gets better at my codebase without me doing anything extra.

the more i automate my dev workflow, the more my job looks like product management. define the problem. clarify requirements. approve the plan. review the output. the engineering skills didn't become less valuable. they just moved from typing code to systems design.

claude code PostToolUse hooks are underrated IMO. i added a 17-line bash script that detects gh pr create and automatically starts the code review loop. no manual step between "PR created" and "reviewer comments handled." code.claude.com/docs/en/hooks

do you want AI to cite your brand? own the definition. find the specific terms in your niche that don't have clean, authoritative explanations anywhere. write the best one. structure it so AI can extract it cleanly. I call this the "dictionary definition heist."

three layers to my agentic memory system: 1. raw transcripts. every conversation stored as source of truth. 2. compressed observations. what the agent noticed and extracted. 3. normalized truths. facts with confidence scores and validity windows.

most AI conversations start from zero. you open a new chat and the AI has no idea what you worked on yesterday. I built a memory system so mine doesn't work that way anymore. three layers: raw transcripts, compressed observations, normalized truths with confidence scores.

i've been experimenting with different AI memory systems. i wanted something that can store observations, connect entities, and retrieve context across sessions. my first instinct was to reach for a hosted vector database. that's what everyone recommends. 🧵

caught up with the latest development happenings in WordPress over the weekend. some of my clients are heavily invested in the ecosystem and it's important to stay current. the biggest shift: the WordPress Interactivity API (IAPI). 🧵

icymi: i published my first npm package this weekend, 🏛️ Librarium it's an open source CLI tool that lets you fan out research queries to multiple search and deep-research APIs in parallel for human or agent consumption. let me know what you think! github.com/jkudish/lib...

🪶 announcing plume — a Laravel package for the X API v2. if you're building anything with the X API in Laravel, i hope this becomes the canonical implementation. give it a try and feel free to star/share 🙏 github.com/jkudish/plume

First time sharing my work on Hacker News. I posted about my new CLI tool Librarium. news.ycombinator.com/item?id=471... github.com/jkudish/lib... Appreciate an upvote if you can!

🏛️ Introducing Librarium Fan out research queries to multiple search and deep-research APIs in parallel. Results are collected, normalized, and deduplicated into structured output that your agents can consume or write to markdown. github.com/jkudish/librarium

🏛️ Introducing Librarium Fan out research queries to multiple search and deep-research APIs in parallel. Results are collected, normalized, and deduplicated into structured output that your agents can consume or write to markdown. github.com/jkudish/librarium

the real unlock with AI agents isn't just speed. it's parallelism. i work faster AND on more things at once.

what's one thing you automated this week that you used to do manually? for me: new errors on most of my projects automatically create an AI driven investigation and PR for me to review and approve.

i've been deep in SEO/AEO/GEO across several sites lately. if those terms are new: - SEO: rank on google - AEO: get cited by ChatGPT, Perplexity, Gemini - GEO: show up in AI overviews and AI search here's what's actually moving the needle. 👇

i'm building a Laravel platform that connects everything i do in my business. my obsidian vault, my tasks, my AI agents, my client projects; one operational layer. having everything talk to each other is where i see AI and agents going. not isolated tools; connected systems.

planning is the highest leverage part of the agentic engineering process. AI compressed the build step. lines of code that used to take days can take as little as 15min. but if you skip planning, you just get to the wrong answer faster.

upgraded a client project from Laravel 9 to 12 today. three major versions, swapped Nova for Filament, full test suite and manual testing. Laravel Shift + AI with a few judgment calls on my part but mostly automated in one day. done mostly in the background while working on other projects.

AI helps me leverage the things i'm good at; architecture, taste, client conversations; and speed-run the repetitive stuff i used to procrastinate on. it didn't replace my strengths. it gave me more time to use them.

outsource execution to AI. use AI to partner on decisions and thinking. use AI to help you learn. but don't outsource knowledge and understanding. that's the part that compounds.

the cost of building internal tools dropped so much that "just deal with it" stopped making sense. hit a friction point, build a tool, move on. most of them take an afternoon now. the best ops improvements aren't bought. they're built for your exact workflow.

models get replaced every few months. your system survives the upgrade. the memory files, the workflows, the conventions; they transfer to the next model and the one after that. bet on the system, not the model.

the loudest AI conversation is about which model is smartest. the quiet advantage is the person who spent six months building systems around a "good enough" model and now ships circles around everyone debating benchmarks.

Laravel has mail, notifications, queues, jobs, a package system, Eloquent, and now an AI SDK built in. AI understands the framework well. the ecosystem is mature. you can go from idea to production faster than ever. there's never been a better time to build with Laravel. Laravel forever.

i spend more time tweaking my AI system than i probably should. new agent for this. better prompt for that. one more memory file. but every hour invested pays back across every project i touch. so i keep going.

don't optimize the prompt. optimize the context. your notes, your conventions, your past decisions; all feeding into every interaction. that's where the real leverage is.

i've accumulated years of notes in my obsidian vault. every meeting, every decision, every mistake; all searchable, all linked. the compounding isn't in the notes themselves. it's in how fast i can recover context i would've lost.

your AI system is a mirror of how clearly you think. messy prompts, vague instructions, no structure? you'll get messy output. clear conventions, specific context, well-defined boundaries? the AI looks like a different tool entirely.

most scope creep isn't about clients adding features. usually, the problem is not asking enough questions nor understanding the problem at the start.

nobody talks about maintaining AI systems. your prompts drift. your agents break on edge cases. context files go stale. a model update changes behavior you relied on. it's software. it needs upkeep. the people treating AI as "set and forget" are going to hit a wall.

what does your AI system look like right now? not which model you use. what have you built around it? memory files, custom prompts, agents, workflows, integrations; anything that makes it yours.

everyone has access to the same AI models. the moat isn't which model you use. it's what you've built around it. your context files. your agent workflows. your feedback loops. the stuff that makes the AI work for your specific business, not just in general.

building an AI system feels slow at first. writing memory files. documenting conventions. creating custom agents. setting up context that persists. then one morning you start a session and the AI already knows your codebase, your preferences, and your last three decisions. that's when it clicks.

my AI system isn't one tool. it's a stack. memory files that persist across sessions. custom agents for specific tasks. task tracking that syncs with my workflow. context that builds on itself. none of it is fancy. all of it compounds.

most people use AI tools. not many are building AI systems. a tool speeds up one task. a system remembers your decisions, feeds context into the next session, and gets sharper the more you use it. one saves time. the other compounds.

i don't think AI replaces senior developers. i think it makes the gap between senior and junior wider than it's ever been. seniors know what to ask for, what to reject, and when the AI is confidently wrong. that's not a skill you can shortcut.

the developers mass-adopting AI right now will split into two camps within a year. those who built systems around it. and those who just used it as a faster keyboard. the gap will be enormous.

if your AI can't read your past decisions, it's just a fancy autocomplete. CLAUDE.md, specs, commit history, architecture docs. that's not documentation overhead. that's your compound interest.

building with AI every day outweighs reading about AI every day. the anxiety about the future dropped and got replaced by genuine excitement for what's possible.

the gap between "I have an idea" and "it's live" has never been smaller. which means the gap between you and your competition has never been smaller either. taste, positioning, and who you build for matter more than ever. the code is the easy part now.

your AI workflow is only as good as what it remembers tomorrow. most people start every session from scratch. same mistakes, same corrections, same context loading. build the memory layer. that's where the leverage is.

i use AI to think slower. spec conversations, architecture reviews, decision journals. each one feeds the next session. the AI gets sharper because my inputs get sharper. shipping fast, at this point, is table stakes. compounding context is what I focus on now.

everyone's talking about which AI model is best. the developers actually shipping are using all of them. different models, different tasks, swap as needed. the answer was never "pick one." it's orchestrate many.