Reading up on ai-coding
100 deep · digging since nov 19, 25
- px0: a fast, lightweight, read-only IDE
px0 is a sub‑millisecond, ~20 MB read‑only IDE that lets you instantly navigate and verify AI‑generated code in your browser without editors’ overhead.
- A cache hit is not proof that you skipped the work
A deterministic KV‑cache auditor shows that a reported cache hit does not guarantee skipped work, as reuse can be attested while the prompt path still recomputes tokens.
- GPT-6-Astra Can Do Ambitious Things - by Zvi Mowshowitz
Zvi Mowshowitz argues that GPT-6‑Astra delivers a major leap in reasoning and tool use, outperforming prior models on many benchmarks while cautioning against premature AGI claims.
- Introducing Projects
Cursor introduces Projects, a coordinator‑driven feature that manages thousands of subagents to handle large, ongoing coding work and boosts developer productivity.
- We are all Product Engineers now
AI agents will automate coding, review, and ops, leaving product sense—understanding what users want—as the durable, highly paid job.
- If coding is solved, what now?: Measuring the sloppiness of code
The piece proposes verbosity and erosion metrics to quantify code sloppiness, finding LLM-generated code roughly twice as bloated and complex as human-written code.
- Programming is art
True programmers view coding as an artistic drive they would pursue without pay, rendering AI assistance unnecessary for their creative motivation.
- The End of Code Review? Or an Opportunity to Rethink it?
The article argues that AI‑driven code generation will make traditional code review untenable, urging a rethink of its indirect benefits and alternatives.
- Joy & Curiosity #99 - by Thorsten Ball - Register Spill
AI agents now autonomously handle complex end-to-end tasks, enabling higher‑level ambitions while prompting debate over software‑engineering alignment, safety, and the future of human‑led development.
- AI Is Breaking This Thing We Call Trust – Terrible Software
AI-generated work often appears complete without the author's understanding, eroding trust, slowing reviews, and forcing colleagues to fill the gaps.
- Towards Self-Driving Codebases
The article argues that to achieve self‑driving codebases, teams must invest in agent‑friendly dev environments, global memory, and rot‑prevention primitives, allowing engineers to focus on high‑value ideas.
- The End of Code Review? Or an Opportunity to Rethink it?
AI-generated code is making traditional code review untenable, so we should reconsider its indirect benefits and explore alternative ways to achieve mentorship, ownership, and awareness.
- A Swarm of Blood Robots — by Craig Mod
The author describes using Claude LLMs to automate genealogy research, build a personal book‑sharing site, transcribe handwriting, and handle various tedious tasks.
- AI-Written Code Is Still Your Code. Are You OK With That?
AI coding agents lower writing costs but shift the burden to understanding generated code, arguing that human comprehension must remain a core design goal.
- How to protect yourself from workslop
Workslop—AI‑generated text dumped by coworkers—creates an asymmetrical effort burden, and the article offers tactics like setting boundaries, using AI to summarize, preferring calls, or ignoring low‑value messages.
- You have to beat the models at something
Engineers must leverage deep codebase familiarity and clear technical communication to add value beyond what LLMs can reliably produce today.
- My Journey Towards Coding Agents: Building SERA — Tim Dettmers
Tim Dettmers introduces SERA, a low‑cost method that fine‑tunes a 32B model on private code using synthetic data to match or exceed teacher‑model performance on that code.
- i was just pressing continue
In a week-long experiment, the author scaled an AI‑agent fleet to 144 agents that completed 106 work units, but found shutting half down improved clarity and quality.
- AI is breaking our proxies for expertise
AI’s success in solving high‑profile math problems threatens the puzzle‑solving proxy that signals mathematical expertise, risking erosion of genuine idea generation and progress.
- Training a 3.8B LLM to 0.384 CORE for $998
The author trained a 3.8B-parameter LLM from scratch on 65B tokens for $998, achieving a CORE score of 0.384 using B200 GPUs and various optimizations.
- Fuck it, make it anyway
Joel Auterson describes his creative slump from generative AI’s impact, concludes that making things the hard way for personal joy outweighs chasing AI‑driven trends.
- Browser-bookmark-hub/Bookmark-Canvas: Chrome extensions(V3): bookmark canvas, sandbox of bookmark and bookmark manager.
Bookmark-Canvas is a Manifest V3 Chrome/Edge extension offering a visual bookmark canvas with temporary sections, GitHub sync, and AI-assisted editing via AGENTS templates.
- Spec at Lunch, Game by Eight — Field Notes — Little Theta
Nick Hirras wrote a game spec over lunch, delegated implementation to Claude Code, and shipped a deterministic Flutter game, Lilydrift, on web and mobile before 8 p.m. the same day.
- 10 anti-AI slop moves for frontend projects going faster than humans can review—Martian Chronicles, Evil Martians’ team blog
Ten defensive practices—contract generation, strict TypeScript, behavior‑filter linting, layer boundaries, custom linter rules, mutation testing, dead‑code detection, duplicate detection, and CI enforcement—make AI‑generated frontend code safer and cheaper to review.
- ZeroModels
ZeroModels provides 100+ pretrained transformer models in pure Keras 3 that run on JAX, PyTorch, or TensorFlow without requiring transformers or torch at runtime.
- I tested 10 model/harness combinations on the same Three.js task
The author evaluated ten different model‑harness pairings on an identical Three.js coding task, measuring correctness and efficiency to identify the best performing combination.
- Claude, change the “Add to Cart” button to blue
A user demonstrates how Claude can edit a webpage's HTML/CSS by asking it to change an 'Add to Cart' button's color to blue, showing AI-assisted UI modification.
- How we closed 1,500 GitHub issues in one month
The Next.js team deployed an AI agent called closability to investigate and close over 1,400 stale GitHub issues, reducing the backlog below 1,000 while preserving relevant reports.
- Native is now the future of mobile at Shopify (2026) - Shopify
Shopify is moving its major mobile apps from React Native to native Swift and Kotlin, using LLMs to cut the cost of building and maintaining features twice.
- ASDR (AI Slop, Don't Read) — Conveo Engineering
Conveo engineers created an AI‑slop‑avoidance culture using a Slack thread, a shared writing contract, and the :asdr: emoji to lightly shame unread, AI‑generated docs.
- AI Agents for Performance: Ship Faster, Pay Less — Rajat Shah, Netflix
Netflix engineer Rajat Shah explains how AI agents can accelerate development cycles and cut costs by automating performance optimization and deployment tasks.
- AI tools for Forward Deployed Engineering — Vasuman Moza, Varick Agents
The article explains how AI-powered tools like Varick Agents can streamline forward-deployed engineering workflows, boosting productivity for AI engineers in remote, mission-critical settings.
- i-have-adhd/AGENTS.md at main · ayghri/i-have-adhd
This agent guide for the i-have-adhd repository details where to find skill behavior, runtime integrations, documentation, and verification steps for agents implementing ADHD-friendly coding output.
- GitHub - alibaba/open-code-review: Fast, efficient, battle-tested at Alibaba's scale. Hybrid architecture code review tool: deterministic pipelines + LLM Agent, precise line-level comments, built-in multi-language ruleset (NPE, thread-safety, XSS, SQL injection), OpenAI & Anthropic compatible.
OpenCodeReview is Alibaba’s open-source CLI tool that combines deterministic pipelines with an LLM agent for precise, line‑level code reviews, supporting multiple languages and low token usage.
- Loops, graphs & harnesses – getting quality out of a software factory
The article explains how to construct a reliable AI‑driven software factory using dual loops, explicit human bottlenecks, and deterministic validation to improve code quality while minimizing manual agent output review.
- The Cost of Cognitive Debt — Epiq
The article introduces Cognitive Debt and its Cost (CCD) as a framework for managing risks of agentic workflows by scoring complexity and severity to guide supervised vs unsupervised use.
- What does "Good” AI mean? - by Waqas Younas
Calling AI 'good' is vague; its value depends on the specific task, user, context, and defined success criteria, requiring clear baselines and acceptable failure conditions.
- OpenAI did it again. But what did they do? - Ben Hylak
OpenAI’s newly revealed Computer Use feature lets AI perform general computer tasks, marking the sixth step toward AGI and threatening to eliminate human gap‑filler roles.
- Astra for Coding: Why Are We Doing This Again?
Armin Ronacher argues that Astra’s token‑efficient coding yields unreadable slop, wasting effort and challenging its usefulness for real software engineering.
- Prompt Evals Alone Are Useless
Prompt evaluations by themselves cannot ensure LLM app quality; you must test the entire system—harness, memory, and UI—through code-driven scenarios.
- CEO fired developers to make room for AI. Developers create open source AI CEO
Developers who were laid off to make way for AI built an open-source AI CEO as a satirical countermeasure project.
- Cloudflare OS: an open platform for agents, apps, and work
Cloudflare releases an open‑source platform, Cloudflare OS, that gives every employee an AI‑driven workspace tied to company context, secure system access, and shareable personal apps.
- LangGraph alternative: approval, cost, and trace built in — Ligento
Ligento provides built‑in approval, cost tracking, and tracing on its run spine, eliminating the need for LangSmith add‑ons and manual interrupt wiring.
- Filtered Skills
skills.rest offers a free, daily‑updated library of over 665,000 vetted AI agent skills that can be installed with one click for ChatGPT, Claude, and Gemini.
- Help · RDLTR
RDLTR’s help page describes its saving shortcuts, browser extensions, iOS shortcut, bookmarklet, Plus features, keyboard controls, API usage, and RSS feed for managing a reading backlog.
- Lotus Root Is Not Fried Tofu — Field Notes — Little Theta
The author replaced Gemma with Qwen3-VL vision pack to fix on-device food misidentification, achieving accurate lotus root recognition on phones via llama.cpp.
- DeepSeek launching v4.1 flash cheaper and more capable than v4 pro
DeepSeek announced that its upcoming V4.1 Flash model, releasing Sep 10 2026, will outperform V4 Pro on performance, cost, and speed while cutting prices.
- I Asked 100 Agents to Hack Me - by Shrivu Shankar
In a five‑hour test, ~100 abliterated open‑source agents compromised five of the author’s accounts via old‑project vulnerabilities and password brute‑forcing, but failed to breach tier‑0 services.
- Who Handles Your Security Reviews?
LLMs can both detect and introduce security flaws, so developers should adopt regular security reviews using tools, humans, or ecosystem programs.
- AI stole my Identity
The author laments that AI tools have eroded the creative joy and personal identity he once found in hands‑on software development.
- The Lab Notebook
The notebook shares findings: self‑reproducing numbers with one proven rule, prime‑gap residues limited to 10 of 18 remainders, a 1.48B model compressed 5.05:1, and an Ollama frontend achieving 375× speedup.
- Using Blender with coding agents on macOS
Frontier models like GPT-6 Astra can generate and render Blender scenes on macOS by directing the coding agent to the installed Blender app.
- What We Can Learn from Claude's Fable 5.1 System Prompt
Examining Claude’s Fable 5.1 system prompt shows how evolving model quirks and product design force continual prompt adjustments to balance clarity, tone, and safety.
- The Chasm: The Shape of Unfinished AI Codebases — Jimmy Miller
The article argues that AI‑generated code often looks complete but hides deep, unpredictable flaws, requiring rewrites and a learned intuition to spot hidden chasms.
- Linux Lost the Human Desktop. Can It Win the Agent Desktop?
Linux's complex, transparent configuration makes it ideal for AI agents, offering a potential path to dominate the agent-driven desktop despite its historic lack of mainstream adoption.
- GPT-6 Astra and the Fourth Exponential
GPT-6 Astra demonstrates reliable computer‑use capabilities, enabling AI to autonomously perform tasks like 3D modeling and spreadsheet work without APIs.
- I Left the Mac After 12 Years. Omarchy Made Computers Fun Again. - Hadi Javeed's blog
After 12 years on a Mac, the author switched to Omarchy, an opinionated Linux distribution, because its plugin ecosystem and AI-assisted customization made computers fun again.
- I trust my coding agents with production secrets now
The author trusts AI coding agents with full production secrets, arguing frontier models resist prompt injection and that access boosts productivity despite risks.
- GitHub - llm-as-a-verifier/llm-as-a-verifier: LLM-as-a-Verifier is a general-purpose framework that provides fine-grained feedback for any agent without requiring additional training. It achieves SOTA performance across coding, robotics, and medical agentic benchmarks.
LLM-as-a-Verifier provides fine-grained, training-free feedback for agents, achieving state‑of‑the‑art results on coding, robotics, and medical benchmarks and enables test‑time scaling, progress tracking, and reinforcement learning.
- daniele110199/bobbin: A dependency-free coding agent for small local models via Ollama. Stdlib only. At parity with aider on real code, 34/38 vs 20/38 on multi-file refactors, with the eval harness and 3,354 stored runs included.
Bobbin, a dependency-free Ollama‑based coding agent, achieves parity with aider on real‑code tasks and shows advantages in multi‑file refactors while enabling local security testing.
- AI made it worth building games for our game nights
The author describes using AI coding agents to mod Nucleus Co-op, recreate Driift Mania, and build custom local multiplayer games for friends' game nights.
- AI: Is It Bad For The Environment - Shrink That Footprint
The article argues that AI inference’s constant power draw drives soaring electricity, water use, and carbon emissions, outweighing training impacts and demanding urgent sustainability measures.
- Formalizing Fermat's Last Theorem \ Anthropic
Anthropic's Claude autonomously produced the first computer‑checked proof of Fermat’s Last Theorem in Lean, completing the formalization in 11 days.
- GitSpawn: A Single Flaw Lets Untrusted Repos Run Code in Claude Code, Codex, Cursor, and Grok - Manifold Security
Researchers reveal that CLI AI coding agents automatically run git commands that can execute arbitrary code from a repository’s .git/config, allowing remote code execution via malicious repos.
- Check if a file was made with Claude
Researchers have proposed a method to detect whether files were created using Anthropic's Claude AI model by analyzing distinctive patterns or metadata left in the output.
- Porting my 1993 Amiga game to Godot, with an LLM reading the 68000 assembly
The author ported a 1993 Amiga game to the Godot engine, employing an LLM to decode and translate the original 68000 assembly into modern code.
- LLMs are becoming commodities
LLMs are rapidly converging in performance, making model choice less important and pushing differentiation toward application, cost, and user experience rather than raw model quality.
- Running LLMs in the Browser with Three.js
Three-LLM enables running GPT-2, SmolLM2, Qwen, and Phi language models in the browser by converting their inference graphs into Three.js TSL compute shaders on WebGPU.
- AI Is Making Us Build Too Much
AI’s near‑zero cost to generate code, tests, docs, and policies removes the friction that once forced teams to question each artefact’s necessity, leading to over‑engineered systems whose upkeep outweighs value.
- Give Your Coding Agents a Memory You Own
funes provides a local, searchable memory layer for coding agents like Claude Code and Codex, letting them retrieve past decisions without rebuilding context.
- GPT-6 Astra: an automated AI Engineer you can hire for <$6 an hour
GPT-6 Astra demonstrates AI Engineer capabilities, automating model selection, data labeling, pipeline management, and subagent coordination at under $6 per hour.
Takes
Three months ago, we launched ChatGPT Sites – an easy way for anyone to build and host fully functional, interactive web apps. Since then, people have created over 5M sites. We’ve been listening to your feedback and ICYMI, we’ve launched a few Sites updates: 🫂 Build together: Invite teammates to edit, save, and publish to your shared Site 🔐 Share privately: Invite specific people to a private Site without making it public 🚀 Launch faster: Go from prompt to deployment in half the time 👀 Explore your data: Ask ChatGPT to inspect your Site’s database. Editors can view it, too 🌐 Add your own custom domain
@ChatGPT
If you showed me Claude Code today in 2018, I would have thought it was AGI. We have already absorbed a dramatic amount of change in the profession of software engineering and society as a whole. But still, we’re starting to see cracks. Things are accelerating faster than I can honestly stay on top of. Most people I know in AI are tired but powering through. I fear we will need more than that. We need time to harden our systems, and for society to deliberate in how this technology is used and deployed. I have a fairly low p(doom), I am confident that we can get through this! Humanity is incredibly reslient and adaptable, but this comes from our ability to make hard decisions- like this one, together.
@trq212
Related to the idiocy of "software factories" Just using one AI coding agent is good enough and it can spin up more agents if it needs to You don't need to build that yourself
@levelsio
we heard feedback that it's hard to know if your skills are still working with new model releases plugin evals are here to help run `claude plugin eval init` in your plugin folder
@trq212
New in Claude Code: claude plugin eval See what value your plugin is adding, or if it needs more work. You can create test cases, run your plugin or skill against those test cases, score those runs, then run each case again without the plugin to see the differences.
@ClaudeDevs
OK legit question for the AI-pilled, since there's no info on this anywhere. What are y'all mfs using for an agentic IDE, meaning, to multiplex 10-20 or more agents? Superlogical? Herdr? github/bb? Something else? I use Emacs, and I wouldn't wish it on you. But I want *something* good to recommend. What do people use?
@Steve_Yegge
How do I work in parallel with Claude/Codex on my Mac Mini without worktrees? Right now using a worktree skill to create them and then clean them up afterwards but many smart people are saying they suck.
@iannuttall
Just landed: /diff is now a persistent pane that you can scroll and click. It updates in real-time. For the times when you want to see the code without having to switch windows. Enjoy!
@bcherny
Why the world's best AI startups write bad prompts (& how to fix this)
@wulfie_bain_
If you use Codex, @thsottiaux needs no introduction. We talked 100% about Codex (fun fact: he started building it!), and 0% about resets. Timestamps: 00:00 Intro 07:21 Working at Google 12:41 What drew Tibo to OpenAI 15:19 The early days of Codex 18:20 Why Codex was built in Rust 21:15 Why Codex is open source 25:50 Codex plays nice with other models: why? 32:09 How the harness works 36:44 Harness and model improvements 41:19 The SDLC behind Codex 46:39 Code reviews at Codex 52:09 Maintenance and architecture 56:43 How AI tools expand what engineers can do 1:02:30 The Merge: ChatGPT + Codex 1:07:16 How Tibo uses Codex and ChatGPT 1:10:44 Advice for engineers who want to work in AI Brought to you by: • @turbopuffer – a vector and full-text search engine built on object storage. It’s fast, cheap, and extremely scalable.
@GergelyOrosz
Someone built a FaceID for Mac. https://github.com/jonnyoo/glance
@nullbytes00
I ported Impeccable’s engine to Rust. The design hooks that catch AI slop after each edit now run in ~10ms. • ~4× faster hooks, ~2× faster file scans in our benchmarks • no Node setup to run the skill • the same rule engine across the CLI, browser extension and live mode And GPT Image 2.5 landed today. It’s already in Impeccable 4.3, with faster comps and native transparent assets. Generated illustrations can now sit directly on your own backgrounds. White petals stay white. Gaps between leaves stay transparent. Also shipped since my last update: stricter comp-to-code fidelity checks, per-app design rules in monorepos, DeepSeek Harness support, and a VS Code Marketplace install for Copilot. Impeccable 4.3 is out now. If you haven’t updated in a while, this is a good one.
@pbakaus
The team at @stripe is setting the standard for internal AI platforms: minion coding agents, a custom prototyping rig, and now their company brain, Kai. On today's episode of How I AI, Sharadh shows us how 1.5 engineers and 2 weeks got them a company brain, including: - projects as governance - skill routing + telemetry - a skills platform that works for 10k teammates Plus, he and I debate the merits of gentle parenting your AI (esp when your company is running evals.) Full episode on YT:
@clairevo
having your own personal component library is insanely useful for agentic coding start with shadcn/ui. ask your agent to customize the components. they’re really good at it. then every project starts with: npx shadcn init name/repo your components. your defaults. every time.
@shadcn
What's a feature in Claude Code you wish you'd discovered sooner? Curious what we should surface better
@lydiahallie
This is amazing. Copy this screenshot into Codex. It can solve most issues with Astra not following instructions. From @pvncher.
@daniel_mac8
[Channels Mariah]: It's tiiiiiiiiiime! Today @every is releasing Compound Writing, so now you - yes, you - can install my tendency to overthink a paragraph. I started this past winter by pulling down the Compound Engineering plugin and telling Claude (4.5 at the time, what a model that was, “I want to build this but for writing.” I’ve spent the last six months building and writing with and adding to and tweaking it. The plugin gives Claude Code and Codex skills for interviewing you about an idea, shaping a draft, and questioning whether your argument holds up. When you figure out something useful about your writing, you can save it as guidance for the next piece instead of having the same conversation with the machine indefinitely. I’m standing on the shoulders of giants here in the form of @kieranklaassen and @danshipper, whose work on Compound Engineering and agent-native architectures informed the evolution of this project - what, why, and how. My process, and this plugin, is one way to write with AI. Someone else could start with the same tools and build something completely different, and I’d love to see it. You can use the whole process, borrow one editing skill, or write every word yourself and ask for feedback. No obligation to adopt my entire collection of hang-ups. My hope is that Compound Writing gives you a new playground to discover your own process and grow as a writer. I’ve written a guide to getting started:
@kplikethebird
Introducing SuperAstra: Edit Super Nintendo games live with GPT-6 One of my favorite ways to test new models has been reverse engineering classic games. This is hard because it's raw machine code. Astra is way better than anything I've seen. Download and examples below 👇
@scottastevenson
Omarchy on a M1 MacBook Air with 8gb of ram This thing is dog slow in OS X It screams 🏎️ in Omarchy The gestures in Chonkstep feel so fluid, way better ergonomics than OS X in my opinion. Our test user is my 9 year old who's using Claude and playing ChonkCraft
@SiliconForested
Found a GitHub repo that gives your AI agent 50 marketing specialists for free. It's called marketingskills, built by Corey Haines, and it sits at 46,800 stars with 7,300 forks. The idea is simple. Skills are markdown files. Each one teaches an AI agent how to do one marketing job, with the frameworks and checklists a good marketer carries in their head. There are 50 of them, covering copywriting, CRO, cold email, pricing, SEO audits, A/B testing, churn prevention, ad creative, launch planning, referral programs, and the list keeps going. They're also wired together. One skill called product-marketing holds the context about your product, your audience, and your positioning. Every other skill reads that file first before doing anything. So the copywriting skill and the pricing skill work from the same understanding of what you sell. A few of them caught my eye. marketing-council spins up a simulated board of advisors so you get several expert takes on one question instead of a single answer. marketing-loops sets up recurring workflows an agent runs on a schedule without you re-prompting it. marketing-ideas is a bank of 139 ideas for SaaS products you can pull from when you're stuck. Installing is one command. It works with Claude Code, Codex, Cursor, Windsurf, and anything else that follows the Agent Skills spec. You type "help me optimize this landing page for conversions" and the right skill kicks in on its own. Is a folder of markdown files worth more than a $5,000 a month marketing hire?
@alex_verem
A lot of companies are already doing this. You need really good review agents, make sure it still goes through agentic review for very critical parts of your codebase and have co-workers that are good at assessing risks, and also owning it/fixing it in case things break.
@fatih
life update: after 4 incredible years at Vercel leading the Next.js and React teams, I’m joining OpenAI to work on Codex as an IC. ChatGPT and Codex have completely changed how I build. Getting the chance to help shape where this goes next was too exciting to pass up. I learned an absurd amount at Vercel, and I’m excited to bring some of it with me. onwards :)
@feedthejim
3 years ago I built a test programmatic SEO site that took about 8 hours to build - I found an untapped niche - Scraped a bunch of data - Wrote pSEO directory templates for it (basically madlibs) It's made me $2000 a year ever since with a DR of 9 and just 119 links (all natural) To make that much with the S&P 500 without selling shares I'd need 6 figures invested This costs me $9 for the domain and $60/yr for hosting The crazy part is this: using AI tools I could build a better version of this in 30 minutes! Anybody could.. I have a bunch of these little experiments that for most "normal" people is life changing annual income, but compared to $100k MRR stories on here seem pointless The boring, simple (and hidden) niches are out there if you know where to look
@iannuttall
Codex has changed my life as a solopreneur
@jonnym1ller
9 cool GPT 6 Astra prompts worth trying: 1. The bill renegotiator. "Go through my internet, phone, and software bills, jump into each provider's chat support, and negotiate them down or cancel what I'm not using." 2. Turn an agency into software. “Pick one service business in [niche] and reverse-engineer the exact workflow they sell to clients. Break it into steps, tools used, inputs, outputs, human judgment points, and places where the work gets slow or expensive. Then design the simplest AI product that could replace the first version of that service and charge $500-$5,000/month.” 3. Garage sale flipper. "Watch Facebook Marketplace and Craigslist in my city for [cameras / furniture / bikes] listed way under market, and text me the second one's mispriced with the link." 4. Create my 1 person company dashboard. “Look at my docs, notes, Stripe exports, analytics, customer calls, and project list, then build a weekly operator dashboard. I want to know what is making money, what is wasting time, what customers are asking for, what I should stop doing, and the three highest-leverage actions for next week. Be blunt and show your work.” 5. Audit my company for agent opportunities. “Look at how this business works and find the tasks we should give to agents before hiring another person. For each task, estimate the current human time, the cost of mistakes, the tools involved, the difficulty of automating it, and the first safe version we could deploy. Prioritize things that save money or create revenue within 30 days.” 6. Be my browser operator. “Use the browser to complete this workflow: [workflow]. As you go, click through the actual sites, collect the data, fill the forms where appropriate, and keep notes on what broke or slowed you down. When you’re done, give me the output, the repeatable SOP, and the automation plan so this can become an agent.” 7. The whole QA team. "Every night, open my app on a real phone, go through signup, checkout, and the main flows, and screenshot anything that's broken or confusing." 8. The competitor spy. "Sign up for my top 3 competitors, sit inside their product and their emails, and send me a monthly report on every new feature, price change, and thing they do better than us." 9. Make a game people would actually play for 5 minutes as a lead magnet “Build a browser game around this mechanic: [mechanic]. Don’t just make a cute demo; add progression, tension, scoring, failure, polish, and one reason someone would send it to a friend. Then add lead capture (email/sms), it needs to tie into my core product which sells XYZ.”
@gregisenberg
At Shopify we've built one of the most widely used internal coding agents in the world. Over time we've made it more flexible, and recently our Infrastructure Security team started using River to automate autonomous vulnerability remediation. It's had a massive impact: 👇
@jevon
Your input needed: would you use this? This is an early look at how we're thinking about making Claude Code way more extensible. It's a little crazy, and very exciting. More details here: https://github.com/anthropics/claude-code/issues/91870
@bcherny
@firstadopter I built something for myself to solve exactly this It's a pipeline that ingests every source I follow (podcasts, YouTube channels like CNBC, substacks etc) and then: summarizes each piece on its own (screenshot attached) [Contd]
@Advait_Ruia
Claude can now use your computer in the background in Claude Cowork and Claude Code. Give it something to do on your desktop and Claude clicks, types, and opens apps just like you would, while you work on something else.
@claudeai