R[AI]DAR

Category intelligence

Social Media Briefing

All 416 current items, analyzed and ranked.

Executive synthesis

Social Media Summary

Anthropic's Claude Code dominated the day with a major feature launch: built-in git worktree support enabling parallel agent sessions, announced by engineer @bcherny. A serious safety concern also surfaced—Andriy Burkov documented a bug where multi-agent subagents confused conversation history, leading to unauthorized production commits.

  • François Chollet led a contrarian wave arguing AI coding won't kill SaaS, since code was never the real bottleneck—domain expertise, feedback loops, and distribution are. He pointedly asked whether Anthropic itself uses Slack, Zoom, Figma, and Workday.
  • Andrej Karpathy coined a paradigm progression—'chat → code → claw'—signaling computer-use agents as the next frontier beyond coding assistants
  • Greg Brockman announced a new Codex API via `codex app-server`, while Hamel Husain built a tool automating adversarial code review between Claude Code and Codex
  • Ethan Mollick highlighted the alarming asymmetry between billions spent on model training and minimal investment in independent benchmarking
  • Burkov argued the real opportunity is small teams wiring MCP servers into legacy corporate software, not building new apps

Key Themes

Claude Code Worktree Support · 10AI and SaaS disruption · 10AI agent reliability and safety · 4AI coding tool landscape · 10AI paradigm evolution (Chat → Code → Claw) · 4Multi-Model AI Coding Workflows · 4AI hardware (ASICs) · 2Claude Code Worktrees & Multi-Agent Workflows · 14Mobile OS and AI agents · 3Claude 4.6 Models (Opus & Sonnet) · 2

Primary evidence

All Ranked Signals

Showing 10 of 416 signals 2 remaining
Social Twitter Feb 21

Introducing: built-in git worktree support for Claude Code Now, agents can run in parallel without...

By @bcherny

92 score
AI Analysis

As first covered in Reddit yesterday's Claude Code 2.1.49 changelog, Anthropic engineer @bcherny announces built-in git worktree support for Claude Code, enabling parallel agent sessions without interfering with each other. Major feature launch with very high engagement (9.4K likes, 1M views).

Source details
Introducing: built-in git worktree support for Claude Code Now, agents can run in parallel without interfering with one other. Each agent gets its own worktree and can work independently. The Claude Code Desktop app has had built-in support for worktrees for a while, and now we're bringing it to CLI too. Learn more about worktrees: t.co/JFkD2DrAmT
Claude CodeAI coding toolsdeveloper workflowproduct launch
Social Twitter Feb 21

Cloning any random piece of SaaS is something that could already be done before agentic coding, and ...

By @fchollet

88 score
AI Analysis

Chollet argues that cloning SaaS was always possible and cheap (0.5-1% of target company valuation), and AI coding only marginally changes this to 0.1%. The real barriers are domain expertise, user feedback loops, distribution—not code cost.

Source details
Cloning any random piece of SaaS is something that could already be done before agentic coding, and the economics of it haven't changed meaningfully. Before, writing the clone would cost 0.5-1% of the valuation of the legacy SaaS company. Now it might be 0.1%. It doesn't make a difference -- if you can pull it off profitably today you could also have done it profitably in the past. The code is a very small part of the process of making such a clone successful, and the reason legacy software has
AI and SaaS disruptionSoftware economicsAI coding limitations
Social Twitter Feb 21

Situation: I submitted an error message to Claude (the top most message on the right). Claude then ...

By @burkov

87 score
AI Analysis

Burkov reports a dangerous Claude Code bug: subagents asynchronously updating conversation history caused the model to confuse who said what, leading it to attempt unauthorized git commits. Highlights risks of multi-agent architectures losing track of user vs. agent messages.

Source details
Situation: I submitted an error message to Claude (the top most message on the right). Claude then asked, "Commit these changes?" I have no clue what changes it wanted to commit, so I asked, "What changes?" And this fucker starts committing! After I stopped it and asked, "What the hell," it started to show me an approval modal with the question, "Do you allow me to commit?" I rejected, but it kept asking. Eventually, I made it shut up and showed it this screenshot, and it said that it thoug
AI safetyAgent reliabilityClaude CodeMulti-agent systems
Social Twitter Feb 21

First there was chat, then there was code, now there is claw. Ez

By @karpathy

85 score
AI Analysis

As first explored in Social yesterday, Karpathy coins a progression: 'First there was chat, then there was code, now there is claw'—identifying a new paradigm in AI interaction beyond chat and coding interfaces.

Source details
First there was chat, then there was code, now there is claw. Ez
AI paradigm evolutionAI agentsComputer use
Social Twitter Feb 21

codex provides a very nice api, accessible by running “codex app-server”:

By @gdb

83 score
AI Analysis

OpenAI's Greg Brockman announces that Codex provides an API accessible by running 'codex app-server', enabling programmatic access to Codex's development capabilities.

Source details
codex provides a very nice api, accessible by running “codex app-server”:
OpenAI CodexDeveloper toolsAPI infrastructure
Social Twitter Feb 21

We don't see an avalanche of new groundbreaking apps despite an incredible ease of coding provided b...

By @burkov

82 score
AI Analysis

Burkov argues the software business of tomorrow is small teams installing MCP servers for legacy corporate software and connecting them to agentic LLMs—like traveling salespeople. Claims vibe coding hasn't produced groundbreaking apps because all good ideas were already explored.

Source details
We don't see an avalanche of new groundbreaking apps despite an incredible ease of coding provided by AI because the issue was never the lack of people capable of building apps. The issue was that all ideas worth exploring have been systematically explored by various startup incubators since about 2005. There are just so many apps a normie needs while niche apps don't yield billions. Trying today to build a new Google or Amazon it's like trying to find a golden nugget in a pile of sand sifte
AI business modelsMCP serversVibe codingSoftware economics
Social Twitter Feb 21

I've been having codex review claude so much that I made a thing to automate the ping-pong https://...

By @HamelHusain

82 score
AI Analysis

Hamel Husain built a tool to automate 'ping-pong' code review between Claude Code and OpenAI Codex, having one review the other's work. Very high engagement (730 likes, 81K views).

Source details
I've been having codex review claude so much that I made a thing to automate the ping-pong t.co/pnm4atDhHO
AI coding toolsClaude CodeOpenAI Codexmulti-model workflowscode review
Social Twitter Feb 21

The best way to use AI is an interface to information that lets you deepen and improve your own know...

By @fchollet

80 score
AI Analysis

Chollet states the best AI use is as an interface to deepen your own knowledge and mental models; the worst is outsourcing and forsaking your own cognition.

Source details
The best way to use AI is an interface to information that lets you deepen and improve your own knowledge and mental models. The worst way to use AI is as a crutch to outsource and forsake your own cognition
AI philosophyHuman cognitionAI best practices
Social Twitter Feb 21

Codex for end-to-end dev workflows:

By @gdb

78 score
AI Analysis

Brockman shares Codex being used for end-to-end dev workflows, showcasing broader application of the tool beyond individual code generation.

Source details
Codex for end-to-end dev workflows:
OpenAI CodexDeveloper workflowsAI coding tools
Social Twitter Feb 21

yesterday we chatted with @martin_casado and @sarahdingwang on the pod and he happened to do basic m...

By @swyx

78 score
AI Analysis

swyx highlights Taalas launching their HC1 ASIC achieving 17K tok/s inference on Llama 3.1 8B. Notes the ASIC-GPU performance gap will converge to zero in ~2 years based on HC2 roadmap, urging people to 'build accordingly'.

Source details
yesterday we chatted with @martin_casado and @sarahdingwang on the pod and he happened to do basic math™ on the logic of asics today @taalas_inc launched their HC1 asic that can inference 17k tok/s. Sure, it's a shitty 3.1 8B today which is a 1.5 year gap. But read the details to the HC2 this winter, and do the math — this timeline will converge to 0 in the next 2 years. Build accordingly.
AI hardwareASICsInference optimizationAI infrastructure