Wiredframe Radar logo

Category intelligence

Social Media Briefing — March 10, 2026

524 current items analyzed and ranked.

Executive synthesis

Social Media Summary

Andrej Karpathy's autoresearch experiment dominated the AI conversation — an autonomous agent found ~20 improvements to nanochat over two days, cutting 'Time to GPT-2' by 11%. This sparked broad discussion about AI self-improvement loops, with Harrison Chase (LangChain) building an "autoresearch for agents" variant.

A strong contrarian thread emerged around agent reliability: svpino's viral post declaring "these agents don't work as promised" drew 668K views and resurfaced concrete failures from Chevrolet, Air Canada, and others. Ethan Mollick noted that six weeks after Claude Cowork launched, no competitor has emerged — a telling signal about the gap between lab claims and shipped products.

Key Themes

Autoresearch and AI-Driven Optimization · 10Claude Code Review Launch · 7Anthropic vs US Government Conflict · 4AI Agent Reliability & Failures · 8OpenAI Strategic Moves · 2AI Acqui-hire & Open Source Market Dynamics · 4Perplexity Computer Expansion · 4Multi-Agent Architecture & Verification · 6AI Coding Tools Competition · 7Agent Optimization & Autonomous Improvement · 5

Primary evidence

Top Ranked Signals

Social Twitter Mar 9

Three days ago I left autoresearch tuning nanochat for ~2 days on depth=12 model. It found ~20 chang...

By @karpathy

97 score
AI Analysis

Building on yesterday's Reddit discussion of autoresearch, Karpathy's major post: autoresearch agent autonomously found ~20 improvements to nanochat over 2 days, reducing 'Time to GPT-2' by 11%. Improvements included fixing attention scaling, regularization, attention bandwidth, AdamW betas, weight decay, and initialization. He predicts all frontier labs will adopt agent-driven optimization and envisions multi-agent collaboration for research at scale.

Three days ago I left autoresearch tuning nanochat for ~2 days on depth=12 model. It found ~20 changes that improved the validation loss. I tested these changes yesterday and all of them were additive and transferred to larger (depth=24) models. Stacking up all of these changes, today I measured that the leaderboard's "Time to GPT-2" drops from 2.02 hours to 1.80 hours (~11% improvement), this will be the new leaderboard entry. So yes, these are real improvements and they make an actual differen
autoresearchAI agentsneural network optimizationfuture of ML researchautonomous AI research
Social Twitter Mar 9

New in Claude Code: Code Review. A team of agents runs a deep review on every PR. We built it for o...

By @bcherny

95 score
AI Analysis

bcherny (Anthropic) announces Claude Code Review: a team of agents that performs deep review on every PR. Reports Anthropic engineer code output is up 200% this year with reviews being the bottleneck. Says it catches real bugs he wouldn't have noticed.

New in Claude Code: Code Review. A team of agents runs a deep review on every PR. We built it for ourselves first. Code output per Anthropic engineer is up 200% this year and reviews were the bottleneck Personally, I’ve been using it for a few weeks and have found it catches many real bugs that I would not have noticed otherwise
claude-codeai-code-reviewproduct-launchdeveloper-productivitymulti-agent-architecture
Social Twitter Mar 9

NEW: Anthropic just filed two lawsuits against the U.S. government 👀 The complaint: "The Constituti...

By @TheRundownAI

88 score
AI Analysis

Following Saturday's News coverage of the Pentagon-Anthropic standoff, Anthropic filed two lawsuits against the US government, alleging retaliation after refusing to drop Claude restrictions on autonomous lethal warfare and mass surveillance

NEW: Anthropic just filed two lawsuits against the U.S. government 👀 The complaint: "The Constitution does not allow the government to wield its enormous power to punish a company for its protected speech." It also says officials are "seeking to destroy the economic value created by one of the world's fastest-growing private companies." Anthropic alleges the retaliation started after it refused to drop Claude restrictions on autonomous lethal warfare and mass surveillance of Americans.
AnthropicAI_safetyAI_policygovernment_regulationlegalAI_ethics
Social Twitter Mar 9

We’re acquiring Promptfoo. Their technology will strengthen agentic security testing and evaluation...

By @OpenAI

85 score
AI Analysis

OpenAI announces acquisition of Promptfoo, an open-source security testing and evaluation tool. Technology will strengthen agentic security testing in OpenAI Frontier. Promptfoo will remain open source.

We’re acquiring Promptfoo. Their technology will strengthen agentic security testing and evaluation capabilities in OpenAI Frontier. Promptfoo will remain open source under the current license, and we will continue to service and support current customers. t.co/xhmLmJRoUZ
OpenAI acquisitionAI securityevaluationagentic AIopen source
Social Twitter Mar 9

I'm excited to announce Context Hub, an open tool that gives your coding agent the up-to-date API do...

By @AndrewYNg

82 score
AI Analysis

Andrew Ng announces Context Hub, an open-source tool that provides coding agents with up-to-date API documentation. Solves the problem of agents hallucinating outdated APIs. Agents can annotate docs with notes for future sessions, with plans for agents to share learned workarounds community-wide.

I'm excited to announce Context Hub, an open tool that gives your coding agent the up-to-date API documentation it needs. Install it and prompt your agent to use it to fetch curated docs via a simple CLI. (See image.) Why this matters: Coding agents often use outdated APIs and hallucinate parameters. For example, when I ask Claude Code to call OpenAI's GPT-5.2, it uses the older chat completions API instead of the newer responses API, even though the newer one has been out for a year. Context H
developer toolsAI agentsopen sourcecoding agentsAPI documentation
Social Twitter Mar 9

btw if you can build a category leader open source project in ai engineering right now the market ac...

By @swyx

82 score
AI Analysis

swyx argues that building a category-leading open-source AI engineering project can lead to acqui-hires at $10-100M per engineer. No need for business model, GTM, or funding - just build what companies want

btw if you can build a category leader open source project in ai engineering right now the market acquihire rate is ~$10-$100m per ai engineer. you do not need to figure out a business model, you do not need GTM, you do not need funding. just build things clankers want.
AI_businessopen_sourceAI_talentacqui_hireAI_engineering
Social Twitter Mar 9

@Rahll 👋 Roughly, the more tokens you throw at a coding problem, the better the result is. We call t...

By @bcherny

78 score
AI Analysis

Echoing yesterday's Social discussion on multi-agent systems, bcherny (Anthropic) explains how test-time compute and separate context windows make subagents effective for code review. Argues multiple uncorrelated context windows help catch bugs similarly to how different engineers catch each other's mistakes. Predicts agents will eventually write perfect bug-free code.

@Rahll 👋 Roughly, the more tokens you throw at a coding problem, the better the result is. We call this test time compute. One way to make the result even better is to use separate context windows. This is what makes subagents work, and also why one agent can cause bugs and another (using the same exact model!) can find them. In a way, it’s similar to engineers — if I cause a bug, my coworker reviewing the code might find it more reliably than I can. In the limit, agents will probably write p
test-time-computemulti-agent-architectureai-code-reviewai-engineering-philosophy
Social Twitter Mar 9

"Build a company that benefits from the models getting better and better" — @sama devin brain uses ...

By @swyx

75 score
AI Analysis

swyx provides detailed analysis of Cognition's Devin: uses dozens of model groups, extensively evals every model, does complete rewrites every few months. Notes agents are 'really working now' and credits long-term investment since Nov 2023

"Build a company that benefits from the models getting better and better" — @sama devin brain uses a couple dozen modelgroups and extensively evals every model for inclusion in the harness, doing a complete rewrite every few months. hearing a lot of "devin is good now" feedback but its largely the same process that the team has been running since @ScottWu46 bet on cloud agents in November 2023. agents are really, really working now and you had to have scaled harness eng + GTM to prep for this m
AI_coding_agentsCognitionDevinmodel_evaluationAI_engineeringmulti_model
Social Twitter Mar 9

Perplexity Computer can be connected to your Google and Meta Ads APIs. When you do that, it can run ...

By @AravSrinivas

75 score
AI Analysis

Arav Srinivas announces Perplexity Computer can connect to Google and Meta Ads APIs to autonomously run ad campaigns at superhuman frequency.

Perplexity Computer can be connected to your Google and Meta Ads APIs. When you do that, it can run your ad campaigns autonomously at a frequency that’s not possible to match humanly.
Perplexity Computerautonomous agentsadvertising automationproduct launch
Social Twitter Mar 9

Last week we debuted our new real-time video agents with one of the hardest demos possible: live tel...

By @c_valenzuelab

73 score
AI Analysis

Cristóbal Valenzuela (Runway CEO) announces BBC is using Runway Characters for live television programming with their new real-time video agents.

Last week we debuted our new real-time video agents with one of the hardest demos possible: live television. The BBC is now using Runway Characters to augment segments of their programming. Wild to see this live. So excited for all the new possible applications to come. t.co/E0JgOYx33Z
Runwayreal-time video AIenterprise adoptionBBCAI in media