Wiredframe Radar logo

Category intelligence

Social Media Briefing — February 20, 2026

492 current items analyzed and ranked.

Executive synthesis

Social Media Summary

The day was dominated by Google's launch of Gemini 3.1 Pro, announced across multiple executives including Demis Hassabis, Jeff Dean, and Noam Shazeer. The model scored 77.1% on ARC-AGI-2 — more than doubling Gemini 3 Pro — with Perplexity CEO Arav Srinivas confirming it as second most-picked engine behind Claude 4.5 on their platform.

  • Andrej Karpathy laid out a compelling vision of ephemeral, AI-generated bespoke software replacing traditional app stores, drawing 905K views
  • François Chollet offered a conceptual breakthrough framing agentic coding as analogous to machine learning — specs as loss functions, agents as optimizers, codebases as opaque models
  • Jeremy Howard flagged a serious security vulnerability: LLMs calling tools outside their allowed list, affecting all major US providers except OpenAI
  • Jerry Liu shared LlamaIndex's internal memo revealing coding agents have reduced implementation costs to near-zero, reshaping engineering team structures
  • A biosecurity RCT study surprised forecasters by showing mid-2025 LLMs provided no meaningful uplift for novices attempting wet lab tasks, a significant AI safety finding

Meanwhile, strong negative sentiment swirled around Anthropic, with Scobleizer's blunt "Anthropic really fumbled" post hitting nearly 1M views — likely tied to Claude Sonnet 4.6 backlash.

Key Themes

Gemini 3.1 Pro Launch · 45Anthropic Fumble / Claude Sonnet 4.6 Backlash · 4Future of Software & App Store Disruption · 3LLM Tool-Calling Security Vulnerability · 4Agentic Coding as Machine Learning · 2AI Coding Tools & Post-IDE Evolution · 12Claude Code / Anthropic Tooling Update · 7Gemini 3.1 Pro Release & Model Comparisons · 1Perplexity Product Expansion · 6Gemini 3.1 Pro Preview Launch · 6

Primary evidence

Top Ranked Signals

Social Twitter Feb 19

Introducing Gemini 3.1 Pro, our new SOTA model across most reasoning, coding, and stem use cases! ht...

By @OfficialLoganK

97 score
AI Analysis

Logan Kilpatrick (Google) announces Gemini 3.1 Pro as their new SOTA model across reasoning, coding, and STEM use cases. Massive engagement indicates a major release.

Introducing Gemini 3.1 Pro, our new SOTA model across most reasoning, coding, and stem use cases! t.co/1fvO3oPTtb
model_releasegeminigooglebenchmarks
Social Twitter Feb 19

Very interested in what the coming era of highly bespoke software might look like. Example from thi...

By @karpathy

95 score
AI Analysis

Karpathy presents a detailed vision for the future of bespoke, ephemeral software generated by LLM agents. He describes vibe-coding a custom cardio tracking dashboard in 1 hour, argues the app store model is outdated, and calls for AI-native APIs/CLIs for all products and services.

Very interested in what the coming era of highly bespoke software might look like. Example from this morning - I've become a bit loosy goosy with my cardio recently so I decided to do a more srs, regimented experiment to try to lower my Resting Heart Rate from 50 -> 45, over experiment duration of 8 weeks. The primary way to do this is to aspire to a certain sum total minute goals in Zone 2 cardio and 1 HIIT/week. 1 hour later I vibe coded this super custom dashboard for this very specific exp
future_of_softwarevibe_codingAI_native_infrastructureagentic_AIapp_store_disruption
Social Twitter Feb 19

Sufficiently advanced agentic coding is essentially machine learning: the engineer sets up the optim...

By @fchollet

90 score
AI Analysis

François Chollet draws a deep analogy between agentic coding and machine learning: the spec+tests are the optimization goal, coding agents are the optimizer, and the generated codebase is a black-box model. Predicts classic ML problems (overfitting, shortcuts, data leakage) will plague agentic coding. Asks what will be the 'Keras of agentic coding.'

Sufficiently advanced agentic coding is essentially machine learning: the engineer sets up the optimization goal as well as some constraints on the search space (the spec and its tests), then an optimization process (coding agents) iterates until the goal is reached. The result is a blackbox model (the generated codebase): an artifact that performs the task, that you deploy without ever inspecting its internal logic, just as we ignore individual weights in a neural network. This implies that a
agentic_codingmachine_learning_theorysoftware_engineeringAI_toolsfuture_of_software
Social Twitter Feb 19

https://t.co/taVRqwm4dG

By @trq212

90 score
AI Analysis

@trq212 shares a link that goes massively viral (657K views, 2573 likes). Based on context of their other replies about caching, model switching, system reminders, and light mode themes, this is likely a major Claude Code or Anthropic developer tooling update.

t.co/taVRqwm4dG
claude_codeanthropicdeveloper_toolsproduct_launch
Social Twitter Feb 19

Piotr discovered something worrying: if you give an LLM a list of tools it's allowed to call, it mig...

By @jeremyphoward

88 score
AI Analysis

Jeremy Howard reports a security vulnerability discovered by Piotr Czapla: LLMs may call tools not provided in their tool list, affecting all major US providers except OpenAI. He calls it an instance of Simon Willison's 'lethal trifecta'.

Piotr discovered something worrying: if you give an LLM a list of tools it's allowed to call, it might decide to also call a tool you didn't provide! Impacts all major US providers except @OpenAI. Be sure to check LLM tool call requests! (Lisette/Claudette check automatically)
llm_securitytool_callingvulnerabilityai_safety
Social Twitter Feb 19

Gemini 3 Pro has been upgraded to Gemini 3.1 Pro for all Perplexity Pro and Max users (consumer and ...

By @AravSrinivas

85 score
AI Analysis

Arav Srinivas announces Perplexity has upgraded Gemini 3 Pro to Gemini 3.1 Pro for all Pro/Max users. Notes Gemini 3.1 Pro is the second most-picked enterprise model after the Claude 4.5 Sonnet/Opus family.

Gemini 3 Pro has been upgraded to Gemini 3.1 Pro for all Perplexity Pro and Max users (consumer and enterprise). It's the second most picked model by our Enterprise customers after Claude 4.5 Sonnet/Opus family. Enjoy! t.co/E5SH1WxnH5
perplexitygeminiclaudeenterprise_adoptionmodel_preferences
Social Twitter Feb 19

Excited to launch Gemini 3.1 Pro! Major improvements across the board including in core reasoning an...

By @demishassabis

83 score
AI Analysis

Demis Hassabis announces Gemini 3.1 Pro launch with major improvements in reasoning, 77.1% on ARC-AGI-2 (2x Gemini 3 Pro), rolling out across Gemini App and Antigravity.

Excited to launch Gemini 3.1 Pro! Major improvements across the board including in core reasoning and problem solving. For example scoring 77.1% on the ARC-AGI-2 benchmark - more than 2x the performance of 3 Pro. Rolling out today in @GeminiApp, @antigravity and more - enjoy! t.co/hOgEFtJ57w
Gemini_3.1_Promodel_releasesbenchmarksARC_AGI
Social Twitter Feb 19

Today, we’re continuing to push the boundaries of AI with our release of Gemini 3.1 Pro. This updat...

By @JeffDean

82 score
AI Analysis

Jeff Dean officially announces Gemini 3.1 Pro release, highlighting 77.1% on ARC-AGI-2 (more than double Gemini 3 Pro), with side-by-side comparison of code generation quality.

Today, we’re continuing to push the boundaries of AI with our release of Gemini 3.1 Pro. This updated model scores 77.1% on ARC-AGI-2, more than double the reasoning performance of its predecessor, Gemini 3 Pro. Check out the visible improvement in this side-by-side comparison, showing Gemini 3.1 Pro’s crisp animation built with pure code. Read more about today’s 3.1 Pro update: t.co/vABdcMSE3f
Gemini_3.1_Promodel_releasesbenchmarksARC_AGI
Social Twitter Feb 19

The model is a step forward in reasoning, designed for workflows where a simple answer isn’t enough....

By @GoogleDeepMind

82 score
AI Analysis

Google DeepMind highlights Gemini 3.1 Pro's reasoning improvements, noting it more than doubles Gemini 3 Pro's score on ARC-AGI-2 benchmark for novel logic patterns.

The model is a step forward in reasoning, designed for workflows where a simple answer isn’t enough. On ARC-AGI-2 – which tests for novel logic patterns – it more than doubles 3 Pro’s score. This means it can help you visualize complex topics, organize scattered data, and bring creative projects to life.
Gemini 3.1 Pro launchAI benchmarksreasoning capabilities
Social Twitter Feb 19

Comet iOS is almost there! You can pre-order now! As promised: it will be smooth as butter and feel ...

By @AravSrinivas

82 score
AI Analysis

Arav Srinivas announces Comet iOS, Perplexity's AI-powered browser, is available for pre-order. Described as 'Safari grade browser with Perplexity powering every webpage'.

Comet iOS is almost there! You can pre-order now! As promised: it will be smooth as butter and feel like Safari grade browser with Perplexity powering every webpage and the browser for assistance. t.co/fnUQHJEdUw t.co/4hks13Li07
perplexityproduct_launchai_browsermobile
Social Twitter Feb 19

Anthropic really fumbled.

By @Scobleizer

82 score
AI Analysis

Scobleizer declares 'Anthropic really fumbled' — with massive engagement (2604 likes, 947K views). This appears to reference a major issue with Anthropic's Claude Sonnet 4.6 (released Feb 17, two days before).

Anthropic really fumbled.
anthropicclaude-sonnet-4.6model-releasesai-industry