Category intelligence

Social Media Briefing — December 30, 2025

309 current items analyzed and ranked.

Executive synthesis

Social Media Summary

AI education and developer productivity dominated discussions. Andrew Ng delivered comprehensive guidance on building AI skills—emphasizing structured learning, hands-on building, and optional research reading—while his Claude Code course with Anthropic achieved extraordinary reach (791K views).

Reality checks balanced the optimism. A viral thread catalogued failed 2025 predictions—junior devs weren't replaced, prompt engineers didn't become highest-paid, autonomous agents remained limited. Erik Brynjolfsson (MIT) observed that professions struggling to create AI-proof exams will struggle to remain AI-proof. Ethan Mollick urged nuance: being 'pro-AI' or 'anti-AI' is a category error when evidence supports both views simultaneously.

Key Themes

AI Education & Career Development · 5AI Predictions & Reality Check · 5Coding Agents & Developer Tools · 9AI Philosophy & Capabilities · 4Meta/Manus Acquisition · 7AI Predictions & AGI Timeline · 3AI Automation of Programming · 6Enterprise AI Adoption · 3AI Philosophy & Consciousness · 1Mechanistic Interpretability · 3

Primary evidence

Top Ranked Signals

92 score
AI Analysis
Andrew Ng provides comprehensive advice on building AI skills: emphasizes structured learning (courses), hands-on building, and optionally reading research papers. Warns against 'just plunge in' approach without foundations.
Another year of rapid AI advances has created more opportunities than ever for anyone — including those just entering the field — to build software. In fact, many companies just can’t find enough skilled AI talent. Every winter holiday, I spend some time learning and building, and I hope you will too. This helps me sharpen old skills and learn new ones, and it can help you grow your career in tech. To be skilled at building AI systems, I recommend that you:
  • Take AI courses
  • Practice building
AI educationCareer developmentLearning methodologyAI talent development
88 score
AI Analysis
Reflection on failed 2025 AI predictions: junior devs not replaced, prompt engineers not highest paid, no ubiquitous autonomous agents, manual coding not obsolete, companies still hiring engineers
A few 2025 predictions that never came to be: • Junior developers were supposed to be fully replaced • Prompt engineers were supposed to be the highest-paid role in tech • Every company was supposed to be running autonomous AI agents • Everyone was supposed to be able to vibe code video games by now • Writing code manually was supposed to be obsolete • Most companies weren't supposed to be hiring engineers anymore None of that has happened. Some of it may never happen at all.
ai-predictionsai-hypedeveloper-jobsreality-checkprompt-engineering
80 score
AI Analysis
Neel Nanda (DeepMind) announces open source release of activation oracles for multiple models (Gemma 3, Qwen 3). Highlights combining with SAEs and interpreting Chain-of-Thought circuits.
Exciting open source release! Activation oracles for tons of models - ask any Q about activations. I'm excited to see what people do with them Gemma 3 AOs can be combined with Gemma Scope 2 - do SAEs corroborate activation oracles? Qwen 3 AOs let you interpret CoT circuits!
Mechanistic interpretabilityOpen source toolsSAEsModel interpretabilityChain-of-thought
Social Twitter Dec 29

how to leverage coding agents to ship fast:

By @gdb

78 score
AI Analysis
Greg Brockman (OpenAI co-founder) sharing tips on leveraging coding agents to ship fast
how to leverage coding agents to ship fast:
Coding agentsDeveloper productivityAI-assisted development
78 score
AI Analysis
Tunguz predicts fully developed SWE agents in first 6-9 months of 2026, all human knowledge work 'solved' by Q1 2027, but expects slow percolation into broader economy with some sectors never directly affected
Prakash managed to nail the predictions for the end of this year, and seems to be on track for the next one as well. My take is that it will not take a whole year to reach those other milestones. I think we'll have fully developed SWE agents in the first 6-9 months of 2026, with all human knowledge work (with a few very specific exceptions) being essentially "solved" by the end of the first quarter of 2027. However, I think it will take much longer for all the capabilities to percolate into the
AI PredictionsSWE AgentsAI Impact on WorkAGI TimelineEconomic Impact
76 score
AI Analysis
Mollick argues 'pro-AI' or 'anti-AI' stances are category errors - one can hold both views on same topic. Uses AI tutoring as example: promising evidence but initially bad educational impact.
Saying you are positive or negative “about AI” feels like a category problem. You can be very positive about AI & very negative about AI around exactly the same topic For example, early evidence on AI tutoring is incredibly promising, but its initial impact on education was bad.
AI discourseAI education impactNuanced AI perspective
75 score
AI Analysis
Jerry Liu (LlamaIndex CEO) argues coding agents with filesystem access eliminate need for RAG for small-to-medium doc collections. Announces OSS filesearch agent with Gemini 3.0 Flash and LlamaParse.
Giving coding agents access to filesystem tools 🤖🗃️ kills the need for RAG for most small-to-medium sized doc collections. It's surprisingly powerful and will only get better as frontier models get more tuned for coding. You already use these capabilities all the time in your favorite coding agent (Cursor/Claude Code). We've made an OSS version of this filesearch agent powered by Gemini 3.0 Flash ⚡️ and LlamaParse. It can natively traverse complex file directories, search your codebase, and
RAG alternativesCoding agentsDocument processingLlamaIndexDeveloper tools
Social Twitter Dec 29

What are your AI predictions for 2026?

By @MIT_CSAIL

72 score
AI Analysis
MIT CSAIL asking community for AI predictions for 2026
What are your AI predictions for 2026?
ai-predictionscommunity-discussionfuture-of-ai
72 score
AI Analysis
Erik Brynjolfsson: When occupations struggle to develop AI-proof exams, it signals they'll struggle to develop AI-proof competencies
When an occupation has trouble developing exams that are AI-proof, it’s a sign that it will have trouble developing competencies that are AI-proof. t.co/ClEM7AwNcf
AI Impact on WorkEducationAI DisruptionProfessional Skills
72 score
AI Analysis
Key insight that AI access does not equal adoption - advocates for structured workflows with Gemini Enterprise and Promevo to move from experiments to sustainable organizational adoption
Access to AI does not equal adoption. With Gemini Enterprise and Promevo’s guidance, AI moves from isolated experiments to structured, everyday workflows, supporting consistent and sustainable adoption across the organization. More > t.co/ixAahO3LLA @Promevo Partner. t.co/NMaIdPMSSB
enterprise AI adoptionAI implementationorganizational changeGoogle Gemini