A personal AI news intelligence system for a CS grad student pursuing startups and SDE/MLE/AI roles.
Every day gets its own folder. Every story has a source and a direct "why it matters to you" section.
YYYY-MM-DD/
├── 00-tldr.md # 60-second skim: 5–8 bullets + one action + watchlist deltas
├── 01-big-lab-moves.md # OpenAI, Anthropic, Google, Meta, xAI, Apple — strategy, products, policy
├── 02-new-emerging.md # New models, startups, tools, funding rounds, paradigm shifts
├── 03-practical-skills-and-tools.md # Hands-on workflows, tools, prompting, productivity — act on this TODAY
├── 04-research-progress.md # arXiv papers, benchmarks, breakthroughs — what's moving the frontier
└── 05-career-and-startup.md # Job market, VC trends, skills to build, startup playbook
Every entry has:
- What happened — the fact
- Sources — verified links, no secondhand summaries without attribution
- Why it matters to you — direct implication for startup or job hunt (three lenses: Job · Startup · Insight)
- Tags —
#anthropic #funding #voice #agentsetc., so you cangrep -r "#voice" .across the archive
Cross-cutting docs at repo root:
ME.md— profile / goals / focusing decisions; every edition is written to this profileSOURCES.md— master tiered source list (8 tiers, ~100 sources)WATCHLIST.md— running threads that span multiple days, so nothing drops between editionsACTIONS.md— personal task tracker (extracted from WATCHLIST so what-to-do-this-week is scannable in one file) NEW 2026-05-19STARTUPS.md— running wedge log with market signals, anchor competitors, and your-fit scores NEW 2026-05-19APPLICATIONS.md— job / outreach / program application tracker; closes the loop between weekly "apply this week" notes and outcomes NEW 2026-05-19
| Date | Highlights |
|---|---|
| 2026-05-22 | THE STATE STEPS BACK, THE LABS GO PUBLIC, THE TALENT MAP REDRAWS — all in 24 hours. Trump's AI executive order was POSTPONED (reversal of yesterday's "signing today") — Trump "didn't like certain aspects" and "hates regulation" ("I don't want to get in the way of [US AI] leading"); draft survives = voluntary 90-day pre-release frontier review + Treasury-led cybersecurity "clearinghouse" (negotiated w/ Cairncross + OpenAI/Anthropic/Reflection AI) → pre-deployment-eval career lane delayed, not dead · OpenAI files a CONFIDENTIAL S-1 as early as today — targeting Sept 2026 IPO at ~$852B–$1T (Goldman + Morgan Stanley; financials private until ~15d pre-roadshow), unblocked after Musk lost his lawsuit; Anthropic eyeing October → equity goes liquid, hiring gets structured, public S-1 = the best revenue-by-segment hiring map you'll get · Andrej Karpathy joined Anthropic (announced Tue, started this week) — OpenAI founding member → Tesla → OpenAI → Eureka Labs → pre-training team launching a "use Claude to accelerate Claude's training" group = recursive self-improvement, staffed; loudest talent signal of 2026 → validates the Anthropic-stack focusing decision · Emerging: the IPO wave itself — SpaceX + OpenAI + Anthropic public inside ~12 mo = frontier AI becomes a public-market asset class · Exaforce $125M Series B (HarbourVest/Peak XV/Mayfield/Khosla; total $200M) — agentic SOC: real-time security knowledge graph + agents ("Exabots"), ~10× faster investigations, fewer tokens; pairs w/ the EO's surviving cyber half = thin, two-tailwind hiring lane · Research: real-tool agent benchmarks — MCP-Atlas (Scale; real MCP servers, agent must discover tools) + Tool Decathlon/Toolathlon (ICLR 2026; 32 apps, 604 tools — K8s/BigQuery/Notion); eval bar moved from mocks to your actual stack; "Agentic Reasoning" survey gives the 3-layer taxonomy (foundational/self-evolving/collective) · Practical: the agent-team cost lever — Opus-4.7 orchestrator + Sonnet-4.6 workers ≈ 40% cheaper than all-Opus + the plan→annotate→"address all notes, don't implement yet" reliability loop (T-24 to June-15 metering) · Friday action: ship the dual-model sanitiser project reframed around real-tool verification (cite MCP-Atlas/Toolathlon) + per-step cost; 30-min Meta-alumni reply window |
| 2026-05-21 | THE STATE STEPS IN — pre-release review goes federal, the compute bill comes due, the machine does new math. Trump signs an AI/cybersecurity executive order today — first US federal framework asking labs to hand "covered models" to government 90 days pre-release (+ pre-access for critical infra like banks); labs lobbying for 14 days; driven by cyber-risk (Mythos, GPT-5.5-Cyber named); OpenAI + Anthropic at the table → opens a "pre-deployment evaluation / AI-assurance" job market inside labs, banks, and GRC startups · Anthropic's Colossus tenancy now CONTRACTUAL in SpaceX's S-1: $1.25B/month through 2029 (~$15B/yr, $40B+ total) — entire Colossus 1 (300MW, 220K+ GPUs H100/H200/GB200); the May-9 rumor is now a filed liability · Anthropic projects its first profitable quarter — ~$559M operating profit ~2 yrs early, Q2 rev ~$10.9B → aggressive revenue-role hiring, de-risks the Anthropic-stack focusing decision · An OpenAI general-purpose model disproved an 80-year Erdős conjecture (planar unit-distance problem; infinite family of better constructions via algebraic number theory; verified by Noga Alon + Thomas Bloom) — first prominent open problem advanced autonomously by a general model · ChatGPT Ads Manager opens to all (CPC, no minimum spend; $2.5B→$100B/yr target) — direct contrast with Anthropic's ad-free pledge · Scout AI $100M Series A (largest US defense-tech Series A; Fury model for unmanned warfare); climate narrowed to edge / scarce talent / stack-control ($18.8B into post-2025 AI startups) · arXiv live-benchmark wave: LemmaBench (live research-grade math) · RepoReason (repo-level reasoning) · PostTrainBench · single→multi-agent eval · test-time scaling · Practical: the 2026 Claude Code orchestration stack (CLAUDE.md + subagents + MCP + hooks) → ship a hook-guarded MCP mini-agent + cost trace · Skill re-price: verification/eval design + cost-aware routing got scarce; raw prompting got commoditized · Thursday action: send 10 pre-staged Meta-alumni DMs at 8 AM PT (segment pools a/b/c), add pre-deployment-eval + bank-AI-assurance to apply list, ship the orchestration artifact this weekend |
| 2026-05-20 | THE MORNING AFTER — I/O scorecard + Meta cut executing. Google I/O graded ~7/9 predictions hit: Gemini 3.5 Flash GA same-day at $1.50/1M in · $9/1M out (1M ctx, multimodal in), "within 2 points of Anthropic's flagship at ~⅓ the price," already GA in GitHub Copilot; Gemini 3.5 Pro ships June · WebMCP — the under-weighted headline — an open web standard built on Anthropic's MCP lineage, origin trial in Chrome 149 (sites expose callable tools; agents call instead of scrape; MCP is now the industry default) · Antigravity 2.0 + Managed Agents in the Gemini API + ADK 2.0 — "one API call → sandboxed agent that reasons/uses tools/executes code," near-verbatim Anthropic Managed Agents → agent-runtime primitive is now table stakes; Chrome DevTools-for-agents supports 20+ non-Google agents · AI Ultra $100/mo + Gemini Spark 24/7 proactive consumer agent · Code w/ Claude London: NO new model — ratified an already-shipped roadmap (Dreaming/Outcomes/multi-agent/Claude Finance/Add-ins); re-confirmed Colossus 1 (220K+ GPUs/300+MW), Claude Code 5-hr limits doubled + peak throttle lifted, ~80× YTD demand · Meta 8,000-cut EXECUTING today (≈10%; +6K canceled reqs = ~14K; Singapore→UK→US; ~7,000 redirected into new AI teams: Applied AI Engineering / Agent Transformation Accelerator XFN / Central Analytics; $145B AI infra; H2 cuts planned) · Prompt injection now empirical: Google threat report +32% malicious IPI (Nov 25→Feb 26) with real PayPal payloads hidden in HTML, recommended fix = cheap dual-model "sanitiser" = same primitive as TrajAD verifier + JADE per-claim eval (one primitive, three research lanes, enabled now that Flash made guard-models ~free) · arXiv: DyTopo · RuleSmith · CommCP (conformal prediction over agent messages) · Wednesday action: publish graded I/O comparison table + fix LinkedIn skills to real terms (Antigravity/Managed Agents/WebMCP, NOT "Vertex AI Agent Platform"); apply 1 OpenAI FDE + 1 Anthropic Solutions; ship dual-model sanitiser + cost-router this week |
| 2026-05-19 | GOOGLE I/O 2026 DAY (Tue 10 AM PT) — final pre-keynote consensus: Gemini 3.5 (most leaks now name 3.5, not 3.2 Flash or Gemini 4), Gemma 4 open-weights, Gemini Omni unified video+image+audio + chat-edit, Gemini Spark / Remy proactive agent, Android XR Gen 2 (Samsung Galaxy XR + Warby Parker + Gentle Monster), Aluminium OS / Googlebook formal name + OEM ship windows (Acer/Asus/Dell/HP/Lenovo), Android 17 SDK with system-level agent hooks, Vertex AI Agent Platform pricing (current rate $0.0864/vCPU-hr + $0.25 per 1K stored events; Gemini 3 endpoint pricing live July 1) · CORRECTION: Code w/ Claude London is TODAY May 19 (not May 20–21), Tokyo June 10 (not June 5–6) — same-day collision with I/O = sharpest counter-programming move of 2026. Day-1 livestream + customer presenters Asana · Cursor · GitHub · Replit · Vercel · Meta May 20 layoff T-minus 1 — 8K cuts confirmed = 10% of 78,865 workforce; restructure under CAIO Alexandr Wang's Superintelligence Labs pods; $115–135B 2026 AI infra spend; severance 16 wks + 2 wks/yr + 18 mo health · Isomorphic Labs $2.1B Series B confirmed: Thrive lead + Alphabet/GV + MGX + Temasek + UK Sovereign AI Fund = $2.6B total; CEO Demis Hassabis; partners Novartis + Lilly + J&J; first publicly-disclosed Lab+VC+Sovereign+Industry four-corner template · Sierra $950M at $15B (GV + Tiger Global lead; 3-year-old; largest CX-agent valuation) · Parallel Web Systems $100M Sequoia (Parag Agrawal; total $230M; agent search/research infra) · Runware $50M Series A ("Sonic Inference Engine"; 2M+ HF models EOY target) · Oboe $16M (personalized course generation) · OpenAI Deployment Company $4B + Tomoro acquisition (~150 FDEs) — TPG lead, 19-investor consortium, biggest FDE-market gravity shift of 2026 · Anthropic ad-free policy commitment (first explicit Claude post) + Workday Foundation × Anthropic Solopreneurship Accelerator (15 slots, apply this week) · arXiv: AIRS-Bench (20-task science-agent benchmark, frontier 17–34%) · JADE (per-claim eval against expert KB) · TrajAD (Haiku-verifier + Opus-agent rollback, ~10× ratio) · AgentScope distributed multi-agent · Tuesday action: run 15-min-block I/O monitoring + publish 1-page Gemini-vs-Claude-vs-GPT post by 12:30 PM PT + 12-min Code w/ Claude London slice at 1 PM PT + 5 PM PT follow-up post + update LinkedIn keywords by 11:55 AM PT |
| 2026-05-18 | T-1 to Google I/O 2026 (Tue 10 AM PT) — final pre-keynote consensus: Gemini 3.2 Flash (not a Gemini 4 rebrand), Gemma 4 open-weights, Android 17 SDK, Aluminium OS / Googlebook desktop launch + OEM ship windows, Android XR Gen 2 (Samsung + likely Warby Parker), Vertex AI Agent SDK pricing (the line item to watch most carefully) · Anthropic counter-programs I/O — Code w/ Claude London May 20–21 opens 36 hours after Sundar walks off stage; Tokyo June 5–6. Keynote panel: Ami Vora · Boris Cherny · Angela Jiang. Customer presenters: Asana, Cursor, GitHub, Replit, Vercel (the 5 dev-tools partners Anthropic cannot afford to lose to Vertex Agent SDK) · Meta May 20 layoff T-2: 8K cuts + 6K req cancellations = 14K effective. $135B 2026 AI infra spend; new Superintelligence Labs pods under CAIO Alexandr Wang; severance 16 wks + 2 wks/yr + 18 mo health · Isomorphic Labs closed $2.1B Series B (May 12). Thrive lead + Alphabet/GV + MGX + Temasek + CapitalG + UK Sovereign AI Fund = $2.6B capital base. First Lab+VC+Sovereign+Industry four-corner template · Mustafa Suleyman 18-month "white-collar automation" forecast named accounting/legal/marketing/PM as first to fall — read it as a Microsoft Copilot-attach commercial vehicle, not a literal forecast; the four named verticals are the FDE/Integration-Engineer TAM expansion · Ramp AI Index — three enumerable threats to Anthropic's 2.1pt adoption lead: incentive misalignment · cheap-inference-platform attach growth · Claude Code concentration · Anthropic $30B raise at $900B+: no term sheet signed as of May 18 (3rd consecutive week characterized as "imminent") · Agent SDK June 15 credit doesn't auto-activate — manual toggle required (5-min fix tonight, silent failure if skipped) · arXiv: CHAL (hierarchical agent dialectic) · MemReread (memory-guided rereading for long-context) · ARIS (cross-model adversarial-collab research harness — open-source counter to Anthropic's Dreaming) · Storage Is Not Memory (the storage/recall/memory trichotomy) · Multimodal Procedural Knowledge (visual agents with reusable skill cards) · Monday action (60 min): pre-stage I/O comparison doc · toggle Agent SDK credit · apply to 2 FDE roles before the Thursday Meta-cohort flood |
| 2026-05-17 | T-2 to Google I/O 2026 (Tue May 19, 10 AM PT) — Gemini "Omni" (unified image+video+audio + NL video editing) leak hardens with public clips; Aluminium OS desktop reveal expected; Android XR Gen 2 (Samsung Galaxy XR + rumored Gentle Monster); Vertex Agent SDK watchpoint · Anthropic × Gates Foundation $200M / 4-yr for global health (polio, HPV, eclampsia/preeclampsia, vaccine R&D), K-12 tutoring + sub-Saharan Africa/India literacy, smallholder-farming agriculture — Anthropic's 5th distinct distribution channel in 10 days · OpenAI ships Codex to ChatGPT mobile (May 14, all plans incl. Free/Go) — phone-as-control-surface for async coding agents · FDE postings +800% YoY — Google Cloud 59 open roles, Salesforce/Anthropic/Palantir/Cohere/Databricks/EY hiring; $215–310K base senior, $500K+ TC at frontier labs · Prompt caching = 60–90% input-cost savings (the highest-ROI mitigation before the June 15 Agent SDK metering change) · Karpathy CLAUDE.md at ~109K stars / 28 wks #1 trending — drop it into every project tonight · arXiv: DyTopo (dynamic agent topology rewiring) · AIRS-Bench (20-task science-agent benchmark) · TrajAD (runtime trajectory verifier w/ precise rollback) · "Bayes-consistent orchestration" position paper · Sunday action: drop CLAUDE.md + enable prompt caching + apply to 2 FDE roles |
| 2026-05-16 | Anthropic puts Claude agents on a meter — June 15. Programmatic Claude (Agent SDK, claude -p, GitHub Actions, OpenClaw) moves to separate credit pool billed at API list rates: Pro $20 / Max-5x $100 / Max-20x $200. 30-day window to audit your own bill before the subsidy disappears. · Claude for Small Business shipped (May 13) — toggle Claude inside QuickBooks/PayPal/HubSpot/Canva/DocuSign/Workspace/MS365 with 15 ready-to-run workflows; SMBs = 44% of US GDP; free 10-city in-person tour started May 14 · OpenAI ChatGPT Personal Finance (May 15) with Plaid + 12K+ FIs (Schwab, Fidelity, Chase, Robinhood, Amex, Cap One); Pro-only US preview, Intuit coming · Google I/O T-3 days — "Gemini Omni" leak strengthens (unified video+audio, NL editing) · GridCARE $64M Series A for AI-data-center power acceleration · Sprouts.ai $9M (Revenue Agents) + Nectar Social $30M (agentic marketing OS) · arXiv "Cattle Trade" multi-agent bluffing/bidding benchmark · Career lens: "AI Integration Engineer" is this week's under-priced lane |
| 2026-05-15 | Anthropic in advanced talks to acquire Stainless (≥$300M) — would own the SDK toolchain shipping OpenAI, Google, Meta, Cloudflare client libraries · PwC × Anthropic alliance expansion: 30,000 trained + certified on Claude Code, scaling to 364K global; Claude-native Finance practice spins up · Google I/O preview (May 19 keynote — Gemini 4, Remy + Spark agents, Android 17 SDK, Googlebook SDK) · AI Engineer = #1 fastest-growing US job title (+143% YoY; $206K avg; AI-skill wage premium jumped 25%→56% in 12 months) · Karpathy 4-rule CLAUDE.md playbook · Weekend project: ship one MCP server · arXiv "Attractor Models — Solve the Loop" (fixed-point latent reasoning) · "Many Faces of On-Policy Distillation" unified taxonomy · Chapter Medicare-AI $100M Series E (Generation IM) + Performativ €5.5M + Marloo $10M — vertical-AI-for-regulated-industries thesis hardens |
| 2026-05-14 | Anthropic overtakes OpenAI in US business adoption for the first time (Ramp AI Index: 34.4% vs 32.3%; Anthropic ~4×'d adoption in 12 months) · Anthropic raise talks now at up to ~$950B · US + China agree to launch a formal AI safety protocol at the Trump–Xi Beijing summit · Google's "Googlebook" confirmed — Aluminium OS, Gemini as the OS layer · Cisco +15% on $9B AI-infra order guidance while cutting ~4,000 jobs · Claude Code now ~4% of all public GitHub commits · OpenClaw hits 210K+ stars · Appier "Answer, Refuse, or Guess?" — LLMs miscalibrated on risk · Hint (Martha Stewart) $10M seed |
| 2026-05-13 | Anthropic "Claude for Legal" — 12 practice-area plugins + 20+ MCP connectors (DocuSign, Ironclad, iManage, NetDocuments, LexisNexis, Thomson Reuters, Box, Everlaw); TR CoCounsel Legal now rebuilt on Claude Agent SDK · Google Threat Intel: first-ever AI-built zero-day caught in active mass-exploitation campaign (2FA bypass) · Meta Avocado/Mango delays + closed-source pivot confirmed at C-suite · Wispr Flow in talks at ~$2B ($260M Menlo Ventures lead) — "Voice OS" rebrand · Judgment Labs $32M Seed + Series A (Lightspeed ×2) for deep-agent eval · arXiv 2602.16666 Agent Reliability — 12 metrics, "reliability decoupling" thesis · Q1 2026 layoffs revised to 78,557 / 47.9% AI-attributed · Meta 8,000-person cut scheduled May 20 |
| 2026-05-12 | Google "Android Show: I/O Edition" — Aluminium OS desktop platform reveal (HP/Lenovo/Acer/ASUS/Samsung), Android 17 agentic features, Android XR glasses, system-level Gemini · Anthropic $50B/$900B raise board decision expected this week · EU vs Mythos escalation: Spain Minister Cuerpo publicly cites AI Act Article 51 (Aug enforcement window) · Cognition (Devin) raising at $25B, 80× enterprise growth · xAI Speech-to-Text + TTS + Grok Imagine Quality Mode GA · On-Policy Distillation sweep (Thinking Machines Lab + survey + SDPO + OPSD) · Constraint Decay arXiv paper (May 7) — quantifiable failure mode in coding agents · CS new-grad Q1 data: 52,050 layoffs · MLE +41.8% YoY vs SWE -40% |
| 2026-05-11 | IBM CAIO study: 76% of orgs now have a Chief AI Officer (up from 26% in 2025) · Anthropic ARR crosses ~$44B (May), full monthly progression disclosed · NVIDIA Nemotron 3 Nano Omni open-source (30B-A3B hybrid Mamba-Transformer, 9× throughput) · Apple iOS 27 multi-AI "Extensions" framework · Anthropic blocks Mythos from EU CAISI equivalent · Karpathy autoresearch (630 LOC) overnight ML experiments · Mythos 94.6% GPQA Diamond · Q1 2026 venture data ($300B, 80% AI) · Mollick's May 2026 model picks · Air Street State of AI: May 2026 |
| 2026-05-10 | Anthropic targeting $900B / $50B raise · GPT-5.5 Instant becomes ChatGPT default (52.5% fewer hallucinations) · GPT-Realtime-2 / Translate / Whisper voice APIs · Sierra $950M at $15B · Moonshot AI $2B at $20B (Kimi K2.6 #2 on OpenRouter) · IBM Sovereign Core · Grok 4.3 on OCI · On-Policy Distillation sweep (Lightning OPD, SDPO) · Mem0 + EverMemOS memory architectures · Air Street State of AI: May 2026 · FDE role playbook · Outcome pricing as 2026 default |
| 2026-05-09 | Anthropic rents all of Colossus 1 from xAI/SpaceX (220K GPUs) · 80× quarterly growth · OpenAI GPT-5.5-Cyber for vetted defenders · Pentagon picks 8 AI vendors (Anthropic excluded) · Code w/ Claude conf — Managed Agents "Dreaming" + finance templates · Sierra $15.8B Series E · Moonshot AI $20B valuation · Cloudflare cuts 1,100 jobs at record revenue · EU AI Act delayed to 2027/2028 · Karpathy: vibe coding → agentic engineering · Single-agent beats multi-agent under matched compute (Stanford) |
| 2026-05-08 | Anthropic-Google $200B compute deal (1M TPUs) · Anthropic Wall Street + Jamie Dimon · DeepSeek V4 (MIT license, runs on Huawei Ascend) · Pit $16M a16z launch · Parallel Web Systems $2B · GPT-5.5 Codex browser agent · Mem0 graph memory · 2026 resume formula · Wedge of 2026 = "AI product team as a service" |
| 2026-05-07 | Apple iOS 27 multi-AI Extensions framework (Gemini + Claude + GPT picker) · Anthropic $1.5B PE deployment JV (Blackstone, Goldman, Apollo, H&F, General Atlantic) · OpenAI mirror enterprise JV · Apple $250M AI marketing settlement · CAISI pre-deployment review signs MS/Google/xAI · Gemini 3.1 Flash-Lite GA at $0.25/M input (1432 Arena Elo) · Anthropic "Dreaming" agent technique · Single-agent vs multi-agent (Stanford recap) |
| 2026-05-06 | Anthropic Claude Mythos (cybersecurity model restricted at launch) · OpenAI $25B ARR + IPO · Google "Remy" personal agent · Cursor 3.0 Agents Window · Vibe coding security risks · CS job market reality check · Vertical agent startup formula |
See SOURCES.md — organized by type with notes on reliability and signal quality.
| Time budget | What to read |
|---|---|
| 60 seconds | Today's 00-tldr.md |
| 5 minutes | 00-tldr.md + bold headlines of one category file |
| 20 minutes (recommended) | One full category file, deep — rotate through the 5 across the week |
| Weekend | Pick one item from 03-practical-skills-and-tools.md and actually do it. Re-read WATCHLIST.md and update personal threads. |
Every entry tags with #topic. Search the archive across days:
grep -rn "#anthropic" . # all Anthropic stories
grep -rn "#voice" . # voice-AI thread across days
grep -rn "#funding" . # every funding round notedCommon tags: #labs #anthropic #openai #google #apple #nvidia #funding #vc #seed #agents #voice #multimodal #open-source #research #policy #eu #jobs #fde #startups #playbook #pricing #benchmarks #memory