Feed
The latest from the newsletter firehose. Every fact, every signal, as it lands.
Monday, May 4 — Yesterday
Phill Agnew (host of Nudge) found that appearing on other podcasts drives far more listeners than viral TikTok videos because borrowed audiences already trust the format
Psychologist Ian Begg's research shows concrete phrases (e.g., 'fast car') are 8x more memorable than abstract ones (e.g., 'cutting-edge technology'), a principle marketers should apply to all copy
Showing social proof visually — like a coffee shop displaying loyalty cards on the wall — is more persuasive than simply claiming popularity in marketing copy
Sunday, May 3
Starbucks reversed its automation push and is hiring more baristas, with economist Alex Imas arguing that human involvement becomes the scarce premium good as AI commoditizes everything else.
Anthropic crossed $1 trillion valuation surpassing OpenAI for the first time, while OpenAI restricted GPT-5.5-Cyber after AISI flagged record offensive cyber capabilities.
A three-folder CLAUDE.md pattern (Root, Workstation, Project) is the recommended scalable approach for building personal agents in Claude Cowork without requiring terminal skills.
Stanford and Google/MIT research shows multi-agent systems can amplify baseline errors by up to 17.2x and suffer 2x–6x efficiency penalties on tool-heavy tasks when compared to optimized single agents.
Single-agent systems should be the default architecture; when they fail, fix reasoning depth with pre-answer scaffolding prompts before escalating to multi-agent orchestration.
Use multi-agent systems only when context windows break under noisy RAG data, tasks decompose into fully independent sub-tasks, or strict regulatory verification (healthcare/finance) is required.
Lance Roberts (RIA Investors) flipped bearish on equities, predicting a 10-15% correction between May and midterms and raising 10-20% cash in portfolios.
Franklin Templeton now officially recommends 3-6% portfolio allocation to digital assets, with institutional clients having 'grown-up' conversations about crypto portfolio construction across beta, venture, and yield strategies.
Dr. Alex Wissner-Gross of Physical Superintelligence predicts AI will bulk-solve entire disciplines — math is largely done, physics is next, and Dyson swarms by 2032-2035.
Salesforce costs rose 83% YoY to $22K despite cutting human seats from 10+ to 2, because AI agents consume more API usage — signaling a fundamental shift in SaaS pricing.
Atlassian and Twilio reported accelerating growth with $7B and $5.6B run rates respectively, challenging the 'SaaSpocalypse' narrative for B2B software.
The simplest path to making your product agentic is ensuring your API is truly agent-friendly — most vendors are overlooking this critical step.
Max Schoening (Head of Product, Notion) argues that cultivating agency—not skills—is what separates thriving professionals from those who fall behind in the AI era.
Vibe coding has caused an explosion in the quantity of software but not quality, and Schoening says this gap is a major opportunity for builders who can ship both fast and well.
Schoening's 'tiny core' theory holds that great products are built around a single defining mechanic—citing iPhone multitouch, the GitHub pull request, Notion blocks, and Dropbox's menu bar icon as examples.
Charm pricing (ending prices in .99) originated in 1879 as an anti-theft mechanism in cash registers, forcing clerks to open the register for every transaction.
The left-digit effect — studied by Cornell's Manoj Thomas — causes consumers to anchor on the first digit, making $2.99 feel closer to $2.00 than $3.00, a bias that persists even when you're aware of it.
The US stopped producing pennies in 2025, sparking proposed legislation to round cash prices at the register, but retailers show little appetite to abandon charm pricing.
Saturday, May 2
World Markets on MegaETH is the first fully onchain exchange with no backend servers, combining spot, margin, perps, and lending under a single unified margin account.
Bitmine acquired over 101k ETH last week plus 10k ETH OTC from the Ethereum Foundation, while Ethereum hit a new ATH of 3.6M daily transactions.
1inch surpassed $800B routed onchain and Circle launched a USDC nanopayments gateway across 11 EVM chains.
The Pentagon signed AI access deals with SpaceX, OpenAI, Google, NVIDIA, Microsoft, and AWS for classified networks, while banning Anthropic as a supply-chain risk over military guardrail disputes.
Meta acquired robotics startup Assured Robot Intelligence to supply the AI intelligence layer for humanoid robots, competing with Google DeepMind's Gemini Robotics strategy.
xAI launched Grok 4.3 with a 1M-token context window and voice cloning from 120-second clips at $1.25/$2.50 per million tokens — 40-60% cheaper than Grok 4.2.
April 2026 was crypto's worst month for hacks with ~30 exploits and $625M stolen, with Drift and KelpDAO as the largest victims.
x402 eliminates stored API keys for agents by enabling per-call micropayments to data endpoints, shrinking the credential-exposure attack surface.
A three-layer defense stack — Zauth for endpoint trust, Ampersend for spending controls, and Vaults.fyi for vault risk — enables safer autonomous DeFi agent deployments.
MegaETH's MEGA token launched at a $2B valuation and doubled the network's DeFi TVL, though the token price fell from $0.22 to ~$0.15 shortly after launch.
April 2026 set a record with 28 crypto exploits and $635M stolen — roughly 4x the total stolen in all of Q1 — according to DefiLlama.
Meta re-entered crypto with USDC creator payouts in Colombia and the Philippines on Solana and Polygon, four years after shelving Libra.
Atlassian and Twilio hit $7B and $5.6B run rates respectively, accelerating growth and challenging the SaaSpocalypse narrative.
SaaStr is paying Salesforce 83% more year-over-year due to AI agent seats while having dropped Notion entirely, illustrating the uneven impact of AI on SaaS spend.
Thoma Bravo wrote down $5.1B on Medallia, signaling that legacy B2B software without strong AI adaptation faces severe valuation risk.
Friday, May 1
Kalshi and Polymarket face lawsuits from 40+ state AGs over unlicensed gambling claims, with states citing Kalshi's own 2023 admission that football games qualify as gaming.
Polymarket signed an exclusive MLB deal worth up to $300M over 3 years, and sports betting now accounts for 86% of Kalshi's $23B annual volume and $10B of Polymarket's Q1 2026 volume.
Paradigm proposed 'PACTs' to protect Bitcoin from quantum computing attacks, involving burning funds from quantum-vulnerable addresses that fail to migrate.
Alphabet added $450B in market cap on quarterly earnings of $110B revenue, and the three cloud giants now hold a combined $1.5T order backlog — making the AI boom larger than the entire dotcom telecom buildout.
Dylan Patel of SemiAnalysis predicts large-scale anti-AI protests within three months, driven partly by AI CEOs like Dario Amodei warning of 20% unemployment from the very products they're building.
A Stanford study puts AI consumer surplus at $172B — exceeding all AI revenues collected — while IEA data shows running an LLM uses less power than charging your phone.
Securitize x Computershare and Ondo x Broadridge partnerships close the two biggest gaps in tokenized equities — transfer agent infrastructure and on-chain shareholder voting rights — covering ~58% of the S&P 500.
Meta re-entered stablecoins by paying creators in USDC on Solana and Polygon in Colombia and the Philippines, while PayPal restructured to make PYUSD and crypto a core merchant-facing business unit alongside Braintree.
Meta, PayPal, and DoorDash all made stablecoin moves within the same two-week window, all using existing regulated rails rather than proprietary infrastructure — enabled by the GENIUS Act regulatory floor.
GPT-5.5 tops the Artificial Analysis Intelligence Index (60 pts) and ARC-AGI-2 (85%) but ranks third on knowledge calibration behind Gemini 3.1 Pro Preview and Claude Opus 4.7 due to an 85.53% hallucination rate.
GPT-5.5 API is priced at $5/$30 per million input/output tokens — roughly double GPT-5.4 rates — while GPT-5.5 Pro runs $30/$180 per million tokens with parallel reasoning inference.
Andrew Ng launched AI Prompting for Everyone, a no-technical-background course covering deep research mode, multi-document context, and agentic AI use across ChatGPT, Claude, and Gemini.
The Strait of Hormuz closure has created a 6–11.4M barrel/day oil shortfall, pushing Brent crude to $105–110 and the global oil bill toward $5–8T annually.
New Fed Chair Kevin Warsh views AI as deflationary, giving him cover to cut rates despite 3.3% U.S. inflation, which combined with ISM PMI above 50 historically sets up Bitcoin and crypto to outperform.
The SOX semiconductor index rose 18 days straight to hit 43% above its 200-day moving average — the widest spread since June 2000 — as compute scarcity drives a narrow group of winners dominating S&P 500 earnings upgrades.
Bitcoin gained 14% in April and Bitcoin ETFs pulled in nearly $2B in net inflows, while Morgan Stanley's MSBT ETF launched with the lowest fees and $163.86M inflows in its first month.
DeFi suffered $635M in April exploits led by the $285M Drift Protocol drain and $293M Kelp DAO bridge hack, causing Aave's TVL to collapse ~40% to ~$15B.
MegaETH's MEGA token launched at a $2B FDV across 13 CEXs, putting all community raise participants in profit.
Apple hit a record $111.2B quarter driven by $57B in iPhone sales but warned 'RAMageddon' — AI-driven memory chip demand — will significantly raise costs starting June.
Tesla began mass production of its Semi electric truck after 9 years, offering a 500-mile Long Range version at ~$300K, undercutting all major electric trucking competitors on both price and range.
Meta fired 1,108 Kenyan AI trainers from contractor Sama after employees exposed that Ray-Ban footage used for AI training included banking info, private conversations, and intimate encounters.
AI search prompts average 23 words vs. 3.4 for Google, creating unlimited answer inventory that startups can win by matching specific user context rather than competing for top-3 rankings.
85% of brand mentions in AI-generated answers come from third-party sources like listicles, Reddit, and editorial coverage — making off-site presence more important than your own website.
Structured content with comparison tables (2.5x citation lift) and statistics (41% visibility boost) dramatically outperforms prose, while fully AI-generated content fails — 82% of AI-cited articles are human-written.
LinkedIn has overtaken YouTube as the #1 platform for B2B video sharing, according to Wistia's 2026 State of Video report.
30–40 minute webinars generate ~70% higher engagement than shorter webinars under 30 minutes, per Wistia's analysis of 79 million hours of viewing data.
Video budget growth is cooling — only 49% of teams plan to increase spend in 2026, down from 57% in 2024, while 48% plan to increase distribution and promotion budgets.
Claude Security launched in public beta, scanning codebases for complex vulnerabilities like auth bypasses and memory corruption without any API integration, with findings pushed to Slack or Jira via webhooks.
Cursor released an open-source SDK (`npm install @cursor/sdk`) that lets developers embed its full coding agent engine into CI/CD pipelines, custom products, or cloud-sandboxed background agents.
Nous Research's Hermes Agent now natively controls ComfyUI end-to-end, automatically installing it, managing models, and running full media generation workflows via REST and WebSocket.
xAI acquired Cursor for $60B to gain a consumer application surface before the SpaceX IPO, while Grok 4.3 launched with higher intelligence scores at lower cost than its predecessor.
Anthropic nears a $900B valuation on a ~$50B funding round with revenue approaching a $40B run rate, and Claude Security entered public beta using Opus 4.7 to find and patch software vulnerabilities.
KV cache locality is identified as a major hidden variable in LLM serving cost, where load balancers unaware of token locality cause avoidable GPU recomputation that degrades throughput and latency.
CVE-2026-31431 is a Linux kernel logic bug exploitable with a 732-byte Python script to gain root on Ubuntu, Amazon Linux, RHEL, and SUSE via controlled writes into setuid binary page cache.
GitGuardian found 28,000 LLM-generated passwords in 34 million GitHub commits, with Llama-3.3-70b-instruct repeating the same substring in 96% of outputs and weak credentials appearing in 1,800 .env files.
A DPRK-attributed supply chain attack disguised as a 0G Labs Web3 job interview used an npm prepare hook RCE to beacon victim environment variables and MAC addresses every 5 seconds to a Texas-based IP.
Google Cloud hit $20B quarterly revenue (up 63%) as Alphabet poured $35.7B into capex to expand AI infrastructure.
Railway rolled out a 48-hour soft delete policy after an AI agent destroyed a production database using a long-lived token, signaling a new class of agent-caused incidents.
Anthropic launched Claude Security in public beta for Enterprise customers, offering automated code vulnerability detection and fix generation.
X has completely rebuilt its advertising platform with AI-powered ranking and real-time ad delivery, marking the most ambitious ad platform overhaul in its 20-year history.
Adding an 'Ask AI' button (linking to ChatGPT, Claude, or Perplexity with a balanced pre-filled prompt) to landing pages converts better because the explanation feels like a second opinion rather than marketing.
Businesses should right-size AI models, avoid unnecessary reasoning steps, and stay flexible with providers as token costs are expected to rise alongside model capability improvements.