Hot takes.

Abstract phrases like 'leverage synergies' are 8x less memorable than concrete ones, so marketers should replace jargon with specific, tangible language

Algorithms are designed to keep users on-platform, so viral posts rarely convert to off-platform actions like podcast listens — borrowed audiences from podcasts and conferences convert far better

The lawsuit will make a great Aaron Sorkin movie, with Dane DeHaan playing Musk and Jesse Eisenberg going two-for-two as Sam Altman.

Getting paid dividends for your contribution to AI training data makes sense, but would require better mechanistic interpretability to quantify individual contributions and blockchain to manage ownership.

The relational sector will likely mirror Spotify's distribution — a few brilliant teachers and craft brewers make fortunes while everyone else competes on platforms that take their cut, with the 100,000th-ranked artist earning just $7,300.

Orchestrating multiple agents introduces massive hidden costs and error amplification; single-agent systems should be the default and multi-agent complexity only added when workload characteristics strictly demand it.

The only chart in crypto that is consistently up and to the right is the number of crypto hacks per month.

AI will bulk-solve entire disciplines — math is largely done, physics is next, and by 2032-2035 we will be building Dyson swarms and putting GPUs in sun-synchronous orbit around Earth.

Flipped bearish on the CLARITY Act, saying there is 'very, very, very little margin of error' remaining.

Consensus expects BTC to bottom on the 4-year cycle in Oct/Nov 2026, but when everyone is positioned the same way, the market usually pulls the rug in the opposite direction.

Corn, wheat, and sugar are coiled at multi-year resistance and are next in the macro rotation after metals/mining and energy, with farmer bankruptcies up 46% on fertilizer prices alone adding to supply chain stress.

The 'rip-your-face-off rally' is over — breadth is garbage, semis are vertical, only 5 stocks are holding it up, and a 10-15% correction is coming between May and midterms.

Atlassian and Twilio's accelerating earnings suggest the SaaSpocalypse narrative may be shifting, with the B2B turnaround potentially underway.

AI agents for SMBs are the hardest problem in the space, but Owner has cracked the code with $1B+ in restaurant sales and is one of the fastest-growing investments Jason has ever made.

Traditional seat-based SaaS models are under permanent threat as AI agents don't need seats, causing tools like Notion to be churned silently while API-heavy platforms like Salesforce cost more.

Agency—the drive to take initiative and ship—matters more than any specific skill set as AI tools commoditize execution.

Vibe coding has caused an explosion in the volume of software being created, but has not improved software quality, and that gap represents a significant opportunity.

The predicted collapse of SaaS due to AI (SaaSpocalypse) is overstated and the threat to existing SaaS businesses is being exaggerated.

Despite the US eliminating the penny in 2025, retailers are unlikely to abandon charm pricing because the perceived costs of giving it up are too great.

World Markets on MegaETH represents the long-awaited gold standard of going bankless — a feature-complete exchange with an entirely onchain codebase and no servers.

Early DeFi apps like Uniswap, Aave, and MakerDAO were constrained by slow Ethereum blocktimes into fragmented, incomplete financial primitives — dubbed 'Slow DeFi'.

Zyfai's report that its agents successfully avoided KelpDAO losses should be taken with a grain of salt as it is the company's own account.

x402 reduces credential attack surface by eliminating stored API keys, but does not fully solve endpoint trust, spending limits, or data integrity — those require additional tooling.

Agents' most plausible near-term value in DeFi is defensive monitoring and capital protection, not yield optimization or novel strategies.

April set a grim new benchmark with 28 exploits and $635M stolen — the worst month DefiLlama has ever logged, approximately four times the entire Q1 total.

Atlassian and Twilio's accelerating growth challenges the prevailing narrative of SaaS decline, suggesting the SaaSpocalypse may be overstated for companies successfully integrating AI.

Many B2B software companies still have moats, but moats alone are no longer sufficient for long-term survival as AI agents can replicate features rapidly and customer expectations are evolving faster than incumbents can adapt.

The massive upside potential for prediction market platforms will require winning legal battles ahead while also overcoming negative public sentiment around insider trading and market integrity.

Kalshi's prior 2023 concession that football games qualify as gaming undermines its current legal argument that sports contracts are financial swaps, creating a major contradiction states are exploiting in court.

Amodei's 20% unemployment prediction is unfounded because even Anthropic's own economists concede that the impact of major economic disruptions on labor markets is unclear even in hindsight.

The AI infrastructure buildout — with $1.5T in cloud order backlogs and up to $8T in projected datacenter spending — makes the dotcom boom look small, suggesting the current market concentration may be justified.

AI CEOs like Dario Amodei warning of mass unemployment from their own products is terrible PR that is creating an existential political risk for the industry, potentially enabling federal data center moratoriums.

The GENIUS Act created a stable enough regulatory floor that large consumer platforms are now willing to build product roadmaps around stablecoins, as evidenced by Meta, PayPal, and DoorDash all moving within the same two-week window.

PYUSD should be one of the largest stablecoins given PayPal's 439M customers and $1.8T annual volume, but it isn't — primarily because crypto was a siloed product team with no ability to influence pricing or integration decisions across the merchant stack.

Meta's new USDC payout playbook — pure distribution, no proprietary infrastructure — is structurally superior to Libra, and the author expects expansion to more high-remittance markets in Southeast Asia and Latin America following Stripe and Deel's playbook.

While voting rights and transfer agent infrastructure gaps are closing, secondary market liquidity remains thin and T+0 settlement interoperability with legacy clearinghouses is unresolved — the rights gap is closing but the plumbing gap is not.

The ways we prompt AI are very different in 2026 than 2022, with models now capable of thinking for minutes, ingesting many documents, and using tools — but most users still only ask short questions.

GPT-5.5 knows more than its peers but answers incorrectly more often and acknowledges ignorance less often, making it less trustworthy despite leading objective benchmarks.

More S&P 500 stocks are getting earnings revised down than up — six companies account for ~70% of earnings upgrades, making index-level calm deceptive.

Despite BTC being in a downtrend with market positioned for new lows, bullish macro tailwinds are stacking and a short squeeze is on the table, especially if the CLARITY Act lands as a catalyst.

Kevin Warsh's belief that AI is deflationary gives him political cover to cut rates even with inflation at a two-year high, which is net positive for risk assets.

The $300M DeFi United bailout raises questions about whether crypto bailouts are good for crypto and what DeFi must fix before it can scale to the mainstream.

April was a brutal month for DeFi, marked by historic exploits and a continuing downward trend in TVL and trading volumes since October 2025.

Musk ranked Anthropic first among leading AI providers, followed by OpenAI, Google, and Chinese open source models — placing his own xAI behind all of them.

AEO requires cross-functional coordination touching brand, content, PR, partnerships, and product — no single function can own or execute it, unlike traditional SEO which lived in one team

AI search is a genuine blue ocean where startups can win because the inventory of answer slots is functionally unlimited, unlike zero-sum traditional SEO where three spots define all the value

Fully AI-generated content does not perform well in AI search — 82% of articles cited by AI models are human-written, so scaling AEO with AI content farms won't work

Video budgets are cooling, with the share of teams planning spend increases dropping significantly from 57% in 2024 to 49% in 2026.

Subscriber count is now a vanity metric for YouTube, implying other engagement signals matter more for video strategy.

Every layer of the AI stack is collapsing the gap between thinking and doing — speed without sloppiness is the new bar, as evidenced by Claude Security shipping to prod, KAME talking before thinking, and Cursor opening its agent SDK.

Improvements in spatial biology tasks are unlikely to come from general reasoning gains alone and will require explicit training on statistical design, platform-specific analysis, and replicate-aware differential testing.

The Cursor-xAI deal is considered good for everyone: Cursor gets compute and a non-competing model lab, while xAI gets an application surface to show public market investors before the SpaceX IPO.

In an era where both legitimate authors and malicious attackers use the same coding agents to generate code, entropy-based techniques for detecting malicious browser extension code may become less relevant.

Detection-as-Code pipelines may be overrated due to complex infrastructure requirements, and LLM agents could automate much of the process from linting to deployment at the cost of strict determinism.

Autonomous business workflows are forcing companies to rethink work entirely, not just insert AI — leaders must identify specialized human value before agents take over operational tasks.

Everyone should be actively managing token usage now because models will get more expensive as they improve, making cost discipline a competitive necessity.

Most VCs claiming to be conviction investors actually back pattern-matched founders (elite school, prior exit, hot market) rather than taking genuine risk on strange founders in unproven markets.

Stablecoins offer fast, programmable payments but fail to address complex organizational and governance structures of global card networks, requiring crypto advocates to distinguish between the efficient authorization layer and the inefficient legacy settlement layer.

Prediction markets like Polymarket and Kalshi currently offer limited public value compared to expectations, with the primary bottleneck being the difficulty of writing high-quality, actionable questions.

Design systems built for human-scale processes are becoming 'Inference Systems' as AI agents now assemble interfaces dynamically, requiring adaptation metrics rather than simple adoption metrics.

AI chat platforms like ChatGPT, Claude, and Gemini suffer from a 'forgotten conversation' problem because messaging-style interfaces don't properly index content, making past insights hard to retrieve.

AI tools are eliminating the informal colleague interactions that build trust and innovation, with disconnected employees becoming less innovative, more prone to burnout, and more likely to leave.

AI tools lack the critical self-evaluation needed to produce quality design — professionals must apply a rigorous editorial eye before any AI output reaches an audience.

China's open-weight AI approach is making the future of AI more multipolar than Silicon Valley expected, with open-source models shifting from hype to deployment advantage.

AI risks producing convincing but unowned post-mortem conclusions if it replaces human analysis; real value only comes from human synthesis and accountability, not AI-generated summaries alone.

AI reverses the natural degradation of engineering teams by enforcing code hygiene and maintaining institutional knowledge, enabling faster and higher-quality output.

Despite disappointing sales, Apple will not abandon the Vision Pro platform.

Technical interviews overemphasize theoretical algorithmic puzzles that rarely appear in production, where standard libraries and design trade-offs matter far more.

Advanced AI models like Claude Opus 4.7 can identify authors from tiny text samples, making traditional online anonymity protections obsolete for frequent writers.

Product marketing works best when it builds systems that let anyone launch, not when PMM controls every release — centralized control creates bottlenecks that slow teams down.

Site refreshes frequently drop conversion rates not because new designs are bad, but because aesthetic improvements inadvertently weaken trust signals, CTA visibility, and mobile scanning experience.

AI is producing more marketing output but not improving results — 40% of marketers report more work despite AI, and 64% say personalization lacks real impact.

The faster models get, the more value shifts to taste in deciding what NOT to build; personalization layers will look more like Pi than today's all-in-one platforms within two years.

Big AI labs that built their empires using distillation are now using lawyers and policy to prevent competitors from doing the same, pulling up the ladder behind them.

The West needs a strong open-source AI stack to beat China, and edge models should be open-source because once on a device they're already exposed.

After interviewing 30+ engineering teams, code quality has dropped across the industry with serious projects shipping what he calls 'vibe slop.'

Agent armies create complexity their own future selves can't untangle; slowing down and building minimal toolsets is preferable to scaling agent swarms.

Amazon's AI-hosted product review podcast is potentially 'one of the funniest, closest endpoints to human civilization,' with bots that often can't answer basic product questions.

Building AI plumbing directly into the browser causes severe negative consequences to interoperability, updateability, and neutrality of the web platform.

WordPress is killing itself by blindly following rules and ideals and may be in irreversible decline.

AI will displace millions of jobs and create a permanent underclass as power and wealth concentrate among AI companies and capital owners, while ordinary people lose economic leverage.

Talking about AI adoption or transformation is insufficient; leaders win by running consistent, high-signal operational habits, not just strategy.

Strategy doesn't die in the roadmap — it dies at the intake form, where lack of ownership causes teams to build low-value features instead of executing strategy.

AI making it easy to build features is a trap — the real PM challenge is now deciding what NOT to build to avoid product bloat.

Fixing your process framework is a time commitment, but you are wasting far more time if you don't do it — the cost of inaction is either a broken campaign or burnout.

LLMs do not have a theory of mind and likely face a persistent upper bound on interactive reasoning, meaning they may never replicate the human ability to deduce what other traders know from observing their behavior.

At $2B market cap — last seen in summer 2021 — it takes far less money to move NFT prices today than at peak ($16B in April 2022), so current price surges are less meaningful than they appear.

The current NFT rally likely lacks a singular catalyst and may be temporary hopium in the absence of higher-beta ETH DeFi plays, not the beginning of the next major NFT bull market.

The bull market was a disappointment — Bitcoin made new ATHs but grinded there with no euphoric blow-off top, no magazine covers, and no parabolic candle, while TOTAL3 collapsed ~50% after a single weekly close above its prior ATH.

Widespread bearish consensus and heavy short positioning mirrors the over-leveraged bull setup before the October 10, 2025 crash, suggesting an early reversal is possible before the expected Oct/Nov bottom.

AI is quietly collapsing into one unified system as tools merge and costs scale, with Claude's multi-app connectors and Copilot's token billing signaling a shift toward unified agentic compute.

Different AI models trained on different data are converging on the same internal map of reality — we're not building a zoo of AI tools, we're building one mind.

Brands that engineer content to be clippable — like Skittles and Ramp — can generate organic reach similar to how politics and sports content spreads virally online.

When people say they're 'too busy' to adopt AI, it's almost always procrastination — they're waiting for an external condition to change rather than taking the smallest possible first action

High energy prices from the Hormuz closure mean sticky inflation, making the Fed less likely to cut interest rates

Investors have shifted focus from AI apps to AI infrastructure — chips, data centers, and energy — and those on the winning side of scarcity are doing very well

The U.S. appears happy to keep Hormuz closed longer as a negotiation tactic, but if it is one, it's an expensive one

Search professionals will fork into two paths: those who move upstream into strategic executive conversations, and those pushed downstream into increasingly automated production and reporting work.

Self-promoting listicles have already hit a wall as a strategy for getting brands mentioned or recommended by LLMs.

Calling everything 'AI search' is lazy — it fragments into completely different problems depending on business type, and treating it as one thing leads to wrong strategies.