Hot takes.

Chop is the signal to buy, not tune out.

Lighter's token faces $50M monthly unlocks starting January, so he won't touch it despite liking the product.

CoreWeave's bond yield is not the right metric; average borrowing cost has fallen to 8.3%.

Gas tokens powering blockchains will become strategic commodities as more assets move onchain.

Bitcoin will follow silver's script to $400K.

The B2B apocalypse never ended for anyone under $1B in ARR.

All three ledger failures (bOSMO, NEAR Intents, Pirate Chain) came from claims exceeding assets; protocols should add a check that automatically pauses when the ledger diverges from holdings.

AI companies are wildly successful despite confusing pricing, but this will not last with the mass market; early adopters tolerate mystery credits, mainstream users will not.

BTC could test $78-80K before November, and SOL rarely gets a free pass when Bitcoin dips.

Solana is a high-speed trading engine whose real rivals are Hyperliquid and Lighter; NEAR is better for AI agents, but $100 is a great place to add.

Solana will win the AI agent economy, with machines paying each other in stablecoins for data, compute, and APIs.

Technical accomplishments gain business credibility only when their impact is clear, like reduced disruption and lower recovery costs.

AI safety warnings cannot be dismissed as corporate marketing or attempts to block competition.

Age-based social media bans are easily circumvented and not the right approach; we need better strategies to keep kids safe online.

Accenture tells clients to invest heavily in AI but gives employees only $29/month for Claude.

Leaders need to stay hands-on and current.

The startup-or-big-company choice is increasingly a false dichotomy.

Leadership skills are trending up relative to being a builder.

You no longer need to work at OpenAI or Anthropic to work on the frontier.

The next unlock in AI isn't smarter models but smarter orchestration — running more parallel sessions, resetting often, and staying in control.

Micron looks cheap at ~6x forward earnings; long-term deals cover 35%+ revenue through 2030; demand outrunning supply through 2028.

Once China learns to make something, it can often outproduce the rest of the world; RAM was a low-margin commodity 3 years ago; HBM will be harder to crack.

Market got way ahead of itself; CXMT still has to prove yields, pricing, and volumes; HBM not affected; Sandisk NAND irrelevant.

There is no universal content playbook; advice is situational and you can only play the hand you're dealt, so you have to figure out what actually works for you on the job.

Getting excited when prices go down is the proper strategy.

It's still too early to go hunting in alts.

Expects BTC to dip to ~$74K, causing alts to puke, which would be a buying opportunity.

Alts won't collapse like they used to because of new alt ETFs and long-term holders.

Open-weight models are quietly eating closed-model lunch, one vertical at a time.

Never hire a VP of Sales who hasn't hired and managed a small team that hit quota; jumping from top rep to VP rarely works.

AI agents picking vendors is the way to get leads; if an AI agent recommends your product, you get tested first.

These are fuel, not compounding advantages—they won't keep competitors away long-term and a startup must eventually build something more durable.

Self-custody is now narrow: buy direct, generate visible entropy, and split keys across vendors via multisig, but even then the only partial recovery came from Tether's blacklist, not self-custody itself.

Oil chokepoint closure leaves Fed stuck with no clean exit, feeding stagflation risks.

Stock rally is running on just two letters: AI, and breadth is historically bad.

Crypto is in the early innings of a new bull market and shrugging off macro headwinds; once Fed signals done, Bitcoin might already be over $100K.

The equipment sellers are agile, while the plant owners are forced to move at the speed of government.

Visible chain-of-thought is often a workaround for latency, not a valuable feature.

As AI tools make content production easier, human-made work will become more valuable and command a premium.

Direct cache communication may complement text rather than replace it.

Better orchestration will increasingly depend on avoiding unnecessary computation and communication between model calls.

Why own stocks when 10-year Treasuries pay 5.3% risk-free while the S&P at 19x earnings yields roughly 5%?

We are in a lockout rally and won't get good altcoin entries again; he's fully deployed at 0% cash.

SpaceX is an AI compute company, not just a rocket launch company, and his price target is $389.

Coinbase and Robinhood aren't really onchain; high-frequency trading needs 1-2M TPS, which is what Zero chain targets.

BTC is in a new bull market if it holds the 200-day MA around $71K; he'd risk 15% on the chin for 130% upside.

From where I'm standing, this is a fire drill — nobody has shown an actual attack, and the fix for regular holders is cheap: stop reusing addresses and keep long-term coins somewhere you've never sent from.

AI's ability to refuse dangerous requests is overtrusted; failed refusals may lead to global calamity, and relying on refusal lets governments restrict free speech.

OpenAI's unauthorized access issues are a failure of management and engineering, not alignment, and raise legal liability questions.

Building an enterprise infra business from zero is easier said than done, as Google Cloud's struggles show.

Dots launch exposes the same blank-screen and jagged-model problems; agents may need to be horizontal/ambient or vertical/embedded to work.

There's no sign of winner-takes-all effects in frontier labs, and pricing power is limited beyond compute shortages.

Focusing on Anthropic's pre-agentic-coding numbers is deeply unserious; the real questions are inference profitability and future model costs.

When people say they hate AI, they actually dislike tech companies' relentless push; AI's future is not as inevitable as companies imply.

AI technology has become warped by hype, steered by zealots, and inescapable.

Good content can't be boiled down to rules; content that only follows rules is 'correct' but not 'good' — it also needs rule-breaking and an anomalous element.

Micron will likely beat but the stock direction after earnings is a coin toss; guidance and comments on supply and long-term deals are more important than the quarterly numbers.

Small calibrated models beat prompted frontier LLMs on bounded, high-volume security classification.

Ethereum's setup for $10K is finally starting to make sense, potentially triggering the next altseason.

Bullish long-term on SOL but in no rush to buy short-term because it hasn't found a catalyst or narrative to pull energy back in.

Things are going to stay volatile and probably get even more wild.

Bitcoin's price will likely continue higher in Q4 after the second weekly close above $83K.

If inflation starts to blow out, we might be cooked.

It can take 12-18 months to hire a great VP of Sales, and even with a strong process the hit rate is barely .500.

Being a system of record helps retention but alone it won't equal growth; ServiceTitan grew 25%, while Snowflake and Databricks grew faster.

Working a lot is possible when you have a strong why; the answer is less about 'how' and more about 'why'.

Algorithms don't just happen to us; we train them through our behaviors.

The next big agent benchmark may be permissioning; the bigger fight is who becomes the aggregator between you and hot-swappable service providers.

AI companies selling branded merchandise could work as a conversation starter or possibly ender, given public sentiment.

Churn is never conquered.

Growth almost always comes from the core product

Snapchat's ads business never reached its potential

The median PM is bad

Circle owns every layer of Arc except block producers, giving it control over validators, fee asset, FX layer, and token supply.

Onshore tokenized stock trading now requires transfer agent's register, shareholder's vote, and issuer's acquiescence, limiting venues to those with registered transfer agents.

OpenAI's safety team is racing an accelerating capability curve using tools that have already started to fail.

The #1 mistake founders make when hiring a VP of Sales is stepping back from sales; the CEO never gets to step away.

If Slack had stayed independent, it would likely trade at $16B-$20B, far below the $27.7B Salesforce paid.

The pattern is consolidation - one model, one interface, one backbone to rule them all.

The characters in your training data are shaping your AI's identity quietly at scale.

After the vote failed, we're back to waiting for a comprehensive crypto framework.

The slowdown is likely self-inflicted; labs need to clean up faulty products and demand transparency.

Next-generation efficient models might not have fewer parameters; they'll get better at deciding which parameters run, which memories survive, and which parts of a million-token history are worth revisiting.

Parameter count alone says less about production cost; efficiency depends on active params per token, KV cache bytes, persisted state, and retrieval cost.

Markets look like they absolutely want to rip; speculative positioning is flat.

The market being awake is the catalyst for higher prices, not just breaking specific levels.

Rate limits feel too quick because GPT-6 should be used as an orchestrator that delegates to sub-agents.

AI is reaching an inflection point and becoming easier to use; users should try the highest-level plans.

The 21-bank stablecoin effort is FOMO-led 'paying to see'; with no CEO or ownership details, it won't overtake USDC/USDT anytime soon, especially as six banks also back a competing tokenized-deposit network.

Robinhood's legal position is solid because its disclosures state the tokens grant no rights, but its commercial position is exposed since AMC publicly disavows the product.

The practical utility of very long-horizon tasks is amplified beyond reality; most effective coding agent use is a complex, highly iterative process with high-skill human intervention.

Cutting-edge AI technology is less regulated than the average food truck, and that must change.

Using the front line in Ukraine as a model training site requires urgent regulation; battlefield data should not be treated like ordinary commercial material.

The squeeze may be dead, but TEQs are just getting started and represent a new design space for onchain equities.

Cornering tokenized stock supply doesn't force equivalent purchases of underlying stock, so the squeeze is a lie.

Volume is coming for Aerodrome as Coinbase expands tokenized stocks, likely driving gains; already started buying in preparation.

beehiiv is one of the best owned audience platforms for in-house B2B marketing teams in 2026.

Most brands claiming to build community are just collecting comment sections of randos talking to the brand, not real communities.

Likes and comments are dead; brands should stop optimizing for them and focus on watch time and completion rates.

2024 was the discipline year for hiring, but hypergrowth companies have returned to aggressive headcount expansion, with 100%+ growers hiring harder now than in 2022.