source: Imgflip
Crypto
Three years ago, I had ~[N]% of net worth in crypto, mostly on hardware wallets ("self sovereign"), guided in spirit and discovery by the Bankless bros, David Hoffman and Ryan Sean Adams. Also a bit in esoteric crypto stocks. MicroStrategy was considered risky and untested, actually basically failed as a business intelligence software provider I had first encountered over ten years ago, and Michael Saylor pattern-matched as crackpot-in-the-making a la John McAfee. But by holding GTBC and MSTR in tax-advantaged accounts, that risk carried a heads-I-win-tails-I-win flavor of asymmetric upside.
After ZIRP ended, crypto became [X x N]% of my net worth, while my real estate holdings dropped in value.
I learned from the experience that self-custody carries three insurmountable drawbacks: safety, convenience, and tax/legal risk. My very simple Ledger devices had stupid hardware issues, like un-replaceable batteries losing the ability to charge. The company got hacked, exposing me to phishing attempts. And they took funding, which required them to pivot in order to monetize more aggressively.
While carrying a lot of value onchain, I always worried about getting hit by a bus, or catching a virus that might scramble my brain (cough Long COVID, not syphilis!). Hardware + software wallets are unfamiliar and complicated, with no legal guardrails. My family would have trouble understanding anything without a trusted and well-informed guide. So with new ETFs like IBIT coming to market, I moved almost everything back into custodial brokerage accounts, so it would be easier for any estate planner or probate attorney to untangle without my specific knowledge. I unwound most of my onchain DeFi. And in stocks, as Strategy (Saylor's Microstrategy) got riskier, I took my position down to <2% of net worth.
Bitcoin remains king. I lost faith in Ethereum as an investment, although it remains an important technology enabling important and powerful use cases. My positions are now a barbell, with bitcoin at one end, and things riskier than ETH at the other: alt L1s as flyers, and memecoin runners. The entire ETH ecosystem of L1 + L2s + projects is heavily influenced by VC incentives. Hence it exists, at a first approximation, as a vehicle to transfer value-for-risk from later entrants to the early insider investors.
I still appreciate the concept of self custody, especially with social recovery. I like account abstraction and embedded wallets, for their composabiilty and transparency, providing a kind of security as well as portability. The new Gemini Wallet ("Onchain") and Coinbase's Smart Wallet appear to be gud tech.
I onramp/offramp with Gemini and Coinbase (only if going to USDC on Base). Coinbase is significantly more expensive on swaps and fees (and hidden costs). Once you get through their tollbooth, DeFi options on Base are cheap, fast, and transparently competitive. Gemini is secure, friendly, and offers competitive markets where it chooses to compete. As a New Yorker subject to the BitLicense regime, I don't have access to, nor do I need, regulated offchain opportunities for derivatives-enabled leveraged financial speculation.
Daily driver hot wallets are Rabby (desktop) and Rainbow/Phantom (mobile); maybe Family and Trust Wallet are worthwhile alternatives to consider. I think Privy and Dynamic make solid wallet-as-a-service stacks. Privy feels like it makes more compromises to accomodate its project partners (study how it allowed Farcaster to launch a Privy-powered embedded wallet without private key export functionality). Dynamic feels more expensive, hence mostly suited for corporate-scale use cases. thirdweb Connect is an exciting newer development, which probably gives you all the benefits of composability, arms-length security, and right-to-exit. These opinions reflect my observations as a snapshot in time.
DAT's don't make sense to me. They're a money grab, as the convenience/understandability they promise are way overpriced. They may become interesting as vulture plays in a crash/bear market, but they may also become value traps in that situation. It's not obvious that they should, or will ever again, be priced as richly as they are now.
Some have argued that the ability of a company to stake + loop (leverage) + participate in DeFi justifies higher valuations. But nobody has solved principal-agent misalignment, nor has anyone disproved the efficient markets hypothesis. I keep some funds invested with Saylor, in faith that he will continue to find novel ways to unlock value. But there's no reason to believe a company formed on the virtues of copycat following and just-enough differentiation, will eventually demonstrate alpha for the market to handsomely reward.
Stablecoins, including fedcoins and corpchaincoins, make little sense to me. There already exist a handful of stablecoins that cover the entire spectrum of risk, access, and censorship resistance. Outside of Visa/MasterCard/their direct challengers, the rest of the stablecoin industry is just selling picks-and-shovels to gold rushers. The "gold" in this analogy is the Treasury interest that Tether gets to keep because it's illegal to pay it out. The yield is only coming from money printer go brrr.
AI
My mental model of AI is either: a) an extremely precocious 11 year old child with an undeveloped moral compass, or b) a tireless intelligent and sycophantic junior employee. I'm probably less worried than the average enthusiast about the Singularity; probably more worried than the average enthusiast about value redirection en masse (aka "theft").
Right now, I use AI in three ways:
- $20/mo coding assistant via Cursor, mostly TypeScript and Python but also a tiny bit of Rust and sysadmin / deployment advice.
For me, it's 60% productivity gain + scope expansion: blasting through concepts I would otherwise need 1-2 days to learn and implement, including handling common edge cases: e.g., implementing drag-and-drop in React using a battle-tested library; 35% staying at the right point on the learning curve: for now, this means learning about "vibe coding" and agentic development; and 5% for the dual dopamine hits of a) watching something appear out of a mere glimmer of a thought, and b) the agent gassing me up for making brilliant observations and astoundingly perfect technical choices with the consistency of Shohei Ohtani.
-
informing many software products and apps from which I derive value. Special shout-out to Finch, an iOS habit tracker that uses virtual pets to gamify self-care and productivity.
-
free, privacy-preserving chat
- public AI search via Google, Bing, etc.
- DeepSeek, primarily to translate idiomatic mainland Chinese
- self-hosted LobeChat talking to open-source Cloudflare AI models on Pay-as-you-go including the daily free allowance
I had tried running de minimis-cost local ollama, and serverless GPU clusters (fly.io and Runpod), all with poor results. It was impossible to achieve performance at low cost. You're contending for the same resources that game studios and AI startups rely on. But Cloudflare AI has been acceptable so far, in limited testing.
LobeChat is free, sufficient, and appears powerful if you're willing to leak some data to the hyperscalers. It's also got an over-complicated and confusing interface. But it turns out, if you don't pay for any features, it kind of enforces some level of privacy and safety, by not really working well with history and larger context windows.
Cloud
For 20+ years, I strongly avoided the pay-for-cloud treadmill by operating little private networks and cross-platform sync routines, and self-managing time-consuming redundant backups. But the value of my time has grown (slowly) and costs have fallen (much fasterly).
I pay $2/mo for Google storage. I get generous free storage with Amazon Prime, but do not use it for fear of lock-in. Nothing to Apple. Free plan only on Dropbox. A little free Cloudflare R2, and pay-as-you-go Backblaze B2, which will eventually replace the Google subscription.
Am looking into backing up my iPhone photos with Ente + Backblaze.
Until recently, I had never done cloud development before, always preferring local, even on underpowered machines as a masochistic form of proof-of-lean-coding. Back when our entire eng team had top-of-the-line M2 Macbooks, we took pride in building on blazing-fast Apple silicon, while testing on older hardware so we might better understand our customers' worst case experiences.
But two things changed: my trusty Asus laptop, which I had miraculously saved from parts failure during the pandemic, cannot upgrade to Win11 from Win10. And my delightful Intel Macbook Air cannot update any more either, AND some important binary packages are now built using linkers that my decrepit workhorse can no longer handle (cough wrangler).
I migrated to a small VPS with Contabo for $4.95/mo, and ssh into a Ubuntu shell. It doesn't matter if I'm on Windows or Mac, desktop or laptop, my dev environment is always portable and sync'ed via GitHub. I might even be able to work in coffeeshops using a cheap, theft-immune Chromebook.
Also worth discussing the state of databases / data stores. It used to be that you'd run Docker on a VPS to get MySQL + Redis, and manage your own security and db admin including backups. Or, pay a lot of time and money for solutions in AWS/Azure. Or, pay less upfront, but get in bed with a horizontally-scalable PaaS/IaaS like Railway, Fly Postgres, Vercel, etc.
But now, as the age of AI agents is upon us, we enter a golden age of zero config serverless data solutions. Neon (recently acquired by Databricks for $1 billion) raised the DX bar really high. Upstash Redis and Cloudflare KV offer different price tradeoffs, and are trivial to swap and scale. Neon Postgres and Turso Sqlite make SQL databases just as easy, although their pricing plans force you into different optimizations for storage, compute, and even rows touched. Cloudflare D1 has been hit-or-miss for me, and Durable Objects is still tricky to deploy well. I really like Momento, which provides turnkey serverless caching built on Pelikan, a project that came out of Twitter's needs for scalability. I like how Sqlite can feel Mongo-ish with JSON functions, and nearly seamlessly transform into a column-oriented OLAP.
I'm excited about Evolu, a local-first Sqlite database with CRDT-based sync for scaling and distribution. It's still in its preview release, but if it works as promised (it demos decently well right now), maybe we don't need database servers any more, only file space, light compute requirements, and network connections that only need to be resilient, not consistent. Imagine a future of collaborative experiences like Figma or Slack, which don't need centralized servers.
source: Imgflip
