they let this madman build frames again

👀👀

simulatability appears to be the best predictor of how quickly ai will replace humans in a given domain. math can be simulated perfectly and agents are producing remarkable results on auto-pilot. software still needs business context, so agents are mostly on auto-pilot but still need human steering. hardware is a click above this, given slow feedback loops with manufacturing today, but it's likely a matter of time before we see the first agentic factories that are hyper tuned to manufacturing end to end. biology, on the other hand, seems infinitely harder to simulate given the quantum complexity of modeling cellular behaviors. is agentic acceleration on the field going to be measured in percentages and not orders of magnitude given this complexity?

Deel is now live with Tempo Zones! Contractors around the world receive stablecoin payments privately on Tempo. This is a world-first: an at-scale enterprise making private payments to end users on a blockchain. These private payments are fully interoperable with existing systems - users can continue earning rewards, withdrawing to their bank account and eventually will be able to spend with cards too.

back in the 90s, microsoft tried to rewrite ms word with a multiplatform design called pyramid. it went way over budget, fell behind on features and got axed. it probably wasn't the first rewrite that ended badly, but it became prolific and the consensus view became that rewrites were a bad idea for complex systems. i don't think this holds anymore. if you have well specified or highly tested software, agents can rebase your system onto a different language or architecture almost autonomously. most teams with mature stacks should be exploring rewrites more ambitiously, especially if its possible to unlock large performance or safety gains.

opus 5.5 refuses to build a fuzzer for a program it wrote. i appreciate the nuances of the problem guardrails are trying to solve, but we need a better solution asap.

as agents get better at long-running tasks, i feel my bottleneck is shifting entirely to my ability to multi-task across multiple problems at once. it's now quite easy to block the "main threads" on something an agent is off looking into for the next half hour, and have downtime to pick up and debug other problems. the constraint is having a stack rank of problems ready to feed in, and being able to manage the context in your working memory so you can jump between the right threads as they return.

underdog has a 10/10 consumer onboarding experience and the fastest setup path to a getting a local model very exciting to see local ai get easier to use!

congrats to @Cayden_Liao and veria for finding a decade old bug in a ~$100B cryptocurrency writing secure code is a different art today than it was a year ago. you need to be thinking more deeply about agentic scanning, fuzzing, formals methods and dst. it is simply not possible to just produce secure code in complex systems without these techniques.

i find myself using agents with voice commands often, even for coding. transcription has become good and agents can take a rambling train of thought and parse context well. the only time i reach for writing these days is when i need to go slower and clarify my thinking.

Very excited to see OpenUSD go live today. Today, stablecoin companies capture interest from the underlying reserves, making them an economically unattractive proposition for at-scale companies. OUSD shares all rewards with its partners by default, which will be a huge tailwind for enterprise adoption.

glad to see superintelligence is back

increasingly clear that most of math, cybersecurity and a good chunk of software engineering are fully solvable by models. agents are going to rip through any problem space where the feedback loop is fast, deterministic and digitizable. humans need to focus on areas where one or more of those conditions don't hold.

We need private blockchains for real-world payments. It's been hard to get this right for businesses and users. Most solutions are incompatible with existing markets and infrastructure, which is a non-starter. We've developed Zones, a new privacy model, in partnership with Stripe and other fintechs. They are fully compatible with existing defi markets, wallet infrastructure, card products and more. The key insight that lets us build a simple solution is that users often have a counterparty they're willing to trust in a transaction. Your payroll provider, for example, knows how much you're getting paid. Tempo lets your payroll provider create a private chain where all transactions are confidential between them and their users, while being fully compatible with the rest of the ecosystem. The post below goes into the details of how we made this work.

Mercator makes your agent more effective by letting it pay for tools and services. Agents perform well above their reasoning levels when they have access to powerful search and compute tools to augment their abilities.

Meow is a great product! We were happy customers at Farcaster, and now working together at @tempo to bring stablecoins to more businesses.

I've intellectually known that software engineering was going to go away but it's just started to feel real. The little dopamine rushes — learning a language, reconfiguring keybindings, upgrading packages or pushing around code on a page to make it easier to read — fade in the rear view mirror, vestiges of a profession no longer needed. The job is more director than architect or builder, and quickly rising to higher levels of abstraction. Agents may do the work of spewing code into files and servers, but building ambitious real world software is hard in a new way. We are responsible for houses we did not create and whose floors we've never truly walked. The longer you've been an engineer in the previous world, the harder the transition is now. But it's time to rewire your brain to appreciate the new ways of how things are going to be built. The ceiling on what we can do is so much higher now.

the most frustrating part of early mmos was repetitive landscapes and quest design, wow was particularly notorious. it feels possible to procedurally generate quests and dungeons (with some human taste and direction) with agentic llms to now build much more compelling worlds.