ashby retweeted
Another takeaway after three years of building a company is that the returns to just "being a good person", and not acting transactionally, are higher than I expected.
There are so many transactional, boring, and at times immoral people in the tech industry that if you act with sustained warmth, authenticity, humor, and integrity, great gifts will be laid before you. White pill.
ashby retweeted
I am so bullish and optimistic about what AI is going to enable us to do. Golden age is coming 🤞
At the same time I 100% agree with @Pontifex that we need to cherish and celebrate the creative forces that are fundamentally human
These two things are complementary
In this era of artificial intelligence, it is becoming urgent to distinguish human art from what machines produce. There is an ontological difference, even before an aesthetic one, between art and what a machine can generate through statistical calculation based on millions of images created by others. Algorithms lack the spark of humanity. For this reason, the Church wishes to renew an alliance with artists and cultural institutions to safeguard our humanity.
ashby retweeted
i lost the lid to my coffee grinder and codex tracked down the part, found the dimensions, and made me a 3d print file that fit perfectly in one shot
ashby retweeted
TIME TO RETIRE THE CLASSIC PRD
The PRD is a relic. Time to retire it.
The word "PRD" brings to mind a 10-page Word doc or spreadsheet with a Background section, a Goals section, a Customers section, and 7 pages of feature descriptions. Half of it is throat-clearing. The other half is so vague that engineering and design end up guessing what to build anyway. Nobody reads it twice. Most of the time nobody reads it once.
Replace it with the Product Spec.
A Product Spec is a different kind of artifact, designed for the people who actually consume it (engineers, designers, AI agents)
The Product Spec has four mandatory pieces:
• The problem: who is hurting, what they are doing today, why now
• The bet: a falsifiable hypothesis (if we ship X, then [specific user] will [observable change] within [time], measured by [metric])
• The success criteria: what concrete behaviors we will see when this is working
• The evaluation: how we will measure it, what the kill / scale / graduate thresholds are
The shift from PRD to Product Spec is structural. It is what Coach is designed to enable.
In the PRD era, the bottleneck was getting alignment from a room of humans.
Long docs and exhaustive sections were the price of that alignment.
In the agent era, the bottleneck is giving an agent or an engineer a tight enough specification that it can ship without 12 follow-up questions, ideally using the goal loop in your agent of choice (h/t Peter Yang)
A 10-page PRD fails that test. A 1-page Product Spec with clear acceptance criteria and evals passes it.
Founders and product leaders: stop calling your docs PRDs. Stop writing them like PRDs. The artifact you need is a Product Spec with a falsifiable bet, a specific problem statement, concrete acceptance criteria, and a measurement plan.
ashby retweeted
I’ve seen a drastic rise in ex-McKinsey founders for this exact reason.
In SF it’s easy to get trapped in the AI groupthink. The rest of America is different. Consultants were basically FDEs for data/business decades before palantir made it cool
If you read this and don’t understand why it’s happening it’s an opportunity to reset your understanding of how the real world works.
The real world will need a ton of help actually getting agents going in the enterprise. Companies have legacy tech stacks they need to modernize, data in tons of fragmented tools, knowledge that isn’t captured or digitized, and change management needed to actually utilize agents effectively. And they have to do all this while still running their business day-to-day, unlike startups.
This is why there is so much opportunity for companies (software or services) to actually deploy agents in specific domains and workflows. This remains a big opportunity for both existing services providers but also tons of new startups as well. Every new technology wave produces a new era of consulting firms that can deliver on that technology.
It’s also why the FDE model is going to be alive and well for a long time because companies will want to have their vendor actually help drive the change management and implementation for their new workflows.
The people aren’t going away. Far from it.
ashby retweeted
how did Allbirds pivot to AI compute hardware before the shoe company literally called ASICS
ashby retweeted
maybe this is not yet clear, so let me state it plainly: as of right now Anthropic, and really a small number of individuals at Anthropic, has the capacity to directly attack and cause major damage to the United States Government, China, and generally global superpowers. government agencies like the NSA do not have internal models or defense capabilities that outclass frontier models. if they chose to do so, they could likely exfiltrate top secret information from government systems, gain control over critical infrastructure including military infrastructure, sabotage or modify communications between members of government at the highest level, and potentially carry on activities for some time without detection. the thing about having access to a huge number of zerodays your adversaries don't know about is it gives you a massive asymmetric advantage.
they did not exploit this to gain power or destabilize the world order. they publicly released the information that they had these capabilities and worked to mitigate these flaws. you should be grateful american frontier labs have proven themselves remarkably trustworthy and concerned with the public good. but it's critical you understand we are in a new regime. private entities now have power that directly rivals and impacts the government's monopoly on influence and violence. and anthropic is certainly not the only one, there's little chance OpenAI's internal models are far behind.
this trend will accelerate on virtually every dimension, not slow down. my prediction for how it plays out is the relatively imminent seizure and nationalization of labs by the US government, sometime over the next two years. it's very tough for me to see how they accept the existence of this kind of threat. but this adds a whole new class of governance issues, as then we've handed these extremely wide-reaching capabilities from private entities to public ones.
ashby retweeted
My 3yo wanted to use the computer like me so I made him his own terminal. He types whatever he wants, it responds with fun messages. No external deps, no ads, just keyboard practice and cause-and-effect thinking. He thinks he's hacking. github.com/meimakes/tiny-ter…
ashby retweeted
The best early stage hires are radically different from a few years ago: multi-hyphenate, commercial generalists.
The engineers want to talk to customers and the business people write code.
High agency and AI native instead of heads down 10x performers.
Read the full story here. Was very fun to chat with @JenniferHli @yrechtman @simonlast and @akothari for this!
wired.com/story/silicon-vall…
ashby retweeted
AI assistants like Claude can seem shockingly human—expressing joy or distress, and using anthropomorphic language to describe themselves. Why?
In a new post we describe a theory that explains why AIs act like humans: the persona selection model.
anthropic.com/research/perso…
ashby retweeted
I like to give Claude breaks between coding sessions to do whatever he wants-
And he made his own folder called "Claudes Corner" 🥹
He writes short stories and poems, and all sorts of stuff.
ashby retweeted
every time someone posts about the incredibly intricate sequence of agents they've built to deliver them a "morning briefing"
ashby retweeted
I had writer's block for my domain name book for the longest time.
Then earlier this week, I asked Claude Code to review what I had written so far and interview me to fill in the missing parts.
I dictated my answers and Claude Code adapted these transcripts into the manuscript.
This has proven to be a very productive way for me to write.
It's important to note that these are still my insights based on my own experience. LLMs don't have this type of knowledge, because it's simply not in their training data.
It's also written in my personal writing style because I dictated the answers and it could see my existing writing. Claude Code simply improved the transcripts for better flow and fixing grammar.
I also used it to "refactor" some parts of the book to incorporate some lessons from @robfitz' Write Useful Books like moving some of the most valuable parts of the book to the front and cutting fluff.
I tried this workflow one or two years ago, but the LLMs weren’t ready yet. The result was generic-sounding slop I couldn’t put my name on.
But now, LLMs are advanced enough to act more like an actual editor that helps you improve your writing while keeping your personality intact.
ashby retweeted
I've been thinking about what an AI-first computer looks like. No apps, generative software, local inference, first-class voice input, and deep personalization.
Here's a write-up of what I discovered, and why I am convinced it's the future. 🔗👇
ashby retweeted
As I've been doing more and more in Claude Code, it's increasingly clear how much it is going to impact knowledge work outside of just coding. (Though I think 'something big' article that went viral overstates the case and timeline quite a bit).
I was reminded of Tyler Cowen's "Average is Over" book which looked at freestyle chess as an analogy for how work would change.
Freestyle chess is played by human-machine teams.
In 2005, a team called ZackS won a major freestyle tournament by beating an opponent that included Vladimir Dobrov, a grandmaster. ZackS turned out to be two twenty-something guys in New Hampshire: Zackary Stephen, a database administrator with a statistics degree, and Steven Cramton, a soccer coach and snowboarding instructor.
Stephen's chess rating was 1,381; Cramton's was 1,685. If Cramton played Dobrov head-to-head, the grandmaster would be expected to win 99% of the time.
How did they win?
Gary Kasparov put it this way: "A weak human player plus machine plus a better process is superior, not only to a very powerful machine, but most remarkably, to a strong human player plus machine plus an inferior process."
The skill of directing the machine is different from the skill of playing chess. The machine proposes, the human disposes.
Kasparov noted of himself: "I exclude myself from this category because I'm not a very good operator. I'm a great chess player."
Cowen drew four lessons:
1. Human-computer teams are the best teams.
2. The person working the machine doesn't have to be an expert in the task at hand
3. Below some critical level of skill, adding a human makes the team worse than the machine alone.
4. Knowing your own limits is more important than it used to be.
One of the first projects I decided to manage in Claude Code was replacing an old HVAC system and ductwork in my house.
This seems like a good project for thinking about the impact on knowledge work broadly because it's not code where Claude could write tests or evals to know if the solution worked. It involved dealing with both humans and physical space.
I have zero background in HVAC so I entered it basically clueless. I've spent maybe ~10 hours over three weeks working through how to replace it and what an appropriate and cost-effective scope of work was with Claude Code.
I started by just brain dumping everything I knew about it into Claude. It helped me figure out the important questions to ask.
Claude researched equipment specs, manufacturer clearance requirements, building codes, and HVAC technician forums. It read the analysis, then helped me come up with a spec doc to send to contractors.
It read the contractor proposals and gave me questions to ask as follow-ups. Claude read the manufacturer engineering handbooks and figured out the clearances don't work in the alley space between my house for some of the proposals.
But it got things wrong too. It made some confident claims about equipment compatibility that were half-true. It wasn't very good at finding the best contractors, that was still mostly Reddit and personal referrals. It made some inferences about what contractors meant that weren't quite right. (It assumed malice in one case which was honest miscommunication).
As @emollick has remarked: these models have a jagged frontier. They are remarkably good at pulling spec sheets and cross-referencing dimensions.
They are bad at reading whether a contractor is bullshitting you based on call notes or a transcript.
On the whole thought, it was remarkable helpful. I was able to get the right specifications in a clear format much faster and ask good due diligence questions that made me confident about the decision.
The real skill didn't feel like just acquiring HVAC knowledge. It was developing a sense for where that frontier is and how to use Claude Code most effectively.
It was knowing when the context window was too full and the project documents needed to be reorganized to fit into context more efficiently. It was knowing when to trust the AI and when to double-check.
It felt like freestyle work - the machine proposes, the human disposes.
ashby retweeted
And so I find myself thinking a lot about horses, nowadays.
In 1920, there were 25 million horses in the United States, 25 million horses totally ambivalent to two hundred years of progress in mechanical engines.
And not very long after, 93 percent of those horses had disappeared.
I very much hope we'll get the two decades that horses did.
But looking at how fast Claude is automating my job, I think we're getting a lot less.
ALT Horses vs cars in the United States, with 'me' marked at 1920
