- Eventually, the product develops a character through what its makers pay attention to, what they decide not to do, and what they keep working on after it technically works, and that is the soul of a product. You feel that someone cares about how the whole thing comes together karrisaarinen.com/posts/someone-behind-the-software
- It’s natural for people to want to work on things that they know aren’t going to fail. But incremental improvement is guaranteed to be obsolete over time, especially in technology, where you know there’s going to be non-incremental change. To really be different, you need to focus on changes that wouldn’t emerge from incremental improvements to what you already do x.com/kevg1412/status/2118021567766737163
- Writing used to be evidence of compliance, effort, and learning, but it is no longer evidence of anything. LLM-produced writing makes evidence of compliance even more useless than it already is. It totally destroys evidence of effort: if you need a hundred-page report, you can generate one in an hour with a few dollars in inference tokens. Evidence of learning is likewise broken, since we can produce a detailed status report full of information that never passed through anyone’s mind at all (and certainly was never retained) seangoedecke.com/writing-is-no-longer-proof-of-work
- Newer models may need fewer instructions than the ones we accumulated for older models. Give an agent a clear goal and a way to verify it; if the goal is still unclear, have it interview you until it can write a spec, then start a fresh session with that spec. Claude Code removed over 80% of its system prompt for newer models without a measurable drop in coding evaluations, so it’s worth periodically removing our own old rules and adding back only what the agent actually misses x.com/addyosmani/status/2115688494865465370
- Success is not having to know anything you don’t want to, and wealth is not having to do anything you don’t want to. But perhaps the better goal is to move more of life toward things you can do and enjoy https://x.com/vgr/status/1158098372953178113
- Anthropic is preparing for an IPO just five years after its founding, with $33.2B in revenue in the first nine months of 2026, about 5,300 employees, nearly 12,000 customers spending over $100K a year, and an operating profit in Q2. The striking tension is that most revenue depends on ongoing usage, and over 80% of Q3 revenue came from models released in the preceding six months, while Anthropic has committed nearly $900B to compute over many years: it has to keep winning demand for a rapidly changing product using capacity secured far in advance tanayj.com/p/anthropic-s-1-breakdown
What I’ve been reading (W45, 2026)
The soul of a product, non-incremental change, why writing no longer proves anything, giving agents fewer instructions, success and wealth