Slack search is somehow still stuck in 2010. No semantic search, no vector search, no personalization. Zero awareness of who you are, what team you're on, or what you're working on. Just good old, raw, full text search. Type one wrong letter and you may as well have hallucinated

The future of proactive agents is event driven. We don't need scheduled cron jobs or agents running around trying to predict what we want. We need agents that understand enough about us to respond well when something actually happens. For example: someone reports a bug →

The grader hierarchy is practical: 1. code based - deterministic where possible 2. model based - LLM based where necessary 3. human - human for calibration. You can get so much signal from simple pass/fail tests on outcomes, instead of jumping straight to LLM judges. Does the

Assume good intent. People rarely act out of malice. When someone does something frustrating, there's usually a reason: missing information, missing context, a different perspective. Starting from that assumption changes things. You ask questions, look for what's blocking them,

The minimum viable audience is now one person. People are building tools just for themselves, stuff that would never make sense as a product. This wasn't worth doing before, because it was too much effort for a personal payoff. But now you can have something working in an

https://t.co/uUILJP60YN

Mobile app development feels cracked open again. App Store submissions jumped 24% this year. Ideas that would've died in planning are making it to production, and solo devs are shipping what used to require teams. For a while the ecosystem felt closed; everything had been

One way to makes users to stay is embedding yourself into how they distribute their software. Build and release DevOps tools have the highest retention of any software category because of this. For example, ToDesktop can do 99/100 things wrong but if they can reliably gets your

This is one of the more careful treatments of the AI employment/job replacement question I've seen. Distinguishing between substitutability (can AI do the task) VS substitution (does it make economic sense to actually replace humans).

The Pragmatic Programmer nails it again Absolute goldmine in the AI Coding age

Dead code is technical debt that accumulates silently. Code hygiene shouldn't be a quarterly crisis, but rather regular maintenance built into how you work. Here's an example of a Claude code sub agent whose job is to find and delete old code.

The paradox is that "real" is about to become the most manufactured aesthetic. What actually holds up will be the stuff that's costly to fake: long track records, shipping when no one's watching, relationships that predate the narrative, outcomes you can verify.

I think about something an English teacher told us a lot: kill your darlings. The things you're most attached to are often the things holding you back. This applies to building software too. AI capabilities are moving so fast that you can't afford to be precious about your

>be me, looking for founding engs for @andocorporation >jordan shows up >first thing he does is open a doc >defines the problem clearly >asks why we're even doing this >lays out a few implementation options >talks through tradeoffs for each one >maintainability, extensibility,

Something I’ve noticed in Korea is that AI isn't a culture war here. It's just tools. Grandparents use translation apps, kids use it as study tools, just people using it as you’d use electricity. There seems to be less friction for adoption compared to some other countries,

AI capabilities are shifting so fast that the product you're building today might be irrelevant or even obsolete in six months. It's easy to get stuck building features nobody wants because shipping feels like progress and talking to users feels daunting. Especially if you're

The time you save is just time, and what matters is what you do with it. This person used AI to skip the parts they couldn't do and spent their time on the parts that actually mattered; like thinking critically, verifying the issues were real, and approaching the problems like a

I love this roomba analogy. It bumps into the same corner 5 times and you're like "just go to the left!!" But then you come back an hour later and the floor is actually clean. The inefficiency at the micro level didn't matter because it could just keep going. I think for