Zac Gottschall

@longtilwrong @unclebobmartin yes but i think a frontier model is much less likely than a human to write a comment that drifts from the actual code

@unclebobmartin @gott_zac Is it not the same problem? A bad comment is actually worse because the AI could make improper assumptions?

.@unclebobmartin In Clean Code, you wrote that comments are “a failure to express ourselves in code” but in the agentic era they help models index, navigate, and find relevant code faster have your thoughts changed?

this architecture makes intuitive sense has anyone run real evals on agentic workflows where thin harness + fat skills gives an order of magnitude improvement over plain prompting?

@omerokumusx @unclebobmartin uncle bob did have some timeless clean code takes: naming, small functions, single responsibility. I think these only help the models

@gott_zac @unclebobmartin Good point. That brings us to the issue of timing. Namely, the book was written in 2008 for human developers only. It still applies if you are coding on your own. However, with the incorporation of AI tools, the rule become more flexible.

@unclebobmartin I love it

@gott_zac I let the AIs write all the comments they like. They don't bother me because I don't have to read them. If it helps the AIs great. That's their affair, not mine.

@Samaytwt this is silly. anyone who’s actually hit PMF knows AI doesn’t just spit out great products claude’s a tool

Can you call yourself a founder if your entire product was built by Claude?

@bcherny, i desperately want to try out claude code desktop because the features seem really nice. but it feels so laggy. just pressing enter, it takes a second or two for the message to start to process. am i doing something wrong? is it my remote set-up?

Keep seeing posts like this. I’m not a designer, but it reminds me of when non-SWEs say coding is dead. AI writes the majority of code now, but it’s difficult to replace the taste and system thinking that comes with years of experience. How is design any different? Sure, I can generate something that looks functional very quickly. But designing a UX that actually performs is an art form I don’t see going away anytime soon.

@dickson_tsai latency is measurable. models will get good at it with the right tools. what they won’t crack is harder to define: knowing when a correct solution is actually wrong, optimizing the right thing, translating what a user asks for into what they actually need.

As you entrust agents to do more autonomously, be particularly aware of asymmetries with human behavior and expectations. For example, if you ask an agent to test your app interactively, it will do especially well diagnosing error messages but less well diagnosing latency/jank.

people who use both codex and claude a lot, when do you actually use each? is there a real split, or is it mostly vibes? i keep seeing people talk about having the models review each other’s code and i still can’t tell whether that’s an actual workflow or just lore

@howdymary and the workflows / model capabilities keep changing underneath you too. it’s not just adopting early, it’s building systems that get better when the models get better

ai anxiety is so palpable in sf because moats compound on an exponential curve now the divide between companies that have been quick to go ai-native vs the companies that have lagged integrating ai into workflows is widening every day pretty crazy to code switch from talking to an smb still paying $15k for a simple website makeover to a startup mass controlling thousands of social media accounts every day that you procrastinate in adopting the newest agentic workflows is a solidifying of distribution for existing actors, and it will be harder if not near impossible to break into in six months

@captainsafia i get this, but some of the best product bets look like side quests when they start. claude code probably did too, and now it feels pretty central to the whole story

Side quests and scope creep. I think Anthropic should focus on building really good infrastructure and APIs for their models and let these kinds of ideas thrive in the ecosystem.

engineering leaders: dangerous mode in Claude shouldn’t be banned, it should be contained build infra for running it safely otherwise you’re leaving velocity on the table

@ryanstp_ I told you so

Had a colleague tell me ~3 months ago they didn’t use their IDE at all anymore. At the time I kind of rolled my eyes. Now, I haven't touched my IDE in 2 weeks. For me, the last remaining problem was “verification” and now I just have claude run the code and give my HTML reports verifying the output. Wild how fast we’re going from “yeah right” to “of course”

@felixrieseberg Does this support remote ssh development

Today is a big day! We're launching a ~ new ~ version of Claude Code in the desktop app. It's been redesigned from the ground up for parallel work and is a lot faster. It's been my main way to use Claude Code for the last few weeks.

@bcherny might be a TUI bug: can't paste into the 'Paste code here if prompted >' field when logging into claude code on a remote machine. paste works fine everywhere else on the same box (and locally). anyone else hit this? or maybe i'm just being stupid. either way, plz help

We @a16z are leading @elise_ai's $250M Series E. Housing and healthcare are two parts of life where nearly all of us share the same frustrating experiences of how broken they feel, and yet they remain among the most expensive services we pay for as consumers. In my search for my first apartment, I texted 15 brokers. A few replied quickly. The rest took days or never responded. The apartments I toured were not necessarily the best fit; they were simply the first to answer. Once I moved in, interacting with my property manager was a lesson in waiting. Maintenance requests took days or weeks to see a reply. Even something as simple as a question about my rent or billing turned into ghosting by them. Healthcare is just as frustrating. Scheduling a doctor’s appointment means I’m sitting on hold for hours, repeating the same information, and submitting forms and follow-ups. Many of us probably put off care because of this friction. Getting our test results takes days, and the process of checking in or calling back feels like playing phone tag. Prescription refills should be instant, but often require multiple steps and unnecessary delays. EliseAI is helping to fix both of these industries as an always-on AI property manager and AI healthcare administrator. She schedules tours, handles billing, resolves maintenance issues, books appointments, manages intake, follows up on test results, and helps with prescriptions. She does this 24/7/365, without the frustration that we have all come to expect. Elise makes these interactions more efficient, more accessible, more human. No one should have to wait hours—or days—for answers about where they live or the care they need. With Elise, those answers come instantly. That is why EliseAI has passed $100M in ARR and continues to grow quickly. This new round of capital will allow Minna and Tony to continue building toward their mission of making housing and healthcare better for consumers, and I’m excited to support them on their journey.