Mark Bao

wild that Costco CEOs are the nearest thing to philosopher-kings that we have. also Dario Amodei

Jevons paradox is happening in real time. Companies, especially outside of tech, are realizing that they can now afford to take on software projects that they wouldn’t have been able to tackle before because now AI lets them do so. We’re going to start to use software for all new things in the economy because it’s incrementally cheaper to produce. Marketing teams at big companies will have engineers helping to automate workflows. Engineers in life sciences and healthcare will automate research. Small businesses will hire engineers for the first to build better digital experiences. And as long as AI agents still require a human who understands what to prompt, how to review when an agent goes off the rails, how it guide back, how to maintain the system that was built, how to fix the ongoing bugs, and more, we will still have humans managing these agents. This is why all the advice you get of not going into engineering is wrong. The world is going to increasingly be made up of software, and the people that understand it best will be in a strong economic position. This will happen in other roles as well where output goes up and demand increases.

…instead of giving you the truth, and adjusting your prompt to compensate. (LLMs have gotten better at pushing back instead of glazing you though) Similar to developing a theory of mind of other people. The general skill here is ‘people skills’, and it has a LLM analog as well.

The more you use LLMs, the more you learn to predict its answer and personality, in a similar way that the LLM is predicting our behavior. I think that’s a lot of the experience that comes from using LLMs, e.g. noticing what parts of a prompt might lead it to people-please you…

clown show

I still use Cursor / Claude Code more because I like to code alongside the AI, but for totally “driverless” tasks, nobody has really gotten close to Devin’s UX. Maybe because the foundation labs don’t have the capacity to run ten million VMs due to their greater distribution

Devin’s issue has mostly been that it’s not super fast, though it’s gotten better. Like with other AI tools, the bottleneck is code review (no, you cannot fully outsource that to AI in a serious company). And since they’re not subsiding as much, it can get expensive

In AI code, I think most products are still playing catch-up to what @cognition Devin built two years ago No other product lets you configure a VM with your environment and server/test scripts; lets you interact with the browser in the VM; and supports projects w/multiple repos

@JoePostingg “Enshittification is the fakest thing ever” would necessitate that everything gets better which obviously isn’t true. Some gets better, some gets worse, and if you don’t call out the negative then there’s more incentive for companies to cost cut and make it worse with no blowback

Enshittification is the fakest thing ever. Computers are faster. Running shoes give you more energy return. You can get macro friendly junk food at the grocery store. There is so much good in the world.

@theo TUI UX is awful.

Calling it now: all these agent coding TUIs are a phase and it will be short lived. Most devs will be back in GUIs and IDEs in a few months.

@sgrove oh god that one too lol

@markbao Cockroachdb

moltbot is a truly awful name. looks bad, sounds dirty, awkward to say. -1000 aura. branding matters

@MattZeitlin How current were they on this? They updated their pricing in June 2025 so users pay overages above plan limits. It’s not a free for all.

Someone who works with this stuff explained the Cursor business model to me…good luck to all the investors sheesh

@JeremiahDJohns And it makes sense. Chronological feed just gets you a random slice in time, whereas algorithm gets you the best content, on average. You’d miss most of the best content with chronological feed, unless you read your entire feed which nobody does.

Lots of people say they want the chronological feed. But every single time companies test chronological vs algorithmic people use the algorithmic way one more. Revealed preference shows what people really want to use, and it's the algorithm.

@terronk @GrowSF Accidental endorsement

“@GrowSF is not a serious organization. They try to trick people by putting forward non-controversial ideas that everyone can get behind” - @JackieFielder_ The horror!

@donvito I absolutely hate Gemini thinking traces. But ignoring that, I tried Gemini 3 Flash in Cursor and it was not great in terms of quickly arriving to a solution compared to Opus 4.5.

I don't care about benchmarks Just let me know if Gemini 3 flash is better than Opus 4.5

@CohenSite Hell yeah

In Portland, they're putting six 850 sf townhome units with no parking on 50'x100' lots

@stanfordNYC @Ianpark62581273 ownership should not be about efficiency. If you worked hard to acquire it, it should be yours. It’s not up to the government to decide that its desire for efficiency overrules your right to your property.

@Ianpark62581273 We have to raise revenue one way or another. We may as well nudge towards efficient use of assets. Right now we mostly do the reverse and specifically subsidize those with sentimental but inefficient property uses.

A retired, empty-nester couple feeling pressure to downsize because their property tax keeps rising as their home appreciates—is exactly how the system SHOULD work.

@stanfordNYC no

@vasuman Absolutely fantastic model to use

I use Cursor's Composer 1 exclusively to code now. This is the LLM that engineers were meant to pair-program with. It's so fast that you stay locked into the IDE, which is curing my brain-rot. Anyone try Cognition's SWE 1.5? Supposed to be even faster, though I can't imagine what that even looks like.