Pinned Tweet
Taico got a major face lift!
I've always tried hard to keep it general purpose and not turn it into another coding platform. But deep integrations with PRs are so nice.
Now it's smart enough to detect when PRs are mentioned and it follows them, surfacing status by color, CI with spinners and other details when you hover.
Threads can also watch PRs and react to changes.
It's just really nice.
But the best part is that none of this shows up if you don't do coding stuff. Taico is still general purpose but can follow you if you are a coder.
Another thing I intentionally didn't build was showing the agent's steps. The rationale was that Taico was built to delegate tasks and move on, working async. But sometimes I found myself wandering why something was slow or stuck. So I integrated a realtime feed showing what the agents are doing. It's hidden by default, tucked away against the right border of the task.
Not gonna lie, I want a MacBook Pro with a glowing Apple logo. Such a nostalgic throwback.
But I am perfectly happy with my 2022 Air. Turns out I need less compute at my fingertips with every passing year. I am accumulating a lot of intense compute but it's all remote.
I have more engineering capacity than I have product decisions to feed it. Think about how absurd that would have sounded 3 years ago.
I'm evolving Taico to support the next leg of growth. I'll need to rework the core identity and authorisation platform. I'll need a tenancy model with users / admins / teams / orgs. I'll need strong isolation. This will require a lot of engineering work to implement.
But here's the thing: I don't think about that at all. I am actually focused on what the product should feel like, where the boundaries should be, and how the journeys should work. At the same time I'm mapping architecture decisions in my mind so I can build the right foundational capability.
And that's the hardest part. Strong product + engineering instincts.
Implementation is not scary at all. In fact I have spare quota of frontier AI that's going to expire in 2 days. I have nothing to feed it yet. I haven't landed on a product and architecture that is elegant and scalable yet. My agents are sitting there waiting for me.
And then I realised this was the situation, almost like a meta observation. It's such a fundamentally different way to build things. I can think big. So big in fact that it takes a lot of mental effort to clearly see the solution I want to land on. And I'm 100% confident that the implementation will get done in a few days with little to no supervision from my end.
We are living in a different world.
Kelsey Hightower is an interesting phenomenon. Absolute legend in the kubernetes world. Completely clueless about AI (which is fine) but with strong opinions and some sort of superiority complex.
Had beers with coworkers. One asked me what I did for fun. “Coding, building” I said.
“I used to enjoy that but I can’t stare at screens anymore” they said.
“What if you didn’t have to stare at a screen?” I said, pointing to my phone, with a beer in my hand.
Feels like they got 14 lite to a point where they’re happy to stop iterating and can focus on lower priority platforms like (checks notes) my Model 3 in Australia?
The fuck do you mean GPT Luna is 7.5x cheaper than Gemini Flash?
I’ve been accepting Gemini’s shitty performance assuming the price was lower. But no. And it’s going to be 15x once the promotional period ends.
Out Gemini - in Luna
1/ This morning on the bus I asked ChatGPT to investigate the new sign in with ChatGPT feature to offload Taico’s threads agent to my subscription.
5/ GPT Luna driving the threads feels soooo much better than the Gemini flash 3.8 I was using before. It’s not even funny. And “free” with my plan.
