Tao Dong

something i've noticed: AI agents create a weird new kind of burnout. esp for young people. a lot of ambitious 22 year olds are going to think the answer is simple: - spin up more agents - ship more code - sleep less - outwork everyone and for a while, it will feel incredible. you can keep multiple agents running, feed them tasks, review outputs, fix mistakes, make decisions, and keep the whole loop moving. the problem is that the work no longer drains you through typing. it drains you through judgment. More attention. More context switching. More verification. More decisions per hour. so instead of 8-10 normal productive hours, you might get 4-5 extremely intense hours before your brain is fully cooked. and you feel numb until you sleep properly and reset some of my friends are already burnt out. they don't say it out loud but i can tell. the agent can keep working 24/7. the human still has a hard limit

This is a nice article demonstrating the trend Interaction Design is increasingly about designing the agent-computer interface and the agent-to-agent collaboration patterns. Same high-level principles, different subjects.

@nlycskn App users are builders 😉

My partner's building an MCP server for basically every app he uses 😂 Now he just orders lunch and browses Zillow through a custom chat app he built himself. He literally never opens the apps anymore. We’re not just in a new era of building, we’re in a new era of app users 😁

Great article! I agree that manual error analysis is so often underappreciated. “If you are not willing to look at some data manually on a regular cadence you are wasting your time with evals.”

Finally had time to check out this course. It includes a nice explanation of the relationship between skills and MCPs and how these technologies can work together to solve realworld problems using agents. Highly recommend!

Nice article on evaluating agent skills by @tessl_io. +1 to the point that providing public visibility of skill quality for creators and users is critical to ensure the health of the skill ecosystem. tessl.io/blog/anthropic…

Great primer on agent harness

Agent engineering is like parenting. Adapt your approaches as the underlying model grows. “But even then we often saw Claude forgetting what it had to do. To adapt, we inserted system reminders every 5 turns that reminded Claude of its goal. But as models improved, they not only did not need to be reminded of the Todo List but could find it limiting.”

“You want to give it tools that are shaped to its own abilities. But how do you know what those abilities are? You pay attention, read its outputs, experiment. You learn to see like an agent.” So beautifully said.

@ZhiruoW @FariaHuqOaishi Nice work!

Most agents either run fully autonomously or interrupt at the wrong times. What if agents know when YOU want to step in? 🚀Introducing PlowPilot - a web agent that adapts to your interaction patterns achieving +26.5% user-reported usefulness Huge credit to @FariaHuqOaishi for leading this project!

@sethladd Congrats on shipping!

One of our most requested features!

@OfficialLoganK Well, it’s technically still there but few users will likely see that section at the far bottom. The simplicity of the new design is probably great at letting returning users jump right into actions but it’s at the expense of inspiring first-timers.

@taodong still there at the bottom, just trying not to overload people

Experimenting with new AI Studio vibe coding start screens today, New vs Old. What do you think?

AI is now "managing up" their human supervisors.

Gemini's browser use feature within Chrome is very handy when I had to deal with user-unfriendly UIs. I just helped me found a "Rename" button that was hidden when the file is running 🤷🏻‍♂️.

Vibe-coding interactive demos to learn Reinforcement Learning in @GoogleAIStudio was fun.

@sethladd Congrats, Seth! Labs is lucky to have you!

Some bittersweet personal news: I’m transitioning to a new adventure within Google joining the Google Labs team.🧪 This means I’ll be stepping back from my active role in the Flutter community here. It’s been a privilege serving you all. I'm now a Flutter user, and I'm cheering on the team and all of you from the sidelines. Thanks for everything, Flutter fam! Keep shipping!💙

TL;DR: Show agency and behave like you were already hired.

Sidenote: I also learned about a companion tool called Petri from the post. While Bloom measures, Petri is for exploration. You use it first to broadly audit a model and find weird behaviors. Once you identify a trait, you move to Bloom to quantify it. anthropic.com/research/petri…

My takeaway: Bloom scales evaluation by turning anecdotes into statistics. However, it remains a simulation. Use it to quantify trends, but do not treat it as a substitute for real-world testing.

I just finished reading up on Bloom, a new open-source framework for automating behavioral evaluations of AI models. It proposes a way to scale up how we measure alignment, but relying on simulations raises some interesting questions. Here are my notes 👇