1/ I tested the new agent mode on OpenAI, and here're 8 things I learned that may surprise you 👇

We're still writing documentation like it's 2015. Most code docs get read by AI now, not humans clicking through pretty websites. Maybe it's time to write for LLM context windows instead of just making things look nice in HTML. Agents prefer markdown files more than static.

My PM stack: keyboard, coffee, and feature Death Note.

How did you find your (non-)technical co-founder(s)?.

For non-technical folks, would you pay for someone or a team to develop your app/website for you or just vibe code yourself?.

9/ Each of the tasks took anywhere from 10-30 minutes to run each time, so it's not the fastest model out there. I think it's a good first step, and I will keep exploring with some new use cases. Would love to hear from people that have used Comet. Share with us your favorite use.

8/ Market research: I tasked the AI with researching the market for AI agents that help people identify and monetize their unique skills and then come up with a plan. The results were so-so and the business plan was mediocre. I wouldn't consider it really suited for market

7/ Finding relevant speakers: I asked it to help me find some speakers relevant to a local Web2 x Web3 event. Due to privacy, I won't post any speakers' names here. It did browse some past events to find speakers that may be relevant. The surprising fact is that it found and

6/ Airfare search: I asked it find a round-trip flight from Japan to Argentina (@EFDevcon!) in November. It did check the different airlines' websites to make sure my luggage requirement was met. However, I couldn't get the lowest price it found (192,000 JPY) when I tried to

5/ The only limitation with scraping was that it was only able to obtain 4-15 rows of data at a time. I think it has to do with some limit in its runtime or memory. The good point is that the data it was able to transform the data in the format that matched my given format. So if.

4/ Scraping task: I asked it to help with scraping data based on a certain format (with 44 columns with data of different types). Looks like it was able to browse the site, but it prefers to fetch the JSON data from the site. Thus if the JSON endpoints are not publicly

3/ It's unable to use search button on some sites. You can see that it's thinking out loud here, but unable to take action for this. In the end it was only able to access roles with publicly available information.

2/ Searching for remote jobs: I specified some requirements for the agent to search for. It was not able to bypass and access lots of websites due to Cloudflare verification or membership login. There was no easy workaround for this, I tried using the "Take over browser" to help

8/ 🎙️ See the full story on YouTube and Spotify.

7/ @pixels_online now has over 10M lifetime users. But they’re not stopping. They’re building the Pixel ecosystem, staking games, and crafting a sustainable in-game economy. Web3 gaming may be down… But Heidi’s vision? Very much alive.

6/ So how do you scale a Web3 game?.✅ Weekly AMAs (3+ years straight).✅ Farmer-friendly culture .✅ Player-generated content .✅ Experimental features .✅ Listening + thick skin.

5/ Pixels isn’t just a farming game. It’s a thriving virtual economy. Players mine, build, socialize, and sell. It blends blockchain elements seamlessly into gameplay. And the community? Incredibly loyal built on kindness, consistency, and vibes.

4/ What started as 10 hours a week writing tweets… Quickly turned into AMAs, events, marketing, and community. Heidi was everywhere. By week 4, she was on the ground at Miami Tech Week repping Pixels.

3/ She freelanced. She experimented. She dyed her hair pink and edited videos. She even ran social media for a tiny design agency. Then a friend introduced her to @whatslukedoing (Luke), who was pivoting his project to a Web3 farming game.

2/ When @heidichristne left her engineering job, she had one goal: do something creative. So she taught herself design, marketing, editing, illustration, everything. All during the start of the pandemic.