one of the most challenging things about being an AI builder is dealing with terrible experiences and bad software. like why does the Xfinity app even exist.
I’m watching #GoogleIO and @rmstein is demoing something I tried this weekend… getting Google Workspace / #Gemini / AI Search / #Antigravity to handle our weekend planning... I can see the vision they’re demoing, but if this is live now, the actual experience still feels a little half-baked and under-delivered. Their demo is the exactly how I thought this would work, but the IRL results were that the wife was not impressed.
I mess with Gemini as my #3 go-to AI and with 0 expectations on Sunday tried giving it an email thread with a bunch of open-house listing links and prompted it to review the listings, build a table with the necessary details, and then plan driving directions for how we can best tackle it from the place we were having brunch.
This to me would be a dream scenario for the Google ecosystem, but each sidebar spawned or dropped context between apps, would randomly pull from emails years or decades ago (probably the last time we looked for a house), and the apps constantly reset because they lose connection or the app gets wiped when they go into the background. For me here's the big 3 things I noticed:
1) Fractured AI experiences. Each of app’s sidebar feels like a different experience and I’m not sure where the best place is to “do the work.” Do I make a Google Doc? (as an .md alternative), do I make a Sheet? (as a pseudo “db”), or is this supposed to be in the Gemini app? Or AI Search?! All of them felt terrible and I ended up going into Gemini… Unlike canvas experiences like the desktop versions, the mobile app just kept making new google docs.
2) Having “all” the context, is not necessarily a good thing. Some of the most /facepalm moments were when it would go into my email. This feels like the weirdest issue to prompt around and most UX here will probably be a huge waste of time. This is what really made me abandon ship on workspace and just go back into Gemini.
3) Technical Difficulties. I bought a Pixel 10 Pro mostly for the AI bundles and because they really won the last gen of AI assistants. But right now it's the little things that don’t seem to connect. The capabilities between mobile vs desktop are wildly different. Hand-off and often times just switching apps kills the entire backgrounded app. I get it, handoffs are hard, but this is the “killer feature” within the Google ecosystem.
This is cool, but in comparison, this just doesn’t feel like magic. Right now the awkwardness of this roll-out is it feels incredibly fractured across all the sidebars, mobile apps, and technical challenges. In the end, if we weren’t comparing this to OpenClaw’s “it feels like magic” this would be amazing. Right now it made me feeI’m nol “stupid” because I assumed it was going to work. It’s hard here because google really does have a higher bar to clear: this needs to work whether I’m on a phone, computer, and while I’m driving and using maps. It looks like they’re promising this with Antigravity in search, but seriously, this needs a clearer place to live.
#googleio #antigravity
this is great! i ran into some similar issues with my harness and connecting to a decision repo made things work like magic, here's what i noticed so far:
1. 'planning' workers should have access, 'implementing' sub-workers should not. (keeps from drifting)
2. 4.7 update. prompt changes and new default tool access have unpredictable results. (i think i'm competing with some memory vs docs etc...)
3. cron pruning and archive on releases. something i haven't implemented yet, but is on my backlog is to start culling old docs and keeping the important ones fresh. probably needs to happen on release steps anyways.
Some days working in the AI deep end all day feels great. You've maxed out your human context window and you're fully spent like a session of mental hot yoga.
Today was not that day. Definitely felt like an unsexy game of twister and my agent forgot how to connect dots.
Most underrated super power in AI work environments is just being able to focus and read massive amounts of .md and streaming chat dialogues.
It's sad to imagine you could get the solution to world peace in an output and most people will just tl;dr summarize for me
RIP: @Tesla premium connectivity + unlimited @grok . I won't lie I loved just using grok in the car while driving. But with the new limit I have save my voice chats like it's ammo.
Won't lie 2x Claude code is a good trade though.
We’ve agreed to a partnership with @SpaceX that will substantially increase our compute capacity.
This, along with our other recent compute deals, means that we’ve been able to increase our usage limits for Claude Code and the Claude API.
@trq212 What a great guide. I'm swapping over from Cursor at home and Claude Code at work and it's been amazing to see how Claude Code has evolved over the past month. Love the insight around tools being grown out of and just giving the model what it finds useful.
🐻 Here's MunchiDex, built in 24 hrs for the @lovable hackathon. It's an inspired Pokedex that lets you use pictures of food to catch Pokemon. Check it out at https://t.co/AK5dWxkXvs
https://t.co/qizuRHarDw
Welcome to AuraVale💫
AuraVale is a cozy live sim where your words shape your story.
🎨 Hand-Drawn World
👭 Co-op w/ Friends
🧑💻 UGC Scenarios
🏆 Early Access 2025
Wishlist today at https://t.co/SxmUC3oNXv
#cozygame#lifesim#auravale
🎉The ultimate VR basketball game Blacktop Hoops, has dropped on Quest, Steam, and Pico. ✅Campaign ✅Bots ✅ Casual ✅Competitive. Do you have what it takes to become the GOAT? @ycombinator
🥽 Buy on Quest:
https://t.co/DYbMS01kA8
🖥️ Buy on Steam:
https://t.co/72W8yuzHHT