G Shiva Shankar (@monozyg) header
G Shiva Shankar (@monozyg) profile picture

Building @flockagents - a personal assistant OS that builds your own suite of agents and apps that talk to each other.

Joined July 2012
Photo by G Shiva Shankar (@monozyg)Photo by G Shiva Shankar (@monozyg)Photo by G Shiva Shankar (@monozyg)Photo by G Shiva Shankar (@monozyg)Photo by G Shiva Shankar (@monozyg)Photo by G Shiva Shankar (@monozyg)Photo by G Shiva Shankar (@monozyg)Photo by G Shiva Shankar (@monozyg)Photo by G Shiva Shankar (@monozyg)Photo by G Shiva Shankar (@monozyg)
Customized software brings the best of design and agents. We just need a way to make it easy for everyone to create ones for them without going through all this hassle.
So Claude can design an app for me and it can code that app for me. I have built a lot of apps now but I always run into dead ends when it comes to usability. You know how I fixed this problem. I asked a designer friend to take a look at my design system for this app I am building. They worked with Claude for 2 hours and sent me a markdown file with their guidelines. I use that app I built everyday now. I think I just learned something valuable about the AI world we live in. I just quite articulate it yet.
1
68
Something like this
Meta's Muse is getting shopping carts powered by connectors. Instacart is going to be first. There is a long list of retailers including Walmart that said theirs are coming soon, too. This is different from ChatGPT/Claude plugins. Those render the plugin's visual output as an in-chat mini app. Instacart's connector on Muse will only handle data exchange - Muse will render the shopping cart and other UI using Muse-native widgets. (screenshot from the connector's docs)
1
39
It's always gratifying to do this small UI delight things. And glad to not follow the trend on thinking spheres/triangles.
38
These guys had the sheer gall to build a space tech company in their final year of under grad and make Indians proud everyday. Haters gonna hate. Keep the great work going guys !
300+ Indians have spent years of their lives building satellites and critical technology for this country. They deserve better than having their work and patriotism questioned by manufactured controversies. I owe them a response. Here it is:
45
Yesterday, @patrickc talked about chatting with your agents inside your apps. Today, @karpathy talks about building your own personalized suite of apps over and above existing app providers. This is exactly the bet we are taking at @flockagents. Build your own personalized apps that can talk to agents & can be shared publicly or a trusted group of people. A whole new world of usecases gets unlocked with this.
Weekend hack: HackerTube. I was annoyed with YouTube's recommendations so I had Claude filter down 43M comments over 20 years of HN to 6,454 videos that keep being recommended. *Significantly* better viewing experience, ~2 hours with Opus 5.5: hackertube.org/ - Excellent recs without any logins (state is kept in local browser session) - Thumbnail is a representative frame from middle of the video instead of the super duper A/B test hyper-optimized user uploads - Underlying data from the comment analysis as one zip file, point your Claude at it and ask for your own version - HackerTube Longs ✊ A bit more on the process: I use Claude Code (desktop app for Mac), Opus 5.5 on Medium effort as my default. I only talk to Claude and have it do everything, the only "manual work" I helped with in tandem was in the browser (reminiscent of MenuGen): 1) I had to create some kind of API key from Google Cloud UI to query YouTube for video metadata, 2) I booted up a VPS box for serving and 3) I bought the domain name. Once the box was up I gave Claude the IP and told it update, secure, harden, deploy and optimize the code, and it did the whole thing. Many awesome projects that would have normally taken weeks/months even can now be few-shotted for breakfast. What else are we not thinking of.
1
1
92
It's crazy that we picked up a paradigm before it became cool. @flockagents helps you build apps that can talk to your agents from within the app. it's not just to improve the app, but also to do tasks within the app.
I'm now building an "improvements" feature into all of my personal vibecoded applications: a freeform text input that's hooked up to an agent that yolo implements whatever is submitted. It's an interesting experience -- the activation energy barrier to sanding any rough edge or implementing any new feature becomes extremely low. Someone else who uses one of the applications I created has already implemented 150 improvements over the course of a few days. This pattern feels like a new kind of tool -- liquid software? -- and every application that doesn't have this feature now has a slightly rigid and anachronistic flavor. Why can't I just tell it how I want it to be different? Now, it's not obvious to me how this pattern can or should scale. Does it become unmaintainable after a while? (Maybe model intelligence improvement outstrips my feature appetite such that it's fine?) How does collaborative software work in this model? Can software designed this way be secure and reliable? Whatever the larger questions about broader deployment, it definitely works great in a personal context. It reminds me a little of the visions of Lisp/Lisp machines, Smalltalk, and HyperCard -- systems that the user can take control of and modify. Those visions never quite worked, though: the vast majority of software today is not end-user modifiable in any useful way. So, while this "liquid" affordance is a very small touch, the way it changes one's relation to the software feels a bit transgressive and empowering in a Lickliderian and Engelbartian way.
1
36
The biggest challenge in India is the latest pricing updates have made it prohibitive for building WhatsApp assistants. And building shared agents for family / friends is also impossible. nitter.cf/signulll/status/211662…
one of the most interesting things about using a personal agent like dots through imessage is that it truly stops feeling like software. as if something is entering your world as opposed to you entering its which is almost every other product. there is this visceral & maybe subtle feeling behind it that is quite powerful. a lot of ppl don’t like it but damn it’s so easy to use, & feels so damn natural esp when ai uses emojis + reactions well etc. kinda reminds me of how even babies intuitively are able to figure out how to unlock an ipad.
39
Making your own apps where you can chat with your agent unlocks so much potential. The fixed UI let's you do quick actions. Agents perform complex tasks and also get to update the app's UI.
2
74
What if this could happen from your own Claude subscription right from your own laptop ? This usecase was the exact genesis for us to build out a locally hosted, personal subscription led personal assistant. Coming soon !
It's happening. My wife told her friends how I set up a @bot to handle weekly meal planning and grocery shopping, saving 2-3 hours per week. They lean forward. "I want that! How do I get that?" The AI factories are pitching travel agents, but mundane convenience seems like a killer feature to usher the Instagram crowd to Muse/Bot/Dot/etc. Most people right now "use AI" the way I use a Swiss Army knife to lift the tab on a soda can, so everything she describes next lands like "The Turn" of a magic trick. Every Sunday, "Sylvia-bot" sends 15 meal ideas and we choose 5-7 based on the week. It knows to assign the easy meals to the nights when our family calendar shows football practice. It knows to use ingredients 2-3 times in different ways, like a chef balancing COGS and flavors. It knows the components of a successful taco night. We had years of weekly grocery orders for training, so it knows the brands, flavors, and sizes we prefer as a family. (I wonder how long it will take others to get to this point without the archive.) We're still working on the idea that "grape tomatoes and cherry tomatoes are interchangeable" and "don't buy berries that aren't on sale!" In 5 minutes, we approve the plan. It does the shopping on H-E-B, double-checks the pantry assumptions (e.g. soy sauce, spices, etc.), adds the kids' usual extras, and emails a clean PDF of every recipe for the week. At the end of each week, we log our favorite to a 'recipe book' memory for future rotations. For now, my wife will scan the cart (another 5 minutes) and confirm tomorrow's pickup window. Soon, I expect our Tesla will do the grocery run on its own.
82
If its 2 % in the US, it's a rounding off error in any other market.
98% of US households aren't paying for AI yet More charts in State of Markets II: a16z.news/p/state-of-markets…
30
Claude literally gave this as the first option when I was designing my agent's avatar. Is this what they mean by regressing to the mean, when everyone uses AI to produce the same thing ?
53
Supporting ChatGPT login with @openclaw and @NousResearch and also launching Dot on the same day is a crazy move tbh
30
Add Chat to it and it essentially kills Microsoft Office, Slack, and a whole bunch of other tools. And makes it hard for a business to move out of OpenAI. The real question is whether you as a business want to get locked into one AI provider forever.
Announcing ChatGPT Space, a collection of new collaborative experiences right in ChatGPT. Edit rich notes with live visualizations together with others and your dots. Whether it’s a TODO list to delegate, a product plan or meeting notes. Working with others and your agents is seamless and they also be automatically kept up to date in the background. For developers, this is AGENTS.md taken one step further.
37
This is a great sign for people building harnesses. Forces Anthropic to keep allowing builders to use the Agents SDK within the plan limits.
Wow sign in with @ChatGPT and letting people use their ChatGPT plan inside your app is a game changer
1
34
Every model company will build a Grok bot competitor. Question is whether you'll keep switching between each one or use a harness that allows you to plug in any model you want - even open source ones.
If Anthropic builds a Grok bot competitor it's basically GG. After Opus 5.5, I never want to touch any other model again.
35
Impact on AI agents is also going to be huge. It was possible earlier to have a few templated messages and send them to initiate a conversation and then make use of the 24 hour consumer window to have infinite conversation for free. With this update, Whatsapp as a channel just became super costly for Personal assistants.
Whatsapp is now going to charge every message sent from October Costs will be 14 paise for each message, a 10,000 daily user support now costs 2L per month from 0 earlier SMS era rates are back, expect a lot more businesses to move to apps Upside maybe reduced Whatsapp spam
1
87
Let's go! Time to build.
Jev by @typesafeai is now on OpenRouter, in beta. Jev is a System One model. Instead of generating text, it takes your app's state plus a typed question and returns a typed decision with a probability attached. There is no JSON prompting, parsing layer, and nothing to validate against.
1
44
One of the most high value things to do while coding with AI is to maintain the product architecture docs mapped to your Claude.md files. Create a skill for updating this documentation and treat it as a pre-PR step just like running tests. Ship the skill for your whole team so that nobody misses to update it. This is useful for not only for your new coding sessions but also for new joinees who want to understand the code. They dont burn tokens for understanding the code.
5
79
And it's already started.
Just nuked my Instinct account. Tried it out for 48 hours and it does feel like the future of personal AI assistants (had it shop for new homeowners insurance, book a doctor's appointment online, search for travel, manage my LinkedIn notifications). I'm just... not yet comfortable hooking up unfettered read/write access for my entire life to an AI.
1
85
This is not surprising. In B school we had a case study, on Tesla . My professor asked at the end of the class - did @elonmusk decide to launch the family car because it was a larger market or was it a deliberate plan to launch the sports car first. We all argued that it was not deliberate plan but just market dynamics. Then our prof pointed us to a small article in the bibliography "My Big Secret plan for Tesla" where he literally drew the plan for this much before he even launched Roadster.
I’d be surprised if Grok Bot stays $300 a month long term, and not because I’m an Elon fangirl. 😅 I was a Starlink beta customer at $90 a month. It went to $120 after beta, then eventually there was an $80 option. They’ve sent me an updated router for free and even offered me a free Mini with a reduced monthly subscription for being a longtime customer. FSD was $10K while in beta. Now I can subscribe for $99 a month and turn it on and off if finances get tight. I got early access to X Money because I bought a year of Premium+ the day it came out. Now it’s rolling out to Premium, and eventually X Money is supposed to be available to everyone in the US. What I’ve learned from watching all of this? The early adopters pay for the product, and sometimes they pay the most while everyone is building it, testing it and breaking it. Then once it’s ready for everybody, the cost can be spread across a much bigger group of users. So for now, I’m perfectly happy sitting on the sidelines watching the nerds build and break Grok Bot. 😂 Then hopefully the rest of us get the cheap, easy version we can just pick up and use. And honestly, this is what makes me an Elon fangirl. 😂😅👇
1
3
291
Also, I am not sure if the same argument can be given for Grokbot. A serious cost item here is not the model. It is the infrastructure of cloud computers that every bot gets. You need massive scale for this to bring a meaningful reduction in cost & I am not sure if the market is so big. Right now it is all tech enthusiasts / VCs who are hooked on it. But running data centres in space with Solar energy is also something that nobody expected. So who knows 🤷
1
35