Introducing Human-in-the-Loop (HITL): a personal AI agent hub running on your phone, on both iOS and Android.
Quick to set up. Any model provider. Your data stays on your device if you want it to.
Try it now in beta 👇
@aiwithsally I agree, I want to be the person to make the final click. The hardest part is finding the item comparing with competitors, discounts and filling in all your details, if the agent can do all this and I can make the final click that saves a me a lot of time already!
@immad I agree in the agentic future principle, but not in the winning product form. Having been deep in this long enough the harness will matter more than the model as things scale and no one yet (imo) yes nailed the human/agent ux of the future
@Junda_Chen_ yes loving antigravity now as well...I am using opus 5.5 a lot less now as both hydrogen and carbon make an impressive pair for most of your development needs
Still hunting for a free table, site after site?
"Book me a table for 2 at Jake's Famous Crawfish in Portland this Saturday at 7:30pm…"
My AI agent opened the booking site, found the slot and held the table. Then it handed it back to me for the final tap.
It does the legwork. You make the final click.
Just to say I am impressed at the speed of development with a new update everyday..the issue I am now overwhelmed with orchestrator, swarm and now dynamic workflows, so not sure when to use what and lean back to my scriped workflows via skills. Suggest an auto mode that then decides how that can be done with the cost, speed and quality tradeoff before work starts.
The privacy bar may favour the incumbents. Most people already share their email, photos, location and messages with Apple, Google and Meta, so an assistant on top is a small step for them. A startup has to earn that trust from zero. But the giants are moving at a snail's pace compared to startups, so the real question is whether trust beats speed.
@SierraPlatform@Meta@Shopify Good to see asking the customer built in. The next question is how often. Some people want a confirm on every purchase and others only above a certain amount. That should be the user's setting.
@shravanthreddyr@viticci Drafts only is a smart default. The agent does the work and you keep the final say. Later you will want some things to just go through, so that should be a setting you control e.g. by certain recipients
Devs have already had their agent moment. Coding agents now do real work in the background without being asked. Most people are still at the chatbot stage, where each upgrade feels small. What is missing is the everyday experience. Most personal agents still need hand holding or a complicated setup, so only the tech savvy get in.
@signulll Default on-device dictation on iOS and Android is poor, necessitating paid tools like whisperflow. This matters less when text goes straight to an agent/LLM, as it handles typos and cleans output. Regardless, cleaner transcripts are preferred
I disagree. I don't want it to ask first, and I don't want it to be completely silent. I want it to set the schedule, tell me that it did it and offer to change if necessary. That is always a personal preference, so you should be able to set the dial yourself. Not every personal agent lets you.
@DavidPawlan The pattern across all four is the same. Agents getting their own way in, whether that is email, a courier booking or a watch things are moving fast!
@venturetwins@omooretweets Retailers who earn from selling will welcome agents. The ones who earn from you being on the page will keep blocking them. Shoppers will go where their own assistant is allowed in just like with the shift of internet and mobile.
@MelinaVale14 The more an agent knows about you, the more it can do, so privacy and usefulness pull against each other. There is no clean fix. The best middle is knowing exactly what you have shared and being able to take it back.
@aiwithsally Fewer things to worry about is the right measure. An agent that chases the loose ends and only interrupts you for a real decision. How much it interrupts should be a dial you set.
Interesting what helped me is that in some of my projects I have asked (this is opus 5.5 main orchestrator agent) to own the technical decisions but I still own the product decisions, so far this has worked well. Let me see how it goes with Fable 5.2 with the more explicit mandate
@jamescoder12 Agentic Siri in 2027 means the people building agents on phones today have a long head start. The giants are moving at a snail's pace compared to everyone else. The real unlock is a standard way for apps to be called, with the same consent screens people already trust.
Disagree on one super agent for everything. Personal stuff, family, work and side projects all in one feels like too much, and I suspect it only works once we reach AGI. Until then people will want a few focused ones, with some things kept separate..or Chief of staff model managing the other ones