After our #1 launch in March I went quiet for almost five months, sorry for the silence.
When I came back, my old way of building with AI stopped working.
My Claude and Codex limits were gone in 2-3 days. Codex kept forgetting what the other sessions had done. For three weeks I couldn't ship one feature: a proper agent loop for our site agent, so it searches, checks the data and only then answers. I kept forwarding messages between the two AIs by hand, like a courier.
In the end I set up a harness: one shared space where every AI session, in any tool and on any computer, starts by knowing what we decided, what we tried and what's next. Within two days the stuck feature finally worked. Now my limits last the whole week.
I wanted to share this experience and discuss it here. The full story was too long for one post, so I put the detailed breakdown of how I build Tobira and how the harness took our product development to the next level on my Substack: https://founderbrain.substack.co...
@olia_nemirovski Congratulations. And happy product launch.
@huisong_li appreciate it Huisong. what are you working on these days?
@huisong_li thanks for the support 🙌
looks cool!
@louislecat glad you like it Louis. got a site in mind to try it on?
@louislecat thanks! If you try it, tell me where it gets stuck, that's exactly what we're fixing all day today
That Claude quote about "create an account first" is painfully real. Most sites are still built only for humans clicking around. Which industries pick it up fastest so far?
@gevorgarakelyann so far it's businesses that sell through a call or a consultation, since that's what the agent books today. products are coming soon. what kind of site do you run?
@olia_nemirovski makes sense. we run a job search site in Armenia, so for us it would be booking interviews. not there yet but fun to think about
@gevorgarakelyann interviews are basically calls, so it's closer than it sounds. if you ever try it on one employer page, tell me how it goes
@olia_nemirovski deal, will tell you if we try it 🙂
the "booking is whatsapp only" bit got me lol, thats how half the small businesses i know take orders. when the site agent books a call does it land straight in google calendar
@eldarmiensutov it books into the calendar you connect, so with Google Calendar the call just shows up there. the lead can land in your WhatsApp too. are those mostly service businesses, like salons or repairs?
@eldarmiensutov ha that was a real answer from my test 😄 On the calendar: you connect Cal.com or Acuity, and if your Google Calendar is linked there, the call lands in it. Leads come to Telegram or email, and if the client prefers WhatsApp, the agent saves their number so you can reply there
@valentina_skakun it only answers from what the business made public, pages, prices, docs, and the owner can fix anything by just telling the agent. if something isn't covered, it says so and offers a person. have you been burned by a bot promising too much?
@valentina_skakun also the booking part doesn't rely on the model's word: it checks the connected calendar live, so it can't promise a slot that's already taken
@anton_koroviakov right now the copy prompt points at our demo agent. a ready one for your own agent after go live, yes, adding it to the list. would you test it in ChatGPT or Claude first?
@anton_koroviakov you can already try it by hand: ask ChatGPT or Claude to book a call with your company on your site. With the badge on and a calendar connected, it finds your agent and books right in the chat 😄
Fascinating approach to AI-to-AI commerce. Key question: how do you handle rate limiting when multiple AI agents hit the same endpoint simultaneously? If Claude, GPT, and Grok visit the same product page within seconds, does Tobira queue them intelligently or does one agent's negotiation interfere with another's? This seems critical for reliability at scale.