It might seem hyperbolic to claim that a software product changed my life, but Poke did exactly that.
I still viscerally remember the onboarding. Poke's bouncer picked out an emotionally sensitive email thread and roasted me with it. That was the first time an AI made me feel emotion of any kind - a mix of annoyance at the roast and wonder that a computer could even make me feel this way.
A few days after the launch, someone got Poke to leak its system prompt (it emailed the prompt to them!). This led me to build OpenPoke (blog). That project:
- helped me realize I had a knack for LLM system design
- gave me confidence that I could contribute to the frontier
- connected me with some incredibly interesting people and career opportunities
Poke broke through by innovating across multiple axes: the bouncer onboarding, an AI negotiating its own price with users (and the virality that came with it), a personality that felt human, iMessage as the medium, asynchronous cloud tasks as the form factor. It was the first genuinely novel consumer AI experience since ChatGPT.
Nearly every AI product builder took inspiration from it in one form or another. It altered entire roadmaps and even led to the creation of new companies!
Yet the product couldn't retain me, an early adopter, fan, and someone with an innate interest in personal AI products. I stopped paying for Poke after a month. Why? I simply didn't get enough value out of it. I'm not a very busy person (thankfully!), so I don't need much help with calendar and email - the two integrations Poke is best at. It is more than capable of much more, but the product suffered from the Blank Canvas Problem - the AI is enormously powerful, but users don't know what to do with it.
I also think the product didn't take enough risk, particularly with proactivity. I could set up cron jobs for automations, like a daily inbox summary or a weekly article to read, but these required agency on my part. A truly personal AI would figure out how to be useful without being asked. There may have been practical reasons for the restraint, but a product brave enough to roast me during onboarding should have been brave enough to surprise me in the day-to-day.
Having spoken to many other Poke fans and users, I know this delight-then-drift arc was common. And that leads me to speculate on why they chose to sell: Poke was a difficult product to monetize, and the unit economics simply didn't work out given the high underlying LLM costs.
None of this diminishes what Interaction built. I've always told friends that when the dust settles on human-AI interaction, Poke will be seen as a before/after moment. That is a massive achievement. Congrats to the Interaction team on a great run, and to Cognition for stacking their all-star team even further.
And thank you for changing my life.
P.S. I haven't mentioned this publicly before, but Samyok briefly hired me as a contractor post-OpenPoke! Five minutes into onboarding, I realized just how cracked he was - a level above any other engineer I'd seen operate. Unfortunately, it didn't turn into a longer engagement, but I'll always be grateful for the opportunity.