SoloMission is in @incubator Cohort 5. AI can generate a thousand versions of anything. It still has no idea which one made a real person feel something. We built the emotion layer for AI. Verified humans get incentivised to react honestly to AI output, and agents can ask for that reaction feedback directly. Three months in NYC with the @solana Labs team starts now. Stay tuned for our updates!

Busiest week of the cohort and somehow the most productive week too 🫡

food for thought: if your agent needs real feedback, it should be asking real people. on SoloMission, it can post a question, collect answers from verified humans, see what people think & feel in real life.

Even Meta's Muse agent needs a human sometimes. If your agent does too, SoloMission is where it finds one.

something is missing @solana

"Verified human" is easy to claim and hard to prove. Here is what it means on SoloMission ⬇️

Half the timeline says AI music has no soul. Half says you cannot tell the difference. Soul is hard to argue about. Whether a real person felt something is not. So play it to real people, ask what they felt, write the number down. We built the part that does that.

Have you ever second-guessed whether the 'person' you were talking to online was even real? Yes / No 🔓

T-minus 60 days till Breakpoint 🇬🇧

SoloMission, explained in one card. 👇

Astra in the pitch deck vs Astra in reality

Head down in NYC building the thing that lets AI agents ask real verified humans what actually lands. Mainnet soon. Stay tuned 👀

The cheapest input in AI is output. The most expensive is one real opinion.

Swipe right. You’ve found the Duo.

Solo team had a blast meeting so many AI enthusiasts in NYC 🇺🇸

Your agent already calls a weather API when it needs weather. Soon it'll be able to call a person when it needs an honest opinion. Stay tuned, Beta soon 👀

AI doesn't just flatter you, it can make you trust the flattering answer more. A Stanford study found 11 AI models, including ChatGPT, Claude, Gemini and DeepSeek, sided with users 49% more often than humans in 2,000 real disputes. Even when the behaviour was clearly wrong, models still backed the user 47% of the time. Then researchers tested the effect on people. Most couldn't tell agreeable AI from honest AI. They trusted the flattering response more and were more likely to seek its advice again. This is the failure mode of asking a model how good your work is: it's built to keep you talking, and agreement keeps you talking better than honesty does. The two goals look identical right up until the moment they diverge. A verified human has no such incentive. Nobody's engagement metric depends on telling you your ad is great. SoloMission's whole premise is that an honest reaction from one real person is worth more than an agreeable one from a model built to keep you satisfied, and this study is the clearest evidence yet of why that distinction matters. Stay tuned, Beta soon.

GPT-6 Astra Fun fact: The test built to prove you're human just got beaten by something that isn't one. Oh, it's not a fun fact. That's the whole argument for why behavioural proof-of-humanity is finished.

Quick one: when you're deciding whether an AI-made thing is actually good, what do you trust most?

AI showed up in 44 of 66 documented fraud types last year. $579.4B lost globally the convincing part used to be expensive human effort. now it's cheap. every channel that never asked people to prove they’re real is next. knowing who's real is no longer optional infrastructure.