we just fired time zones. We’re building Tonebird.ai as your 24/7 personal agent, with a memory for the people in your life: who they are to you, what you promised them, and how you usually talk. Your client shouldn’t need a fresh introduction every time you ask AI to help you reply. So we made a Tonebird film where San Francisco goes to sleep, Seoul keeps working, and the deal closes before the alarm goes off. The sun can keep its schedule. We’d like ours back.

the part I keep getting stuck on is how a startup gets big enough in the first place. Google has a whole graveyard of products it had the money to keep alive, and they still died. money doesn’t make the ball roll. I think the early version of katamari is getting big in a tiny corner. a few thousand people who can’t stop using you, in one very specific job. that’s enough to know which capabilities matter. you don’t need to be big everywhere, just deep somewhere.

we fired the keyboard. your pet got the job. zero typing experience. slept through onboarding. knocked something off the desk. but he does understand “tap the phone twice,” which is apparently enough to get a draft started now. we connected Tonebird to iPhone Back Tap through an iOS shortcut. it reads the chat on screen and suggests replies in your tone, so our newest employee can focus on his core competency: touching things he shouldn’t. a human still picks the reply and sends it. he’s already asking to work from home.

We just fired the keyboard. two taps on the back of your iPhone → three replies to choose from. we built a Tonebird.ai iOS shortcut that works with Back Tap. once set up, it screenshots the conversation on your screen, reads the context, and drafts 3 replies in your tone. pick one. copy. paste. send. your group chat gets a reply. your thumbs get a day off.

X ALGORITHM UPDATE (for accounts under 500 followers) 🚨 I have 236 followers. So today I read X's open-source algorithm to answer one question: what does it actually do for small accounts? More than you'd think. But only for 2 hours. Here's how it works, straight from the code (updated Oct 21–24): 1. Every time someone opens For You, X saves one seat for a small account. It picks ONE post that is: • from an account with ≤50K followers • an original (not a reply, not a repost) • under 2 hours old • seen fewer than 200 times on Home …and lifts it to around #16 in that person's feed. That seat is how strangers find you. At 236 followers, my followers aren't my distribution. That seat is. 2. The seat goes to the post with the best like rate. X runs a tiny lottery on every eligible post: likes ÷ views. Everyone starts at an assumed 1.5%. • 0 likes in 100 views → you look like ~0.5%. You lose. • 3 likes in 100 views → ~2.5%. You're in the race. The top 2 lottery draws go head to head on predicted engagement. The winner gets the seat. So your first 100–200 views are an audition, and the audience is mostly your own followers. 3. This week's update made the audition fairer. X can only score 2,800 posts each time you open the feed. Before, it took them first come, first served, and brand-new posts (under 2h, under 8 likes, under 500 views) were at the back of the line. They got cut first. Now every source gets cut by the same %. New posts are no longer the first thing dropped. 4. What each reaction is worth (weights on predicted probability): • Copy-link share: 20 • Reply from a mutual: 20 (5 + a 15 mutual boost) • Reply / quote / DM share: 5 • Follow: 4 • Repost: 1 • Like: 0.66 • Bookmark: 0 And what a bad reaction costs: • "Not interested": −52 • Mute: −59 • Report: −234 A 1% chance someone taps "not interested" cancels out an 80% chance they like it. Reaching the wrong crowd is worse than reaching nobody. What I'm changing at 236 followers: → Originals over replies. Replies and reposts can't win the seat. → Post when my people are awake. The 2-hour clock starts when I hit post, not when anyone sees it. → One strong post > three okay ones. They'd compete for the same audition. → Write for my 236, not for everyone. Their like rate is my ticket to strangers. → Talk with mutuals. Their replies count 4× a stranger's. → Make it worth pasting into a group chat. That's the 20. Source: github.com/xai-org/x-algorit… Bookmarks are worth 0 to the algorithm. Copy-link shares are worth 20. If this helped, you know which button to press 🐦

the bird has been doing all the talking lately, so hi, it’s me 👋 i’m one of the humans behind Tonebird.ai. we’re building it for that message you keep opening, thinking “i’ll reply later,” and closing again. it remembers your relationships and the conversation context, then drafts a reply that sounds like you. also, we’re giving away gift cards, keycaps, and a smart ring this Halloween 🎃 plus a free Pro offer! everything’s in my pinned post. giveaway ends Oct 31. you can ghost your inbox for another minute. come say hi first.

The scariest thing this Halloween? The message you're ghosting. 👻 Ghosting Awards 🏆 Drop the scariest message you ever got, ghosted or not. Prizes: 👻 Scariest Message → smart ring 💀 Most Cursed → giant keycap light 🕯️ Most Relatable → keycap set 🎲 Random draw → keycap set To enter: 1. Follow @Tonebird_ai 2. Repost this 3. Reply with the message (blur the names). Your boss's Friday "quick question". Your ex's 2am "u up?" 🎁 Plus: everyone gets 1 month of ToneBird Pro, free ($40 value) ↓ Ends Oct 31. Winners Nov 1.

Everyone's racing to be AI-native. Fewer people are asking what gets lost along the way. The answer: friction. The arguments, the pushback that only happens when humans actually talk to each other. Agents don't disagree with you. They execute your spec. At the earliest stage, your spec is almost always wrong. Look at @PodiumHQ. The second run to $150M took 2.5 years, but it didn't start from zero. It started with 8 years of judgment built through customer calls, failed bets, and internal debates. AI didn't create that judgment. It made acting on it faster. That's the trap. AI compresses execution. It can't compress knowing what to execute on. A lot of teams automate to look AI-native, not to get better. Automating before you understand the problem just locks in your wrong assumptions faster. Incumbents have judgment but no speed. Startups have speed but no judgment yet. 
The winners will compound judgment fastest. Automate execution aggressively. Protect the arguments on purpose. AI is a multiplier. A multiplier on zero context is still zero.