01
The spike answered
Two hundred DMs over six hours all get a real answer in minutes — not the eleven you got to before your flight.
X agents

Your agent, on X
AI agents for X
A post that works produces more DMs than you can read, mixed in with pitches and bots. Your agent answers all of them in minutes, asks the questions that separate serious from curious, books the call when it is worth one, and leaves you the five conversations that matter.
TL;DR
What it handles
01
Two hundred DMs over six hours all get a real answer in minutes — not the eleven you got to before your flight.
02
It asks what you would ask — what are you building, what have you tried, what is the timeline — and the answers do the sorting.
03
Forcing questions and one thing to go do before the next message. The people who do it are the people worth your hour.
04
A time is proposed and confirmed in the DM only when the answers justify it. No link dropped into a conversation that had not earned one.
05
Growth pitches and copy-paste openers get nothing, so your attention goes to the humans who wrote something real.
06
Written from your own posts: short, direct, no exclamation marks you would not use.
A real DM
The DM does not get a form, a link, or an auto-reply telling them to email support. It gets the conversation you would have had if you had the time to have two hundred of them.
An X agent vs a DM autoresponder
The standard answer to DM volume is an autoresponder pointing at an email address or a booking link. Both read as a closed door, and both fill your calendar with the people who click, not the people who qualify.
| Capability | Your X agent | DM autoresponder / link drop |
|---|---|---|
| What the DM gets | An answer, and a question back. | A link and a "we will get back to you". |
| Who ends up on your calendar | The ones who did the thing it asked for. | Whoever clicked first. |
| Spam and pitches | Ignored, silently. | Auto-replied to, politely, forever. |
| Your voice | Written from your posts. | One paragraph you wrote in 2023. |
| Memory | Knows this handle DMed you in April. | Sends the same auto-reply again. |
| Handover | You get the thread and the reason it matters. | You get an inbox. |
Start from a complete template
YC office hours in a DM thread: the six forcing questions, real pushback, and one assignment they have to go do before the call.
What comes with it
Evidence
Published research on response speed and AI-assisted replies. None of these studies tested Keiki — they measure the behaviour a DM agent is built on.
Inbound goes cold fast.
A study of firms’ responses to online enquiries found most reply far too slowly to reach a lead while it is still live — the same window a post-driven DM burst has.
The Short Life of Online Sales LeadsHarvard Business Review, 2011
AI-assisted writing is faster and rated better.
In a preregistered experiment with 453 professionals, ChatGPT reduced time spent on mid-level writing tasks and raised output quality as judged by blinded evaluators.
Experimental evidence on the productivity effects of generative artificial intelligenceNoy & Zhang, Science, 2023
AI assistance raises how much one person can answer.
Support staff using a generative-AI assistant resolved nearly 14% more issues per hour, with the largest gains for the least experienced.
Generative AI at WorkBrynjolfsson, Li & Raymond, NBER Working Paper 31161, 2023
Further reading
How DM conversations are read through the X API.
X Developer Platform
What an authorised app may send on your account’s behalf.
X Developer Platform
Access tiers and authentication, if you would rather run your own app.
X Developer Platform
Request access at onkeiki.com, open the Office Hours template, connect your X account. About 10 minutes to every DM getting a real answer while the post is still moving.