AI PRODUCT HOMEWORK·AI Product KOL / Growth Evangelist, Manta Network
ⓘ Independent job-application page. Not affiliated with, endorsed by, or operated by Manta Network. Public information as of 20 October 2026.
Homework for my application to Manta Network AI Product KOL / Growth Evangelist.
What Manta's AI products are, what the X channel data says about reaching their users, and a content plan fed by live AI trend data. The demo is a working launch desk for those products: every product fact re-checked live against its source, a trend radar, EN and 中文 drafts checked sentence by sentence, a channel read and a feedback digest.
AI products live in 2026: Incognitive plus seven under Agentese
703k
@MantaNetwork followers; a typical post reaches about 15k to 20k
843
followers across @Agentese_AI, @Yarrow_ai and @LifeClaw_AI combined
1 of 61
top-level posts since February is a video
15.7k
median views on AI product posts, against 24k for chain news
00
Summary
The situation
Manta has shipped a real AI lineup in 2026: an offline private assistant, an agent that makes phone calls, a job-matching agent, a forecasting engine and developer tools. The homepage now leads with Incognitive. The audience on X came for the chain.
The finding
AI product posts on @MantaNetwork get fewer views than recaps and chain news. The product accounts have under 900 followers between them. Almost nothing is shown on video, though most of these products demo in under a minute.
The plan
Show each product doing one real task, in short video and threads, in English and 中文. Ride the AI news cycle with a data feed that already tracks edge models, OpenClaw-style agents and AI apps. Grow the product accounts in the places their users already are, and send every piece of feedback back to product.
The one line: let the products demo themselves, and point each demo at the people who already use Claude, ChatGPT, Cursor and OpenClaw.
01
Manta in context
A privacy chain since 2020 that now builds consumer AI. The levers a growth evangelist works with, as they stand today.
Lever
Today
Growth read
Brand story
manta.network leads with "Meet Incognitive", a private AI model, and adds "Private AI App" to the main menu. The line running through it is privacy, carried over from the ZK chain.
Privacy is the one story that links the chain years to the AI products. Use it.
AI products
Incognitive (iOS, live since August 2026) and the Agentese family: LifeClaw, Little Pink Potato, Yarrow and YarrowPulse, Meeting Copilot, CodeAutrix, Kopi, plus the open-source CC Remote Approval plugin.
Eight products, each with a one-minute demo in it. See 02.
Chain
Manta Pacific runs on. Manta Atlantic has stopped producing blocks; the token migration window closes 16 Nov 2026.
Migration reminders still out-pull product posts. Keep them, and keep them separate.
Channels
X @MantaNetwork 703k followers. Discord 170k members, Telegram 25k. A YouTube channel is linked from the site. Weekly "Ecosystem Recap" thread every Monday.
Large reach on X and Discord; almost no video. Short video is the open format.
Product accounts
@Agentese_AI 287, @Yarrow_ai 523, @LifeClaw_AI 33 followers. Incognitive has no account of its own.
Each product needs its own audience. The main account can lend reach.
Languages
Agentese and CodeAutrix ship English and 简体中文 copy. The site also runs in Korean and Indonesian; the Kopi domain tests a Malay and English page.
中文 content has a home; the Chinese AI user base is large and active.
Team
About 30 people across Singapore, Hong Kong, mainland China and the US (Manta job board). AI-assisted building is house culture.
A small team that ships fast needs content that keeps pace with releases.
Sources: manta.network homepage, agentese.ai catalogue, App Store listing, Manta job board, @MantaNetwork on X, Discord and Telegram invite counts. Read 20 Oct 2026.
02
The AI lineup
What each product does, who it is for, and the one demo that sells it. I would lead with the first three.
Product
What it does
First audience
The demo that sells it
Incognitive lead
Private assistant that runs on the iPhone. Download a model once, chat and work with documents offline. No account, no cloud. v1.1.1 picks the best model the phone can run (Qwen3.5 0.8B and 2B).
People who paste work documents into AI: lawyers, doctors, founders, finance teams. Frequent flyers.
Airplane Mode on, a contract PDF in, three questions answered. 30 seconds, screen recording, no edits.
LifeClaw lead
Finds the place, phones the business in the local language, books, cancels or asks, and reports back. Also a hosted MCP server and a Skill for Claude Code, Codex and OpenClaw-style agents.
Builders who already run agents; people who avoid phone calls.
"Book me a dentist for Friday" typed into Claude Code; the call happens; the booking comes back. Split screen with the call audio.
Little Pink Potato lead
Searches thousands of live jobs, scores each against your resume (skills, salary, location, fit), sends alerts with a match score, tailors the resume for ATS. Free; agents use your own API key.
Job seekers in tech and crypto, a loud and active group on X.
Upload a resume, get a ranked shortlist with strengths and concerns. Then show the cost of 100 matches on my own key.
Yarrow / YarrowPulse
Multi-agent forecasting with the reasoning shown, graded against what happens. YarrowPulse sets forecasts beside Polymarket odds.
Two maps. What a user would pick instead of each product, and how other crypto teams with an AI story win attention.
Product
What users reach for today
The opening
Incognitive
ChatGPT (about 900M weekly users), Gemini, Apple's on-device model, desktop local runners such as LM Studio and Jan.
The big apps send prompts to a server; the local runners need a laptop and setup. A free phone app with no account sits between them.
LifeClaw
OpenClaw (544k followers on X) and its skills; Kimi Claw; Genspark's agent, which also makes phone calls.
Phone calls from your own number, packaged as MCP and a Skill. It plugs into the agent people already run.
Little Pink Potato
LinkedIn and job boards; asking ChatGPT or Claude to rewrite a resume.
Scoring every live job against the resume, with alerts, is work the chat apps leave to the user.
Yarrow
Polymarket and Kalshi odds; analyst newsletters.
A forecast that shows its reasoning and gets graded, set beside market odds.
Crypto team with an AI story
X followers
How it wins attention
Manta's gap or edge
0G Labs
663k
"Trust layer for AI" chain; large ambassador and KOL programmes, strong 中文 community.
Manta ships end-user apps a person can try today. 0G sells infrastructure.
Kite AI
569k
Agent payments chain; partnership-led news.
Same: Manta has finished products to show.
Virtuals
292k
Tokenised agents and a crypto-Twitter launch culture.
Manta's apps carry no agent token, so the pitch is utility. That reaches users outside crypto.
MyShell
175k
Consumer AI apps with a creator community, strong in Asia.
The closest comparable. Its creator programme is the model to study.
NEAR AI
14k (NEAR: 1.8M)
Founder-led talks on private, verifiable AI; IronClaw, its own OpenClaw-style agent.
Same privacy theme. NEAR has a founder voice on it every week; Manta's privacy story has no face yet.
OpenClaw reference
544k
Demos, skills, a lobster mascot and a creator, @steipete (608k), who shows his own work daily.
LifeClaw already uses the 🦞. The playbook to copy: frequent, unpolished demos by the people who build.
Follower counts read via api.fxtwitter.com on 20 Oct 2026. App usage from data.prompt20.com/api/apps (a16z and company figures). OpenClaw-family rows from data.prompt20.com/api/claws.
★
What the channel shows
From the demo's Channel tab: every @MantaNetwork post from mid-February to 20 Oct 2026 that fxtwitter returns (110 own posts, 61 top-level), grouped by format and topic. Views are X's own counts.
Finding
Numbers
What I would do
The audience came for the chain
Median views: chain and token posts 24k, weekly recaps 22.6k, AI product posts 15.7k. The most-viewed post since February is the Atlantic deprecation notice (99.7k).
Treat @MantaNetwork as a loudspeaker and build the AI audience on the product accounts and in AI circles.
Almost no video
1 video in 61 top-level posts. That one, the Incognitive demo clip on 19 Aug, did 14k views, in line with the account median.
Make short screen recordings the default format for every product. Post natively on X, YouTube Shorts and 小红书 / 抖音 for 中文.
Threads lose readers fast
Thread replies median about 1.2k views against 15k+ for the first post.
Put the product, the task and the link in post 1. Keep threads to four posts.
Product accounts start near zero
@Agentese_AI 287, @Yarrow_ai 523, @LifeClaw_AI 33. When @MantaNetwork quotes them, the quote gets 10k to 17k views; the product accounts keep little of it.
Reply under the AI creators these users follow, with demos. A useful reply from a product account earns followers faster than a broadcast.
Big brand moments work
The 19 Oct reply-quote to a @Tesla post ("AI that works even in airplane mode") got 11k views and showed the right instinct.
Keep a short list of daily AI news hooks (new models, outages, privacy stories) with a ready Incognitive or LifeClaw angle. The radar in the demo produces this list.
Method: api.fxtwitter.com profile timeline, paged to its end (14 Feb to 20 Oct 2026). Topic tags by keyword (recap, chain and token, incubated dApps, AI products). It covers the posts the API returns; the demo reads the same feed live.
04
Public-facts check
Small things I found while reading the product pages against each other. Each is a five-minute fix, and an evangelist is often the first to notice them. The demo's Lineup tab re-runs these checks live.
Where
What I saw (20 Oct 2026)
Suggested fix
@MantaNetwork post, 19 Aug ("See Incognitive in action")
Links apps.apple.com/…/id677697739, which returns 404. The correct ID is 6776977391; one digit is missing.
Use one campaign link per channel for every App Store CTA, so a typo shows up in the numbers.
Agentese catalogue → Kopi
The catalogue describes Kopi as an approval copilot for Lark and links hellokopi.com. That domain now shows a different early-access page, a "GL Appeal Assistant" for Malaysian insurance claims.
Point the Kopi card at the Lark product, or retitle it.
Incognitive App Store listing
The seller is Manta Ray Labs, Inc.; the developer name shown is "asMatch". The website says "by Manta Network".
Match the App Store developer name to the brand, so search and trust line up.
Incognitive messaging
The listing says "No cloud". The October update adds online search, which leaves the phone.
Say it plainly: chats stay on the device; search goes online only when you ask.
05
JD duties, my plan
Duty in the posting
How I would do it
Detailed in
Share product content across X, YouTube and community channels
One demo per product per week, cut three ways: X native video, YouTube Short, Telegram and Discord post with FAQ. 中文 versions for X and 小红书.
07
Create threads, demos and short videos
Screen recordings of the product finishing a real task, captioned, under 45 seconds. Threads of four posts with the whole point in post 1.
07, demo
Track AI and Web3 trends, turn them into content
A daily read of new edge models, OpenClaw-family agents, AI apps and prediction markets, each row mapped to a product angle.
06, demo
Build early user communities, bring feedback to product
Start where the users are (Claude Code and OpenClaw users for LifeClaw, privacy and travel circles for Incognitive). A weekly digest of themes, counts and quotes to product.
08, demo
Launches, campaigns, growth initiatives
A launch kit per release: fact card, demo clip, thread, 中文 thread, community post, three creator briefs, tracked links.
07
Explain complex capabilities simply
One analogy, three steps, one honest limit. Example: "Incognitive is a calculator for words: it works without the internet because the model lives on the phone."
07
06
Trend engine
The trend work runs on prompt20.com, an AI data site I build and run: 16 interactive tools, 74 public JSON datasets at data.prompt20.com, 191 guides on blog.prompt20.com, plus AI news and events boards. Its pages drew 231,821 Google impressions in the 90 days to 18 Oct 2026. The demo's radar reads these datasets live.
Qwen3.5-0.8B, the model Incognitive added in 1.1.1, is in the set; Liquid's LFM2.5-350M launched by benchmarking against it. "What a 350M model can do on your phone" is a ready explainer.
The OpenClaw family: Kimi Claw, IronClaw, secure-openclaw and others, with what sets each apart. LifeClaw adds phone calls to agents like these through MCP or a Skill.
Four pillars, a weekly rhythm, and the first five pieces. Every claim comes from a product's fact card; every piece passes a lint for unsourced numbers and hype before it ships.
Show it working
Unedited screen recordings of one task. The core format.
Explain the tech
On-device models, MCP, multi-agent forecasts, in plain words with one analogy.
React to the news
New model, outage, privacy story: same-day post with a product angle.
Users and builders
What people did with it, MCP recipes, community questions answered.
First five pieces
Format
Angle
The airplane test
30 s video, EN + 中文
Incognitive in Airplane Mode reads a contract and answers three questions. Caption: what stays on the phone, and what the online search mode changes.
I gave Claude Code a phone
45 s video + 4-post thread
LifeClaw MCP installed in one command, then a real booking call. Aimed at Claude Code and OpenClaw users.
100 jobs, scored
Thread + screenshot
Little Pink Potato against a real resume, and what it cost on my own API key.
What fits in your pocket
Explainer thread, EN + 中文
Qwen3.5 0.8B, Gemma, 1-bit Bonsai: how phone-sized models got good enough, with Incognitive as the example.
Forecast vs the market
Thread, then a follow-up
One YarrowPulse forecast that disagrees with Polymarket, posted before resolution and graded after.
Sample: post 1 of "The airplane test"
Airplane Mode on. No Wi-Fi, no signal.
I gave Incognitive a 12-page contract and asked who pays if the delivery is late. It answered from the document, on the phone, with nothing sent anywhere.
Free on the App Store: apps.apple.com/us/app/incognitive/id6776977391
同一条，中文版
开着飞行模式，没有 Wi-Fi，没有信号。
我把一份 12 页的合同丢给 Incognitive，问它：交货延误谁来赔？它直接从文件里找到答案，全程在手机上完成，什么都没上传。
App Store 免费下载：apps.apple.com/us/app/incognitive/id6776977391
The samples describe a test I would record before posting; the contract details would match the recording.
08
First 90 days
Days 1 to 30
Use every product daily; write a fact card for each, checked with its builder.
Campaign links on every App Store and signup CTA, one per channel.
First demo clip for Incognitive, LifeClaw and Little Pink Potato.
Weekly feedback digest to product starts.
Days 31 to 60
Two demos a week across the three lead products, EN and 中文.
Reply programme under the AI creators these users follow.
LifeClaw MCP recipes for Claude Code, Codex and OpenClaw users.
A YouTube playlist that holds every demo.
Days 61 to 90
Five to ten AI creators trying the products, with honest briefs and disclosure.
A launch kit used for the next release.
Keep the formats that beat the account median; drop the rest.
Scoreboard
Measured from
Why it matters
Installs and signups per channel
App Store campaign links; signup UTMs; LifeClaw MCP connections
Reach that does not turn into users is noise.
Product-account followers
Weekly count for each account
The base each product launches into next time.
Views per post against the account median
X and YouTube analytics
Shows which formats to repeat.
Written reviews and ratings
App Store, per storefront
Social proof on the page where people decide to install.
Feedback items closed
Weekly digest, tracked with product
Users who see their request shipped become the community.
Any content that mentions tokens or rewards gets a disclosure and a compliance check for the markets it reaches. The AI product content itself does not need token talk.
09
Method & sources
Outside-in, from public data only (public job posting, 2026). Read on 20 Oct 2026. Every figure should be checked internally before it drives a plan.
Data: api.fxtwitter.com (X timelines and profiles) · iTunes lookup and review feeds · data.prompt20.com · Polymarket Gamma API
Peers: OpenClaw · openclaw repo · 0G, Kite, Virtuals, MyShell, NEAR AI profiles on X
Independent homework by Edward Tay for the Manta Network AI Product KOL / Growth Evangelist application. Public data only, no confidential Manta information. The demo is my own prototype and not Manta's code.