
Building @glasserai to give AI agents an unfair advantage. 2,000+ data tools, 27 providers. Pay per call, no seats. Before: Lark at TikTok & SureThing Agent
California
Joined December 2008
- Tweets343
- Following452
- Followers671
- Likes206
Your agent is only as good as what you feed it.
Garbage in. Garbage out.
Introducing @Glasserai - it feeds your agent premium data instead.
Ahrefs. Semrush. ZoomInfo. Apollo. PDL. etc.
1,900+ endpoints. One key. Pay per call. No more subscriptions.
Send this one line to your Claude Code, Codex, Grok Bot, or Muse:
set up glasser.ai/SKILL.md
Every time you tell a customer “you can build anything,” you're giving them homework.
When I was working on Lark at ByteDance, we spent a lot of time studying Airtable and monday.com.
One thing that stuck with me was monday.com's GTM approach.
Both offered flexible platforms. But monday went much deeper into packaging and selling specific workflows.
And there's a pretty important economic reason why this works.
When you sell a platform that can do 100 things, you're asking the buyer to figure out which 3 matter, how to configure them, who will own implementation, and how to convince the rest of the team to use it.
You've basically made your customer an unpaid product manager.
That homework shows up in your GTM metrics:
-Longer sales cycles: Every deal requires the buyer to figure out their own use case.
-Higher CAC: Your sales team spends hours explaining possibilities and designing custom solutions.
-Poor activation: Customers sign up excited about what they could build, then never get around to building it.
If I were selling a horizontal AI product today, here's what I'd do:
-Study your 20 best customers. Find the 2–3 workflows they actually use repeatedly.
-Package those workflows into specific offers. Show the input, the output, and the result.
-Build separate landing pages and demos for each. Stop sending everyone to the same generic homepage.
-Get customers to their first useful output before asking them to configure integrations, prompts, or agents.
-Measure conversion, time-to-value, and retention by workflow. Double down on the ones that consistently perform.
One more thing: the use case that gets someone to buy doesn't have to be the one that makes them stay.
You can acquire customers through one narrow workflow, then expand into the broader platform once they've seen value.
That's how I'd think about GTM for most horizontal AI products.
Your product can have 100 use cases. Your customer should only need to understand one to get started.
1
1
85
ChatGPT Ads: 100 signups in a week. Google Ads: 10 signups. Same product.
The lesson I took is about keywords.
A keyword is a category name. Our buyers don’t know our category exists. They never search “data API.” They ask ChatGPT to find 50 decision makers at companies that just raised.
New product, no keyword to buy. The conversation is the only place the demand shows up.
Still watching activation and CAC before I move budget.
Check before you post:
1. Spend: 100 vs. 10 only means something if spend was comparable. If it wasn’t, lead with cost per signup instead.
2. “Signups” vs. “users”: I changed it to signups, since the post later says you can’t yet tell if they activate. Change it back if they did run tasks.
3. Internal data: the 70% figure and the two customer examples come from the heavy-user report. They’re anonymized, but publishing them is your call.
4. Optional caveat: the ad pool is limited to logged-in adults on ChatGPT’s Free and Go plans, so these signups may skew toward people who haven’t paid for AI yet. One line on that in the “attention is not demand” section would make it more honest than most posts on this.
1
2
97
agent-first ≠ chat-first.
a lot of people think the ideal agent experience is just telling AI what you want, and it gets everything done.
but I think we're mixing up two different things: what AI can do, and how people interact with it.
the point of an agent is to get work done in the background, not to make every interaction a conversation.
for example, you can ask an agent to research 100 companies with one prompt. but when the results come back, do you really want to go through them one by one in a chat?
a table where you can sort, filter, and edit is clearly better.
same with design. you need previews and controls. data analysis needs charts and visualizations. complex tasks need progress tracking and checkpoints where you can review what's happening.
a chat window isn't the best interface for all of that.
I think the future of agents needs both GUI and CUI.
natural language is great for expressing intent. GUI is better for exploring information, comparing options, and making precise adjustments.
they're not replacing each other.
agents should do the work in the background. the interface should depend on the task, not the technology.
1
3
85
We could have become Grok bot.
In fact, we were earlier than Grok bot.
More than a year ago we were building SureThing, a general personal agent. Back then we already believed an agent should keep one continuous session and message you first, without waiting for you to open a new chat. Most personal agents work this way today.
We saw it early. We still lost. Here’s what I learned.
1. I brought a big-company playbook to a startup.
I built Lark before this. At a big company you can try five directions at once and pour resources into whichever one shows a signal. At a startup, that’s five half-bets, and none of them gets enough time to prove anything.
2. “General” was a decision we kept avoiding.
We never picked one industry, one scenario, or one group of people to start from. Every time we got close to choosing, it felt like giving up the bigger market.
3. You find out you have no edge when the market gets crowded.
When OpenClaw came out, the space filled up fast. We knew what made us different. From the outside, nobody could see it.
4. Our users had already shown us the answer.
The most-used part of SureThing was the data sources we had integrated. People used them to find customers and audit their ICP. We treated it as one feature among many.
5. Think big, start small.
General can be the mission. The starting point has to be vertical: one industry, one scenario, or one kind of person.
After a lot of conversations with other agent builders in San Francisco, I stopped trying to compete with them. We took the part users loved and opened it up to every agent.
That became Glasser.
Being early is worth very little if nobody knows who you’re for.
3
5
233
hey @X , not new here but never did a proper intro
> 40, SF & Singapore
> 7 years at ByteDance building Lark, the workspace the whole company (yes, TikTok too) runs on;
> 0 to 10M+ users; grew my team from 8 to 400;
> then left to built SureThing, a general AI agent. we shipped before Muse & Instinct. they went viral, we didn't;
> now building @Glasserai: 2,000+ data tools for AI agents, pay per call;
> one rule I live by: it is never wrong to do the right thing;
gonna post it all here: the failures, what i learned, and what i'm building next
if you're building something, or just curious about AI and life, would love to connect
14
14
391
Semrush subscriptions are dead.
My agent just reverse-engineered Manus's entire 2026 Google Ads strategy.
20 countries. One prompt. $0.24.
What it found:
→ Their biggest bet was Brazil, not the US
→ May alone: 513K paid visits
→ The ad that won never said "free." It said "AI that does the work."
One month of Semrush ($99) pays for 412 of these runs.
No seat. No subscription. Pay per call.
Steal my setup: glasser.ai/SKILL.md
Nvidia fell 4% this morning. The last 18 times it dropped 6%+ in a day, it recovered 17. Median wait: 15 trading days. Range: 1 to 115.
I didn’t look this up. My agent did, through @Glasserai: Monday’s minute prices plus 12 months of dailies, 22 calls, $0.82 on the statement. Koyfin Premium is $79 a month.
One prompt in Claude Code and your agent has the same account: glasser.ai
One key. Every data source. Pay per call.
Not financial advice.
had my agent pull 5 brands' tiktok vs meta ads via glasser.ai for $0.07
wendy's on tiktok: roasting people for 37M views
wendy's paid ads: "all roads lead to the $7.99 Dave's Single Combo"
all 5 do this. like your funniest friend at a job interview in a tie
1
4
105
"Your competitor’s 1-star reviews are the cheapest market research you’ll ever buy.
I read 418 of them for $0.036.
Lovable: 15 of the last 27 complaints say the same thing. Keep it working, stop racing to ship. If I were Bolt, tomorrow’s headline is “It keeps working.”
Shopify Bundles: 24 of the last 50 reviews are 1-star. 24 merchants naming your next feature, in public.
Run it on yours:
Paste the setup prompt from glasser.ai into Claude Code or Codex. $1 free, no card.
Paste: “Pull the latest 1-star reviews of [competitor] from [G2 / Trustpilot / Shopify App Store]. Cluster into themes, count each, quote one per theme, tell me which theme I could own in a headline.”
Write the headline.
Your competitor’s customers already wrote it."
Everyone wants to copy Duolingo's owl. You can't, and honestly you shouldn't try.
Every product that blows up has an owl, the part everyone sees and nobody can replicate. Underneath it there's a pile of boring stuff nobody screenshots, like which channel does which job, or the ad they've quietly left running for six months.
That boring pile is where the GTM actually lives, and somebody already paid to test it for you.
2
82
My agent just stole Duolingo's ads strategy for $0.45.
Duolingo's top 10 TikToks 512M views of the owl,
But their ads quietly do the "boring" work.
- 68 of their 80 latest Google ads sell an English test.
- 29 of the top 30 Meta ads send you straight to Google Play.
- Chess got its own ad. 175 days, 13 versions.
Fun part: the CEO said no to chess. It became their fastest-growing course 🙃
If you want this teardown for your competitor, just connect glasser.ai data router to your agent.
2000+ data endpoints. Pay per call.
$0.90 and 1m40s.
That’s what it cost my agent to find the 12 companies that raised this week and are about to hire their first GTM.
- announced funding in the last 7 days
- no sales or marketing lead on the team yet
- job post not up anywhere
Once that post goes live, everyone has the same list. Right now you’re the only one who knows.
One key. Every data source. Pay per call.
1
5
85
Last week when I was doing user research, I asked in the Hermes Agent Discord what data sources people use.
Guy goes: "None bro. Hermes finds everything."
ME: 🙃🙃🙃
Monthly search volume? Backlinks? Some VP's verified work email?
Buddy, none of that is sitting on the open web.
Your agent didn't find it. It made up something that looks exactly like it.
That's not research. That's fan fiction with a confidence score.
1
100
I guess I have no chance convincing them.
So now it's a plugin. If you can't beat the vibes, ship into them.
github.com/NousResearch/herm…
1
1
67
And if you are not using Hermes, try send this to whatever Agent you use:
set up glasser.ai/SKILL.md
17
Steal the prompt. Schedule it daily in any agent (Claude Code, Hermes, whatever you use):
Use Glasser to build today's X content digest.
1. Pull the last 24h of posts from karpathy, swyx, simonw, emollick
2. Search top posts on: AI agents, reasoning models, open-source LLMs
3. Drop retweets and duplicates, rank by engagement, keep the top 10
4. For each: what it says, why it's worth writing about, my angle, link
5. End with 3 headline ideas for today
Swap in your niche and your accounts. Same thing works for crypto, design, marketing.
1
65
Why it's this cheap: no $200/mo X API plan. The agent pays per call through Glasser, $0.001 per request, 8 requests today.
Same key also covers Reddit, TikTok, SEO and lead data, so you can build the same digest for any niche.
glasser.ai
22




