Building the Contact Data Layer for AI Agents - CEO @ https://nitter.cf/t.co/EAD7z06Hih - a 2-man startup. https://nitter.cf/t.co/ASqUEAqEMS Recovering cold emailer. 🇺🇸→🇮🇱
Tel Aviv
Joined September 2024
- Tweets476
- Following606
- Followers645
- Likes4.9K
Pinned Tweet
Your AI agent can get you customers now.
SupersonIQ runs over MCP + API.
Tell Claude to find the contact, verify the email, pull a direct mobile, and it just does it. No exports. No tab-switching.
Clean data, on demand, inside your stack.
Someone asked whether they could skip us and just point Claude at the government registries directly.
Honestly? For one contact, yes. Go do it.
For 5,000, across forty states with forty different formats, deduped, enriched and verified, that's not a prompt, that's a pipeline. We spent a year on it.
The reason to pay us isn't that it's impossible. It's that it's boring.
Something Americans get wrong about Israeli directness: they think it's aggression.
It isn't. It's efficiency.
When someone tells you your product is bad, they're spending their own social capital to give you information for free. The polite version costs you three weeks and teaches you nothing.
I'd rather be told I'm wrong on Sunday than be flattered until Thursday.
Every American founder here has the same moment.
You send a long, carefully worded email. Three paragraphs of context, a soft ask, a "no rush."
You get back: "ok"
You spend an hour wondering if you offended them.
You did not offend them. That was a yes.
Your ICP files permits. Renews licenses. Registers with trade boards. Takes out loans.
He generates a paper trail constantly, in public, in databases anyone can access.
He just doesn't update a profile page, so an entire industry has decided he's unreachable.
Embermail had a product people liked and a wall I couldn't climb.
Building warm-up infrastructure at real volume is genuinely hard. Not "we need to hire someone" hard. Architecturally hard, with the resources we had.
I could have raised to solve it. I could have ground at it for another year.
What I did instead was notice that every customer who signed up asked me the same question within two weeks, and it had nothing to do with the thing I'd built.
The workflow I actually use:
I describe a customer in a sentence. The model turns that into filters I wouldn't have thought to set. It pulls the contacts, verifies them, and hands me a list.
The part people focus on is that it's fast. The part that matters is that I never opened a spreadsheet, and neither did the intern who'd normally be cleaning it.
“START THE CELEBRATIONS!!!!
After a period of reflection “The Most Damaging Pop Cultural Force In Recent British History” has declared itself fit for purpose and will once again return to lift the spirits of the people.
Come witness the fitness.
It shall be a sight to behold.”
Oasis will tour the UK, mainland Europe, USA and Ireland in 2027!
For your chance to buy tickets you must register by 17th Sept, 4pm BST.
👉oasis.hq.fan/registration
There will be no general onsale.
🚩Please note that mailing list members are still required to register.
#OasisLive27
Ben Offerman retweeted
The prevailing view of so much of the young left in America is that we’re the worst country ever, so if someone attacks us, they can’t be all bad.
We 3x'd the database and announced it and almost nobody cared.
Which was the right response, honestly.
Nobody wants more data. They want the 200 contacts they pulled this morning to actually be real people with actual inboxes.
We were selling size to a market that wanted accuracy. Took a public flop to see it.
There is a plumbing company in your city doing eight figures whose owner has never received a cold email.
Not because he's unreachable. He has a phone, an email, a business license, and a truck with his name on it.
He's just not on LinkedIn, so he doesn't exist to the tools everyone uses.
More than twice as long as they clapped for Stalin in Gulag Archipelago!
The Gaza documentary “NAZA” earned a massive 24.5-minute #VeniceFilmFestival ovation as the crowd cheered and waved Palestinian flags.
This marks the longest-ever standing ovation at Venice, beating the 22-minute ovation for “The Voice of Hind Rajab” last year, and perhaps the longest at any film festival, period.
The new documentary was shot at night on the rooftops of Tel Aviv and shows “in detail the systems behind Israel’s calculated mass killing of Palestinian civilians in Gaza,” per an official synopsis. Abraham and Szor interviewed 24 Israeli military intelligence officers and soldiers — all of whom spoke on the condition of anonymity — who were involved in the surveillance or remote killing of Palestinian civilians in Gaza. Their appearance and voices were digitally altered to disguise their identities.
variety.com/2026/film/festiv…
Most people don't need 50,000 leads. They need 200 that are real.
The operators quietly running small verified lists outperform the volume guys most months and never post about it, because "I sent 400 emails and closed two deals" doesn't go viral.
Got asked this twice this week: "if it works through Claude, why do I need you?"
Fair question. Claude is very good at reasoning and has no idea what Dave's email address is.
Ask an LLM for a contact and it produces the most statistically likely string of characters. That looks exactly like an email. Sometimes it even is one.
There's no mail server involved. Nothing got checked. It's a very confident guess.
We're the part that knows.
The hardest part of a two-person company isn't the work.
It's that there's nobody to blame and nobody to hide behind. If something didn't ship, it's one of two people, and you both know which one.
That's uncomfortable and it's also why things move.
Every company announcing that it uses AI in 2026 is announcing that its business model can be copied by a competent team in a quarter.
It's an anti-moat.
The actual moat is knowing something specific about a market that nobody else could be bothered to go learn.
OpenAI published numbers on their research intern agents this weekend.
About 3.1 agent-workdays for every human day. Median spend north of $600 a day on the agent side. P90 past $7k.
And more than half the successful 4–8 hour jobs still needed at least one human intervention.
Same rule I want on outbound. Somebody has to green-light the person before the first send.
Who owns that click on your team this week?
Most of our revenue this year has come from custom data work, not the product.
That's not the plan and I'm not pretending it is. The SaaS is the long game and the custom projects are what pays while we build it.
I'd rather say that plainly than post a chart that implies otherwise.
The reason "just ask AI for the leads" fails is the same reason Apollo's verification fails.
Both are pattern prediction. Apollo sees two addresses at a company and assumes the format. An LLM sees a name and a domain and produces the most probable combination.
Neither one pinged anything. Neither one knows.
Different technology, identical failure mode, same bounce rate.