Built With AI, No Users? Start With 20 Named Accounts

Last week an indie founder put it like this, building two products with AI: "I can build the product with AI. Getting the first users is harder."

He's right, and he's not alone. With Brainy and MoneyBrain, he writes, "getting the first users and growing monthly recurring revenue has been harder than building the software." The failure mode he names is the one I see every week: "A post about my AI workflow and a post explaining why someone should use an application ask for different kinds of attention." His LinkedIn feed loved the build. It did not buy the product.

I'm Tarun, founder of Mantis. This is the loop I'd replace "post more" with: one line of ICP, 20 named accounts that fit it, one visible signal per account, one specific opener. Recruit the first users manually, the way Paul Graham described it twenty years ago: "the most common unscalable thing founders have to do at the start is to recruit users manually."

The LinkedIn-feed loop rewards the people who already follow you for the thing you're already good at. It gave this founder a crowd watching him build. It did not give him buyers. His own line — "where to reach people who need the product, and why they would try it, return, and eventually pay" — is still an open question in his post, and that's not his failure. That's the format's.

The 2026 version of this pain is worse than it was. "Getting first 100 users for your AI Startup is not easy, it's a battlefield," one YC-adjacent playbook puts it. "Market is too flooded. Everyone's selling the next big AI SaaS." More posts. More AI tool launches per week. The noise floor rose, and feed-based "lazy content" rose with it.

Replacing it costs nothing — except specificity. That's the trade.

Before any account goes on the list, write three lines.

A worked example from our own guide reads: "companies with roughly 20 to 200 staff that are growing fast enough to outgrow spreadsheets and an informal HR function." That is specific enough that you can tell, in ten seconds, whether a name on your list belongs on it or not. That's the bar.

Account-tiering is older than the AI-MVP crowd. A clear primer puts it this way: "Tier 1 accounts are your best-fit, highest-value targets, usually 5-15% of a list." The research budget by tier is the giveaway: "Tier 1: 20-30 minutes per account… Tier 2: 5-10 minutes per account… Tier 3: 1-3 minutes per account, largely automated." And explicitly on the number: "Some teams add a tier 0 for named strategic accounts, a small handful, sometimes fewer than 20."

Twenty is not arbitrary. Twenty is what one founder can carry in their head for a week — read the site, read the LinkedIn, read the last release notes, write the opener, send it, remember what happened. A hundred collapses into a mail-merge within three days. Thirty well-chosen leads a week is "a realistic pace for one person doing it part-time," I've written before. Twenty is tighter than that on purpose.

I keep it in a table. Nothing else.

Twenty rows. One page. Three hours the first time, under an hour by week three.

The move most AI-MVP founders skip is the signal column. "The feature I built works best on X" is a feature, not a signal. "They posted a job for a sales intern and their careers page still shows 2024" is a signal. "They complained about Tool Y twice in a review" is a signal. If you can't put a signal in that column, the row is not ready.

We built Mantis to populate the first three columns when the signal exists in public — map listings, open web, hiring pages, review platforms, public registries. It's "an Apollo alternative for leads Apollo was never built to see", because bought databases miss the businesses the signal actually points to — in India, for example, 41.2% of Indian businesses checked have no website at all. But the signal column is yours to write. Nobody else will do it for a company you don't know yet.

For why self-built lists beat bought ones even when you ignore price, see bought database vs live search — "a self-built list tends to be better than a bought one, for a reason that has nothing to do with price: you know why every business is on it."

The opener is not a sales email. It's the first turn of a conversation a stranger did not know you were waiting to have. Three rules:

What opener-writing is not: GPT dropping your product name into a Loom template. The opener lives or dies on the signal, and the signal is proof you read their business. Any shortcut on that side of the equation shows on the other side as "no reply."

This is one week. It replaces every "post more on LinkedIn" cycle you were running. The only thing it needs you to defend is the signal column. Everything else is just doing it.

Paul Graham again, because he was right the first time: "A lot of would-be founders believe that startups either take off or don't. You build something, make it available, and if you've made a better mousetrap, people beat a path to your door." They don't. "Startups take off because the founders make them take off."

Describe your buyer in one line — "funded SaaS startups in Bangalore hiring their first SDR", or "MSMEs in Ludhiana running on spreadsheets" — and Mantis pulls the named accounts, the visible signal and a reachable person from map listings, the open web and public registries. We won't write your opener. We'll give you the twenty names and the reason each is on the list.

500 free credits once at signup, no card.