Alexander Van Le Turned a Viral Demo into AI Flow Chat

From a 90 Posts a Day Demo to a Two Week Launch

On April 4, 2025, Alexander Van Le posted a demo on X of a tool that automatically produces SEO articles. He said it generated about 90 posts a day and could be pointed at other topics by changing settings. The screen showed a diagram connecting several tasks, and the connected tasks produced articles one after another. The post drew about 3.2 million views, and the experiment he had built for himself led to a business called AI Flow Chat.

He already had a failed startup behind him. A fintech company with four co-founders had raised investment but failed to win customers and generate revenue. In hindsight, he said he spent a great deal of time on investment terms and pitch material while not focusing enough on understanding customers and selling the product.

This time he focused on turning the demo that drew attention into a service people could actually use. He devoted himself to development starting April 7, 2025, and publicly launched AI Flow Chat about two weeks later, on April 21. He built and ran the early product by himself.

A Workspace That Connects Repeated Work

AI Flow Chat was a tool for placing and connecting prompts that had been typed over and over on a screen. Article generation was handed to external AI models, and the user organized materials, instructions, and results into small task boxes. The output of one task could be sent as the input to the next, or the same material could be split across several tasks. It was a way of expressing with connecting lines the process of copying a result and pasting it into another chat window.

The writing workflow makes the usage clear. After adding a title and reference material, a user could split the work into outline writing, setting an editorial direction, drafting the body, and correcting the style. In the usage example Van Le published, different models were assigned to the drafting stage and the final correction stage. A user could revise the instructions for each stage separately and run the whole procedure again.

Having to design every connection from scratch could make a user feel burdened before even starting. On April 27, right after launch, Van Le offered a copyable SEO writing template. By letting people take a finished example and adapt it to their own topic, he reduced the effort of creating the first workflow.

A Quiet Launch, Then Sharing and Embedding

The popularity of the demo did not carry over to the launched service as it was. He introduced the product to acquaintances and asked the people who had reacted to the viral post for feedback, but many did not reply. A promotional post on Reddit brought visitors, but after a single weekend the growth in visits flattened.

One thing he fixed was the first sentence on the homepage. It initially led with time savings, "three hours of AI work in three minutes," but he later changed it to an explanation about bundling repeated AI work into your own tool. According to his own count, with page views at a similar level, daily signups rose from one or two to ten or fifteen. It was the result of making the explanation concrete so the homepage carried the expectation created by the promotional post.

A feature for passing a finished workflow to other people was added as well. On May 27 he introduced a feature for bundling connected tasks into a single app to share, and on June 18 he made it possible to embed that app in an external website. Users could receive a result by entering only what was needed in an input field, without handling the complex task diagram themselves. Over that period the user count Van Le published rose from 58 to 255.

This feature also gave AI Flow Chat customers a chance to build small products of their own. A marketing agency, for example, could assemble a workflow that takes a website address and suggests improvements, then offer it as a free diagnostic tool on its own homepage. Visitors could get a result without an AI Flow Chat account, and the agency could create a point of contact with potential customers through a useful tool.

Adding Document-Centered Chat and Reworking Billing

In July 2025 a feature for chatting freely over materials was added. In a July 29 update, Van Le announced 470 users and introduced chat task boxes, support for PDF, Word, and text files, and a feature for pulling captions from TikTok videos. A user could connect reference material to a conversation and refine the result by asking questions or changing the style. Automation that runs in a fixed order and work that explores a direction while looking at materials came into a single screen.

This document-centered screen had concrete use for video creators. They could gather and connect reference videos and product descriptions, analyze the structure of an introduction or the way explanations were made, and write a script for their own video. A blog operator could keep existing articles as style references, add material on a new topic, and produce a draft. Because what was referenced and which instructions were used stayed on the screen, the setup could be used again for the next piece of content.

While adding features, Van Le also had to find a billing design that could cover AI usage costs. At first he allowed models to run without limits, and in a May announcement he said he was uneasy about promoting the product for fear someone would use it excessively and rack up a large bill. No such incident actually occurred, but he decided to put a daily usage limit on expensive models. Bringing in more customers and designing the usage allowed per customer had to be solved together.

The first approach divided models into three tiers and assigned usage separately per tier. But when one tier's allowance ran out, the leftover allowance in another tier could not be used instead, and when a much more expensive new model appeared the existing classification no longer fit. On July 29 Van Le announced a switch to a single credit system. By giving customers common credits and setting different deduction amounts per model, he adjusted the customer's choice and the service's cost control together.

Revenue Short of Expectations and a Way to Find Customers

Even as users grew, revenue did not rise as fast as expected. In August 2025, Van Le disclosed on Reddit that monthly recurring revenue was 270 dollars, and wrote that by then he had expected it to reach at least 2,000 dollars. He said there were times he felt like he was carrying unrealistic expectations while continuing to develop on a small revenue, but he made clear he would keep the business going.

The customer acquisition method he proposed at that time was concrete. It was to find people complaining about a problem in posts that mentioned competing products and to contact them directly, starting from the problem they were experiencing. He thought it mattered more to find where actual users were than to join posts where founders leave links to each other's services. He advised that after the first payment came in, a founder should check whether the same method could win the next customer too.

Afterward he met customers by publishing usage guides and videos on Discord and Reddit. Rather than repeating the product name, he provided material people could put to use right away, and people who saw it sent him messages directly. Van Le explained that customers who reached out first this way were more likely to convert into purchases, and that they actively told him about problems or improvements while using the product.

In 2026, Putting Reuse at the Center of the Product

In an interview published on February 18, 2026, he said AI Flow Chat, together with Starpop, which he ran separately, was generating about 20,000 dollars in monthly recurring revenue. The main AI Flow Chat customers he named were influencers and marketers who write video scripts and ad copy. A product first introduced as a general tool for connecting repeated prompts was being used in the work of people who organize reference material and produce content.

In 2026 the features that keep the same material across tasks were reinforced. Reusable groups, introduced on August 5, let users make a bundle of a brand's tone of voice, customer information, product descriptions, and research material, then connect it for use across multiple workspaces. Editing the original also reflected the change in connected workspaces. It was a feature that reduced the effort of copying the same explanation for each piece of content and editing each one.

On August 24, multi-turn conversations could be replayed as well. The instructions a user had saved ran in order in a new conversation, and the answer newly generated at each step was carried into the next step as context. Materials connected earlier or the result of another task could also be used together. It became possible to apply a work order discovered through conversation directly to the next set of materials.

As of September 2026, the monthly plans were 29 dollars for the basic tier, 59 dollars for the middle tier, and 119 dollars for the high-usage plan. Each plan included 8,000, 18,000, and 40,000 credits per month respectively, and extra credits could be purchased. Higher plans offered conditions such as more scheduled runs and credit rollover. It was a structure in which spending rises with the usage of customers who keep producing content.

What stands out about AI Flow Chat is that the workflow a customer has built stays inside the product. Once reference material and brand guidelines are organized and the order of instructions that produce good results is set up, the next job only requires changing the parts that are needed. Van Le layered material organization, task connection, sharing, and repeated execution on top of the function that calls external models, building a work environment customers could keep using.