Alexander Van Le Grew AI Flow Chat and Starpop

From a Failed Fintech to Small Products

Alexander Van Le co-founded a venture-backed fintech company and it failed. There were four co-founders, but the company had no revenue when it took its early investment. The company that drew investor interest never turned into a business that collected money from customers.

A founding retrospective he wrote in February 2024 still carries that experience. He warned against taking investment before a workable prototype and a customer were settled, and he also criticized founders who let their job title limit what they do. His idea was to learn unfamiliar things as they come up and to put improving the product or getting it to customers first. His principle was to spend time on activity that shows real reaction rather than on material explaining a long-term strategy.

His list of personal projects carries traces of several attempts. In 2023 he built SeekJob, which manages a job seeker's application status, and in 2024 he built JustDump, a note app for recording thoughts immediately, and DanishTech, a Danish developer hiring site. Hiring listings, notes, and job-search management each became a product of its own from small problems he could explain from daily life.

The turning point was an AI automation experiment he published on X in April 2025. He described connecting several tasks to generate about 90 posts a day, and that post drew about 3.2 million views. That response led him to quit his job and commercialize AI Flow Chat.

An AI Workspace You Can Run Again

AI Flow Chat began with the repeated work inside using AI. He had to keep typing a question, copying the answer, pasting it into another window, and polishing it, and a person also had to remember the prompts and model combinations that had produced good results. Van Le wanted to lay that procedure out on a screen, connect the parts, and run it again when he needed it.

Building and shipping the first version took two weeks. The product became a visual workspace where users place and connect materials and instructions. A user could add documents, websites, videos, and other references and design a job that produces an article or a video script from them. Even after getting one result, the setup of the job stayed behind and could be used again for the next piece of content.

The user count he published in a product announcement on May 27, 2025 was 58. The same announcement introduced a feature for sharing a workflow he had built as an app other people could run. Once a user finished a complex workflow, it could be handed to a colleague or acquaintance in a runnable form. It was a feature that let other people use an automation procedure one person had made.

A scheduled-run feature arrived in July. Combined with a step that pulls material from websites, it could gather new information every day and produce content, and Van Le explained that he used it for a daily news summary. Scheduled runs were offered only on paid plans. It was a structure that charged a subscription for automation work running repeatedly.

Controlling Usage and Reworking Billing

While promoting the service, the pricing design changed too. At first he allowed every model to run without limits, and Van Le said he was even uneasy about promoting it, fearing that a malicious user would run up enormous AI costs. In May 2025 he introduced a daily limit on usage of expensive models. To take on more customers, he first had to control what a single person could cost.

On July 29 he replaced the three-tier model allowance with a unified credit system. Under the old method, a user who exhausted one tier could not draw on unused allowance from another. As more expensive models arrived, costs could also diverge within a tier. The new method gave subscribers credits each month and deducted different amounts for different models. This reduced the need to redesign every plan when adding a model and let users spend their credits on the models they wanted.

The free offering was reset as well. In an October 2025 announcement he explained that abuse had occurred on the free plan that could be used without confirming a payment method, so it was replaced with a 14-day free trial. Customers who used up their credits were allowed to buy more, and separately purchased credits carried over into the next month. Monthly subscription became the default, and demand from heavier users was met through additional payments.

Starpop, Focused on Ad Creation

The second main product, Starpop, was built with co-founder David Ishag. The product helps ecommerce brands and ad agencies create ad images and videos with AI. A customer can pick an ad template, add product material, and start production. Compared with AI Flow Chat, where users arrange the work order themselves, it was a product that narrowed the purpose to product advertising from the start.

When building Starpop, he looked at similar services that were already selling and settled on a customer and a problem to solve. The first version shipped in a week, and the first subscription sold in the same week. The gap between putting a purchasable product out and confirming payment response was short.

Code from the earlier product helped cut development time. Van Le started Starpop from AI Flow Chat's code and used the same file structure and libraries in both repositories. A bug found on one side could be fixed on the other right away. The burden of re-learning unfamiliar code every time he moved between products also shrank.

Growing Two Products by Meeting Customers

Early sales for both products began by messaging potential customers directly. Later he posted useful explanations and videos on Discord and Reddit to draw inquiries. He kept product pitches to a minimum and used content that showed examples helpful for actual work.

A demo of the image generation canvas posted on Reddit drew a comment that the product's purpose was hard to see. Van Le offered blog posts with a featured image and fairy tale production with illustrations as use cases. To people asking whether an AI service had to be connected separately, he answered that they could use it right after signing up. Even after publishing a demo, there remained the work of explaining what it was for and how to start.

He explained that during signup he also asks where people heard about the product and what they will use it for. The purpose was to get information to consult when deciding the next feature. Because acquisition path and intended use were collected together, it became material for seeing which promotion brought in users for which purpose.

In an Indie Hackers interview on February 18, 2026, he said the two products' combined monthly recurring revenue was about $20,000. As the next goal he named $100,000 in monthly recurring revenue, saying he wanted the room to try a larger business.

Improvements to AI Flow Chat continued afterward, centered on reuse. In August 2026 a feature was added that bundles a brand's tone of voice, product information, and research material so several workspaces can share them. Editing the original also reflected the change in connected workspaces. It pointed toward reducing the effort of customers entering and editing the same background explanation every time.

In his 2024 retrospective, Van Le wrote about a proposal to split a company with no revenue into two products. He worried that founders divided by role would lose sight of the same product and sales problems. His later work on two products suggests that what matters is whether experience from one attempt carries into the next. He sold products with distinct uses while reusing familiar code and ways of reaching customers.