Jon Yongfook: AI is Consuming Entire Categories of Software as a Service
Independent developer Jon Yongfook stated that opportunities for making money as an independent developer seem to be rapidly decreasing after 2026. He wrote that AI is consuming entire categories of software as a service and listed what he would do if starting from scratch.
The first point is software primitives. Tools like email newsletters, forms, and data analysis that have existed for 20 years will not disappear; humans need them to complete work, and AI must also call upon them because the models themselves cannot perform the native functions of the tools. The problem is that the market is already saturated, and new traction may only remain in niches.
The second point is to productize professional services. He believes this is the best window: taking areas where one excels and that previously required one-on-one delivery, and turning them into agents using proprietary data and skill documentation. The third point is AI value arbitrage, packaging capabilities seen on the cutting edge of X for outsiders. A recent example he cited is Opus 5.5 dynamic video. Insiders know it can be done, and outsiders may still pay for it, but this relies on a knowledge gap, which will quickly narrow.
The fourth point is summarized as "just do hard things." Create things that only venture capital firms could previously do; the barriers are not only technical but also market entry and regulation. This does not guarantee success, but it keeps competition out. The fifth point is to "create a social media scheduling tool," which he immediately declared as a joke.
His public profile shows that his self-bootstrapped software business has a monthly recurring revenue of $81,000. In another post on the same day, he discussed using AI to generate a product that allows you and your friends to star in a movie, which could work and has an entertaining storyline, but the unit economics are too harsh for independent developers to pursue.
This is a judgment on the path of practitioners, not a product release. The buyers are non-AI users still paying for the knowledge gap, while the sellers are independent developers repackaging cutting-edge capabilities. Funds are shifting from general subscriptions replaced by models to niche tools from 20-year categories, professional service agents, and difficult projects with regulatory barriers. Those who hold proprietary data and delivery experience benefit, while general software that can be directly replaced by dialogue faces pressure. The speed at which the knowledge gap closes determines the arbitrage window.
Source: Public Information
ABAB AI Insight
Jon Yongfook Cockle has served as the head of digital and product design for Aviva in Asia and has worked in brand digital marketing at Dentsu and a Japanese design firm. In 2018, he resigned from his position in Singapore and in 2019 publicly embarked on "12 startups in 12 months," launching digital goods open-source store Zipsell, app store asset tool Promomatic, and wireframing tool Montage. Among these, PreviewMojo ranked second on Product Hunt for the day, with a monthly recurring revenue of about $400. He stripped down the interface and changed it to an integration with Zapier and Airtable, receiving the first subscription of $49 within 7 days; this integration later became Bannerbear.
The funding did not come from venture capital. Bannerbear charges marketing teams for automatic image and video generation, with public reports indicating annual revenue exceeding $900,000, while the team retained only two customer service representatives. His current profile marks his self-bootstrapped business at a monthly recurring revenue of $81,000. The path is to first solve repetitive banner tasks for himself, then sell the same action to others with similar repetitive tasks. The proprietary assets are not models, but templates, rendering interfaces, and customer workflows.
In contrast are Basecamp and ConvertKit. Basecamp survived multiple rounds of general software wars through project collaboration, while ConvertKit harvested email newsletters for the creator niche instead of creating another general marketing cloud. The scheduling tools occupied by Buffer and Hootsuite are precisely the type he jokingly crossed out. Independent development is currently at a transitional stage after contraction: general functions have been consumed by models, leaving behind 20-year primitives, one-on-one service agentification, and hard projects that emerge simultaneously with regulation and market entry.
This is technological substitution. What is being replaced are replicable software interfaces, not the work clients need to complete. The mechanism is that models first take away generation and operation, leaving tools only when the model cannot natively complete tasks, such as delivery, form submission, and measurement. Knowledge arbitrage depreciates according to the speed of information dissemination, while hard tasks keep competition out through regulation and sales cycles. Pricing power shifts from feature lists to proprietary data and non-skippable workflows.
ABAB News · Cognitive Laws
- Models consume functions but cannot eliminate the work that needs to be delivered.
- Knowledge gaps can be arbitraged, with dissemination speed determining expiration dates.
- Hard tasks block competition but do not guarantee payment.