AI Agents Engineering: AI-агенти на Google ADK 2.0 + Go SDK ➡️
Наталя ХандусенкоAI Eng
8 September 2026, 17:32
2026-09-08
A Ukrainian IT guy created an interactive reading app called GrammaLens that explains English grammar right in context
Have you been studying English grammar for a long time, but still get confused when reading real texts? Specifically to solve this problem, Full Stack Engineer and co-founder of Azon5 Andriy Tretyak created the GrammaLens service, where you can insert text and get an explanation of the rules used in it. In addition, the developer shared the technical details of the project, for which he did not write the code himself, but only set tasks for AI agents.
Have you been studying English grammar for a long time, but still get confused when reading real texts? Specifically to solve this problem, Full Stack Engineer and co-founder of Azon5 Andriy Tretyak created the GrammaLens service, where you can insert text and get an explanation of the rules used in it. In addition, the developer shared the technical details of the project, for which he did not write the code himself, but only set tasks for AI agents.
"I studied English for many years, and as it turned out, not at all as it should be. I dutifully read the rules, did the exercises, nodded my head: well, it's clear. And then I see the same construction in a real text and don't recognize it. It's not that I don't know, it's just that my brain doesn't "highlight" it. For a long time I thought that I lacked practice. But in fact, everything is simpler and sadder: the rule exists by itself, the text by itself, and there is nothing between them," Andriy Tretyak shared on Reddit.
From static rules to contextual analysis
At first, the developer created a website with simple explanations in Ukrainian, but quickly realized that this was the same trap. Then the idea of an interactive lighting system arose.
Now, GrammaLens users can open any story or insert their own text, where the service immediately highlights grammatical constructions (for example, Past Simple or Present Perfect) and provides clear explanations. In addition, on the site, you can click on any unfamiliar word to add it to your personal dictionary and get an analysis of typical mistakes that Ukrainians often make (such as the incorrect use of "I have seen him yesterday").
The developer prepares and proofreads the rules and notes on the site independently as he learns, noting that the project is primarily a personal system summary, not a classic textbook.
A product created by autonomous AI agents
The most interesting technical feature of the project is its development process: Andriy Tretyak did not write a single line of code by hand. The entire product was implemented from start to finish by AI agents.
The project's technology stack consists of standard solutions: Next.js, Postgres via Drizzle, Auth.js, Anthropic API, PDF recognition modules, and deployment in Cloud Run.
To manage the process, the developer created his own wrapper on top of Claude Code called Swarmery , which he uses to collect ideas and generate a clear plan.
“I didn’t write the code by hand. Not at all. I have my own system of agents on top of Claude Code, and they made this entire product, not me… It all starts with an idea, not a task: I just describe what I want to achieve in a few sentences. Then the agents help me break this idea down into a clear plan, they specify the details, divide it into stages, and prescribe readiness criteria for each. The entire plan is stored in a regular text file,” the engineer said.
After the developer approves the plan, the agents autonomously perform the tasks: each stage is implemented in a separate branch, another agent checks the result against the specified criteria, and only after that does the team move on to the next step.
Key takeaways from AI development
Andriy Tretyak emphasized that the main difficulty was not in the ability of AI to generate code, but in the ability to correctly set tasks and determine clear readiness criteria.
“I was sure that there would be problems with the way the agent writes the code. In fact, everything is fine with the code, the complexity is not there. The hardest thing was learning how to set tasks… Everything changed when I started formulating each stage like in a normal team: what needs to be done, how exactly, and, most importantly, what should be considered ready. During this time, I realized the main thing: the quality of the final result almost entirely depends on the quality of the plan, and not on how cool the agent writes the code,” the co-founder developer concluded.
“The hardest part was waiting 10–15 minutes for Claude to implement everything.” The 10-year-old vibecoder, together with her dad, created the game “The Adventures of the Seal Family” over the weekend, which helps teach English and more. How it was
Як нейромережі бачать вільну та незалежну Україну? Тест dev.ua
Нейронні мережі для генерації зображень бачать світ по-своєму, їхню логіку зрозуміти часом зовсім неможливо. Але таки хочеться. На честь Дня Незалежності України редакція dev.ua вирішила провести невеликий експеримент.
Ми задали чотирьом різним нейронним мережам п’ять однакових запитів: «прапор України», «День Незалежності України», «український Крим», «перемога України» та «українці». Отриманими результатами ми ділимося з вами нижче.
У TikTok тепер можна генерувати фон за допомогою нейромережі. Ми протестували її та ділимося результатами
У TikTok з’явилася нова функція «Розумний фон». З її допомогою як фон для тіктоків можна підставляти згенеровані нейромережею зображення. Редакція dev.ua протестувала цю технологію і ділиться своїми враженнями.
