Хочете будувати кар’єру в Data Science та AI? Vodafone Big Data Lab відкриває новий набір —>
Наталя ХандусенкоAI Eng
21 July 2026, 12:06
2026-07-21
“Work on your agent hygiene”: how to stop burning AI limits — advice from an experienced developer
Developers are increasingly complaining about quickly running out of limits and tokens during daily work with AI agents. Perhaps the problem lies not in the strict limitations of the platforms, but in the lack of basic "agent hygiene". Oleksiy Krychevsky, a Software Engineer with over 10 years of experience, shared his observations and advice on how to optimize the work of tools and improve code quality.
Developers are increasingly complaining about quickly running out of limits and tokens during daily work with AI agents. Perhaps the problem lies not in the strict limitations of the platforms, but in the lack of basic "agent hygiene". Oleksiy Krychevsky, a Software Engineer with over 10 years of experience, shared his observations and advice on how to optimize the work of tools and improve code quality.
Why limits are "burning"
Rapid token loss during interaction with agents is almost always a result of setup errors or irrational use of the working environment.
“If you are hitting limits when working with agents, it means you are doing something wrong. And there are a lot of things that can go wrong: from loading hundreds of skills, connecting dozens of MCPs, and so on. Maybe it’s a flaw in your environment, or maybe it’s your project setup,” Oleksiy Krychevsky shares in his LinkedIn post.
According to the IT professional, the main "suckers" are unnecessary settings, an inflated basic context, as well as cluttered project documentation and unoptimized code search.
4 steps to optimizing your agent experience
To stop wasting tokens and budget, Oleksiy advises implementing several practical steps:
Conduct an environment audit: review the application settings or CLI tools you use. You need to analyze the enabled MCPs, skills, and tools and ruthlessly disable anything that is not needed for the current task.
Check the base context: You can ask the agent himself to show you what exactly is included in his base context.
“I say “hi” to the code and get minus 19k tokens. Given that I don’t suffer from limits at all, I see that this figure can be reduced to somewhere around 12k tokens of the base context,” the engineer gives an example.
Use wrapper tools: special wrapper tools help optimize the volume of incoming and outgoing tokens by cutting off excess "water".
Keep track of documentation and logging: it is important to keep project specifications in order, optimize searches for evidence in code and decisions made, and clearly log the work performed.
The main mistake of developers: infinite sessions
Oleksiy Krychevsky calls the lack of session discipline the most common "sin" among those who complain about limits.
"The biggest mistake I see among people writing about their agent work is that they sit in one endless session. People, it doesn't work like that! Follow the rule: one task - one session. And there are also subagents..." - emphasizes the developer.
Instead of a conclusion: it's not just about money
Optimizing your work with artificial intelligence isn't just about saving money or staying within your plan limits. It's primarily about the accuracy and quality of task performance.
“It’s not just about limits or money, it’s also about how clearly and precisely your agent will achieve the set goals. When your context is 80% of your window, the results of your work will be, to put it mildly, not very good. Work on your agent hygiene!” — concludes Oleksiy Krychevsky.
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.
