Former Anthropic dev Thariq calls out the manic loop of AI agents
You have multiple AI agents running, prompts to test, code to review and tasks being completed in the background. It feels like a productive day. But how much meaningful work is actually getting done?
That is the concern raised by former Anthropic engineer Thariq Shihipar, who has questioned the growing obsession with AI agents. In a post on X, Shihipar said coding agents can create an intense feeling of activity, where people feel constantly occupied even when little meaningful progress is being made.
His observation points to a growing tension around AI at work.
While these tools promise to speed up tasks, using them can sometimes become an activity in itself. Let’s break down why feeling busy with AI agents doesn’t always mean you’re moving the needle.
When AI activity gets mistaken for progress
AI agents are designed to perform tasks with some degree of autonomy. They can break prompts into smaller steps and carry out work such as writing code, researching information and testing ideas. Developers increasingly use them to handle repetitive or time-consuming tasks.
There’s 100% a kind of mania you can get into with coding agents, where it feels like you’re getting a lot done and working all the time but nothing is really happening.
I think it’s often a form of avoidance of something else in life.
— Thariq (@trq212) October 29, 2026
Thariq Shihipar, who previously worked on Claude Code, compared the rush of using coding agents to a manic loop. People may keep prompting, testing, rebuilding and experimenting, feeling productive throughout, without necessarily moving closer to a meaningful outcome.
He also suggested that constant interaction with agents can become a form of avoidance. Instead of deciding what needs to be built or fixed, users may find themselves occupied with getting AI to do more.
The hidden work behind AI productivity
A similar concern is emerging across workplaces, where employees are spending considerable time managing AI-generated output.
Moreover, a YourStory report on AI ‘botsitting’ highlighted how workers spend time supplying context, correcting responses, checking facts and fixing AI-generated work before it can be used.
Citing Glean’s Work AI Index, the report noted that digital workers spend 6.4 hours a week on botsitting. About 37% of the time employees spend with AI goes towards supervising it, slightly more than the 36% spent using AI to produce actual work.
The findings point to a gap between adopting AI and getting measurable productivity gains. Although 75% of surveyed digital workers said AI made them more productive, only 13% said it had significantly improved their organisation’s performance.
More AI does not always mean more output
Thariq Shihipar is not alone in exploring the intensity of agent-based coding. AI researcher Andrej Karpathy, who popularised the term vibe coding, has spoken about pushing AI coding tools to their limits. OpenAI CEO Sam Altman has also joked publicly about changes to his sleep patterns because of advances in coding tools.
The answer is not to stop using AI agents. It is to be more deliberate about what they help accomplish. Whether in software development or everyday office work, the value of AI should be measured by the work it helps complete, not the number of prompts, agents or hours spent interacting with it.
