Former Anthropic engineer Thariq Shihipar says excessive use of AI coding agents may create an illusion of productivity while helping users avoid underlying personal struggles.
Artificial intelligence agents are increasingly becoming part of everyday professional workflows. From OpenAI's Codex to Meta's Muse, AI-powered tools are being used to automate tasks, write code and improve productivity.
However, former Anthropic engineer Thariq Shihipar has raised a different concern about the growing popularity of these tools. He believes that constantly working with AI agents could sometimes become a way for people to avoid dealing with other aspects of their lives.
In a recent post on X, Shihipar discussed the psychological experience of using coding agents continuously, suggesting that the feeling of being productive does not always translate into meaningful results.
In his post, Shihipar described how working with AI coding agents can create an intense feeling of activity and constant engagement.
He suggested that users may feel as though they are accomplishing a significant amount of work because AI agents allow them to move rapidly between tasks. However, this activity does not necessarily mean that meaningful progress is being made.
He explained that developers can enter a state where they feel continuously occupied with coding agents, even when their actual output remains limited.
Shihipar also questioned whether this behaviour could sometimes represent a coping mechanism or a form of avoidance.
According to his observation, people may turn towards AI-powered activities to distract themselves from personal difficulties or other issues they would otherwise need to address.
His comments highlight a possible difference between staying busy with technology and making genuine progress in professional or personal life.
Shihipar further connected his observations to a broader discussion about addiction and emotional suffering. In a comment accompanying his post, he shared another discussion that compared addictive behaviours to anaesthesia used to reduce suffering.
The underlying argument was that addictive habits can provide temporary relief from uncomfortable emotions, while addressing the deeper source of suffering may help people move away from those habits.
By drawing attention to this idea, Shihipar suggested that excessive engagement with AI agents could sometimes serve a similar purpose.
However, his comments represent a personal observation rather than an established scientific finding. They do not suggest that everyone who regularly uses AI agents is experiencing addiction or avoiding personal problems.
Shihipar's comments have attracted attention partly because of his previous work at Anthropic, the company behind the Claude family of AI models.
He was associated with the development of Claude Code, Anthropic's AI-powered coding tool.
He reportedly left Anthropic earlier in October 2026. However, he has not publicly disclosed the reason behind his departure.
His observations come at a time when AI coding agents are becoming increasingly capable and developers are experimenting with new ways to automate software development.
Shihipar is not the only prominent figure in the artificial intelligence industry to describe an intense relationship with AI agents.
Andrej Karpathy, former OpenAI co-founder and the person who popularised the term 'vibe coding', previously described his experience of working with AI agents as being in a state of 'psychosis'.
Speaking on the No Priors podcast, Karpathy explained that he was constantly exploring the possibilities of AI agents and attempting to push their capabilities further.
OpenAI CEO Sam Altman has also openly discussed how AI coding tools have influenced his daily routine.
In April 2026, Altman said he was considering switching to a polyphasic sleep schedule, which involves multiple sleeping periods throughout the day, because he did not want to miss opportunities to work with Codex.
His comments highlighted the excitement surrounding increasingly capable AI tools and the pressure some technology enthusiasts feel to keep experimenting with them.
While AI agents are primarily designed to improve productivity and automate complex tasks, their growing adoption has also encouraged discussions about their possible psychological effects.
AI chatbots have already attracted attention from researchers and mental health experts over concerns about excessive usage and their potential influence on users' thinking.
Some reports have raised concerns about a phenomenon described as 'AI psychosis', in which prolonged or unhealthy interactions with AI systems may reinforce paranoia, delusional beliefs or distorted perceptions of reality.
However, this is not an established clinical diagnosis, and the effects of AI interactions can vary significantly between individuals.
Shihipar's comments add another dimension to this discussion by focusing on productivity-driven engagement with AI agents rather than chatbot conversations alone.
As AI tools become more integrated into professional and personal routines, understanding the difference between useful automation, excessive engagement and genuine productivity will become increasingly important.
Conclusion
AI agents are transforming how people work, but Thariq Shihipar's observations raise questions about the psychological side of constant technology use. While these tools can improve efficiency, excessive engagement may sometimes become a distraction from real-life responsibilities and challenges. Finding a healthy balance between AI-assisted productivity and personal well-being remains important.