Ex-Anthropic Engineer Warns AI Agents May Fuel Avoidance

Artificial Intelligence agents are becoming increasingly common in everyday work, with people using them for tasks ranging from software development and research to planning and problem-solving. Tools such as OpenAI’s Codex and Meta’s Muse are part of a growing ecosystem designed to automate tasks and help users work more efficiently.

However, former Anthropic engineer Thariq Shihipar has raised a different question about the growing dependence on AI agents: could constantly using these tools sometimes be a way of avoiding something happening in a person’s life?

In a post on X, Shihipar described the intense feeling that can come from working with AI coding agents. He compared the experience to a state of mania, in which someone feels continuously occupied and productive but may not achieve meaningful results.

Thariq Shihipar Describes the ‘Mania’ of Using AI Coding Agents

Shihipar said that working with coding agents can create an overwhelming sense of activity. Users may feel as though they are constantly building, experimenting and making progress, even when the actual results do not match the amount of time and effort invested.

“There’s 100 per cent 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,” Shihipar wrote on X.

He also suggested that this pattern could sometimes be connected to avoidance rather than productivity alone.

“I think it’s often a form of avoidance of something else in life,” he added.

His comments raise questions about the relationship between constant digital activity and meaningful work. AI agents can make it easier to start multiple tasks, generate code and explore ideas quickly. But the feeling of doing more does not necessarily mean that a person is making progress towards a specific goal.

Former Engineer Links Addictions to Underlying Suffering

In a comment on the discussion, Shihipar linked another post that compared addiction to anaesthesia for suffering. The post suggested that addictive behaviours can help people temporarily escape emotional discomfort, while addressing the underlying suffering may be necessary to move away from those behaviours.

“Addictions help with suffering because anesthesia helps with suffering,” the post said. “Therefore, the way to unlearn all addictions simultaneously is to remove the underlying suffering.”

By sharing the post, Shihipar connected his observation about AI coding agents with the broader idea that people may sometimes use repetitive or absorbing activities to avoid dealing with difficult experiences.

However, his comments were a personal observation about the way people can engage with AI tools. They do not establish that frequent use of AI agents is an addiction or that people who use them regularly are necessarily avoiding problems in their lives.

Thariq Shihipar Recently Left Anthropic

Shihipar left Anthropic earlier this month. He previously worked on Claude Code, the company’s AI-powered coding tool.

He has not publicly provided a reason for his departure, according to the information in the report. His comments about AI agents have attracted attention because they come from someone who has worked on developing tools that are increasingly being adopted by software developers.

His observations also come at a time when AI agents are becoming a regular part of software development workflows, with developers experimenting with ways to delegate more coding and technical tasks to automated systems.

Andrej Karpathy Also Described an Intense Experience With AI Agents

Shihipar is not the only prominent figure in the artificial intelligence industry to describe an intense experience while working with AI agents.

Andrej Karpathy, the former Tesla AI director who coined the term “vibe coding”, previously spoke about the overwhelming nature of experimenting with AI-powered coding tools.

Speaking on the No Priors podcast earlier this year, Karpathy described his experience as being in a “state of psychosis” while exploring the capabilities of AI agents.

“I’m just like in the state of psychosis of trying to figure out what’s possible, trying to push it to the limit,” he said.

His comments reflected the intensity of experimenting with rapidly developing AI tools and trying to understand how far their capabilities could be pushed.

Sam Altman Says GPT-5.5 in Codex Is Changing His Work Routine

OpenAI CEO Sam Altman has also spoken openly about his use of AI tools and their influence on his daily routine.

In April this year, Altman said he was switching to polyphasic sleep, a pattern involving multiple periods of sleep within a 24-hour period. He linked the decision to his enthusiasm for working with GPT-5.5 in Codex.

“I am switching to polyphasic sleep because GPT-5.5 in codex is so good that I can’t afford to be sleeping for such long stretches and miss out on working,” Altman wrote.

His statement illustrated how strongly some people in the technology industry are engaging with AI tools and how the technology is influencing their approach to work.

Concerns About Excessive AI Use and Mental Health

While concerns about the Mental Health effects of AI agents have not yet reached the same level of public attention, AI chatbots have already raised questions about the potential consequences of excessive use.

Reports have suggested that prolonged or excessive interactions with AI chatbots may, in some cases, be associated with experiences described as “AI psychosis”. These discussions have focused on situations in which users may develop paranoia or delusional beliefs.

Such concerns are different from Shihipar’s observations about coding agents. His comments focus on the possibility that constant engagement with AI tools can create a feeling of productivity while distracting users from other aspects of their lives.

AI agents can be useful for automating work, exploring ideas and supporting complex tasks. However, Shihipar’s observation highlights a distinction between using technology to accomplish something and staying occupied simply because the tools make continuous activity possible.

AI Agents Raise a Bigger Question About Productivity

The growing popularity of AI agents has changed how people approach coding and other knowledge-based work. Tasks that once required significant manual effort can now be delegated to AI systems, allowing users to experiment with more ideas and manage multiple projects.

But Shihipar’s comments point to a possible downside of this constant availability. When AI tools make it easy to keep working without a clear stopping point, users may find themselves caught in a cycle of activity that feels productive but does not necessarily deliver meaningful outcomes.

His remarks do not suggest that people should avoid AI agents. Instead, they raise a broader question about why people use these tools continuously and whether the time spent with them is helping them achieve their goals.

As AI becomes more deeply integrated into everyday work, the distinction between genuine productivity and the feeling of being constantly busy may become increasingly important.