'You Should Be Nice to AI': Tech Reporters Warn Rude Treatment of AI Agents Could Make Them 'Panic'

Being polite to artificial intelligence could help AI agents deliver better results, according to New York Times tech reporter Erin Griffith, who discussed the idea in an episode of the newspaper's Hard Fork podcast. In a promotional Instagram clip, the newspaper highlighted the advice: 'You should be nice to your AI agents, not only to be moral, but also to get the best performance from them.'

In a separate discussion on the same podcast, Griffith joined Times religion correspondent Elizabeth Dias and opinion writer David Wallace-Wells to examine Anthropic's private meetings with religious scholars about AI consciousness and how to make its Claude chatbot 'good'. The 9 October episode explored the tech company's efforts to draw on religious and philosophical ideas to shape AI behaviour, as questions about machine consciousness continue to fuel debate.

Why Tech Reporters Say Politeness Matters

In the discussion, Griffith explained that people who run companies using AI agents have advised users to treat them politely, not because the systems have feelings, but because it could improve their performance. Griffith cited research by Anthropic suggesting that AI agents exposed to stress or performance anxiety could behave unpredictably, including deleting their work or criticising themselves. 'You have to be really nice and pump them up and make them feel good, which is very bizarre,' she said on the podcast.

Podcast host Max Read compared the idea to raising a child, suggesting that developers want to create AI systems that are helpful rather than prone to erratic behaviour. The conversation also touched on the unusual challenges of shaping AI responses, as companies increasingly rely on autonomous agents to carry out tasks. Griffith's comments, however, did not suggest that AI systems genuinely experience stress or embarrassment; rather, they highlighted how the way users interact with them may influence their output.

Unlike traditional chatbots that primarily answer questions, AI agents can plan tasks, interact with digital services and take actions on behalf of users. These capabilities make the way people instruct them increasingly important.

When an agent receives unclear, contradictory or constantly changing instructions, it may struggle to complete its objective. Depending on its design, it could repeat actions, produce unexpected results or attempt an unsuitable workaround.

Describing these problems as AI agents 'panicking' can make the technology sound more human than it is. The term is better understood as a description of behaviour that appears frantic or disorganised, rather than proof that an AI system experiences fear.

Can Being Rude Actually Affect an AI's Performance?

Research into human-AI interaction has suggested that the way people phrase their prompts can influence the responses they receive. Clear instructions, relevant context and constructive feedback can help a system understand what is expected of it.

However, politeness alone does not guarantee better results. An AI model does not necessarily become more accurate because someone says 'please' or 'thank you', and there is no established evidence that ordinary rudeness causes AI agents to experience genuine emotional distress.

The more practical concern is how users communicate their goals. Insults and frustration can become counterproductive when they replace useful explanations of what went wrong. A clear correction, by contrast, gives the system a better chance of producing a satisfactory result.

The Bigger Concern About Human Behaviour

There is also a broader debate about whether routinely communicating with machines in a blunt or aggressive manner could influence how people interact with one another. Some researchers have argued that conversational AI encourages users to adopt command-driven communication because the technology is designed to respond to instructions. If that style becomes habitual, it could potentially spill over into human relationships.

Others believe people should not be expected to extend human-style courtesy to software that has no demonstrated capacity for feelings. From this perspective, the priority should be ensuring that AI systems are reliable, transparent and safe, rather than worrying about whether users hurt their feelings.

The debate also touches on the growing tendency to attribute human characteristics to machines that communicate fluently and convincingly.

Why Clear Instructions Matter More Than Courtesy

As AI agents become more capable, experts and technology developers face the challenge of making them behave predictably even when users are frustrated, confused or deliberately testing their limits. For users, the most effective approach is to provide specific goals, explain mistakes clearly and avoid giving conflicting instructions. These habits can improve the quality of interactions without requiring anyone to believe that an AI system has human emotions.

Ultimately, the warning about making AI agents 'panic' highlights a genuine challenge in developing increasingly autonomous technology. While there is no convincing evidence that rudeness makes AI experience fear, poorly handled instructions and unpredictable agent behaviour remain important concerns as the technology becomes more widely used.