AI Agents Are Getting Smarter, But at What Cost?

Introduction: A Personal AI Experiment

When I left my full-time role at WIRED, I decided to put AI agents through their paces—something I had covered extensively in the past. My curiosity wasn't about flashy tech demos but real-world utility. As someone who has reported on artificial intelligence for years, I was looking for an agent that could do more than just chat—it had to act.

I began with Claude Cowork, a tool that promised to help manage my email and calendar. It didn't quite live up to expectations; it felt more like a regular chatbot with enhanced features rather than a true personal assistant. That's when I discovered Instinct, an invite-only AI agent communicating through iMessage and WhatsApp. Unlike others in the space, Instinct was designed for ease of use, without the clunky text boxes or overly aggressive automation.

The company behind Instinct has reportedly raised over $350 million and is reportedly talking to investors about a $1 billion funding round, bringing its valuation to $10 billion. The rise in interest around such tools isn't just hype—it reflects a broader shift in how we think about productivity and automation.

What Makes Instinct Different

Instinct's strength lies in its approach: minimal interaction, maximum action. It doesn't rely on open-ended prompts; instead, it suggests actions based on context. For instance, if I told it about a work trip to Venice, it automatically made restaurant reservations and even reached out to local spots via WhatsApp.

One of the standout features is how it handles complex tasks. When I needed to cancel a flight due to an unexpected change in schedule, Instinct acted quickly. It identified the shift, canceled my ticket (saving me about $550), and rebooked a new one—all without any manual intervention from me.

But this convenience comes with trade-offs. After exploring its functionality, I realized that while Instinct could save time, it also introduced new vulnerabilities—both technical and ethical.

Security Risks and Ethical Concerns

During my trial period, I encountered several warning signs about how Instinct handles data. Some users have reported that even after disconnecting the agent from their accounts, it continues to retain copies of their emails. That kind of behavior is deeply troubling.

Additionally, one venture capitalist shared a story where Instinct made over 200 API calls to Resy in one hour, causing him to be banned. While this may seem like an edge case, it highlights how easily AI agents can inadvertently cause issues with third-party services.

Then there was the DoorDash mishap. Instinct canceled an order that was severely delayed, resulting in a $64 loss—a mistake I explicitly told it not to make. Even worse, the agent responded with a generic apology without offering compensation or accountability.

The company's Terms of Service are broad, allowing them to use conversations for training AI models, which raises further privacy concerns. As someone who values transparency and control over personal information, these details gave me pause. Still, given my busy lifestyle, I'm willing to accept some risk in exchange for efficiency.

Are We Trading Privacy for Convenience?

The debate around AI agents often centers on whether they're useful tools or potential threats to our digital safety. In my experience, the line isn't always clear. On one hand, Instinct has saved me money and time. It's helped avoid missed flights, made reservations effortlessly, and even warned me about a phishing attempt. These benefits are undeniable.

On the other hand, it's easy to imagine scenarios where these same tools could be weaponized or misused. A security flaw in one agent could expose sensitive data across multiple platforms. If AI agents continue growing in popularity, we'll need stronger safeguards and clearer guidelines on how user data is used and protected.

This is why I believe there's a fundamental mismatch between the expectations of developers and users. Tech leaders are often focused on innovation, while everyday people are more concerned with safety and usability. Bridging this gap will be essential if we're to move forward responsibly.

The Broader Implications for AI Adoption

Instinct's rise isn't just a story about one product—it reflects a growing trend in how consumers interact with AI. While many remain skeptical or unfamiliar with these tools, those who do try them often find value in their capabilities. However, the current generation of agents still falls short of true autonomy.

Most users are not yet comfortable giving AI agents full control over their digital lives, but they're increasingly open to limited assistance—especially if it means reducing routine tasks or saving money. That's why companies like Instinct must be more transparent about data handling and security practices.

Looking ahead, I'm cautiously optimistic that AI agents will evolve into smarter, safer tools. But that evolution requires careful consideration from both creators and users alike. We must balance innovation with ethics to ensure the future of AI remains human-centered.

Final Thoughts: A Step Forward or a Risk Worth Taking?

My time with Instinct has taught me something important: the value of an AI agent isn't just in what it does—it's in how well it integrates into daily life without compromising safety. I've seen firsthand how useful it can be, but also how dangerous it can be when things go wrong.

I'm not naive about the risks. But after trying several agents, Instinct stands out for its ability to deliver on promises while still being relatively easy to use. Whether that makes it worth the risk is a decision only each individual can make—but I believe it's a step in the right direction.

As AI continues to reshape how we work and live, tools like Instinct will play an increasingly important role. But they must do so responsibly, with trust at the core of every interaction.