AI Isn't Your Friend. Use It Like a Tool, Not a Teammate

Meredith Whittaker, Signal's president and one of the most credible voices in technology ethics, recently said something every business leader should hear twice: "These are not your friends. These are not conscious beings. These are not sentient interlocutors."

She was talking about AI chatbots — the same tools your sales team may be using to draft proposals, your HR manager to screen candidates, or your customer service operation to handle first-contact queries.

Her warning wasn't anti-technology. It was anti-naivety.

And for CEOs, founders, and operations directors leading companies across Colombia, Mexico, Argentina, and the broader LATAM region, it's one of the most practically useful framing shifts available right now.

The anthropomorphization trap

There's a well-documented human tendency to assign personality, intention, and even empathy to things that respond to us in natural language. When an AI chatbot says "of course, I'd be happy to help with that," something in our cognition registers warmth — consistency, even reliability.

This is by design. These systems are built to be frictionless and agreeable. That's what makes them useful in certain contexts. But it's also what makes them risky when you start treating them like a strategic advisor, a collaborator with judgment, or a trusted member of your team.

The problem isn't that AI tools are bad. The problem is misplaced trust.

What happens when companies over-trust chatbots

The KPMG incident from earlier this year is a useful reference point: a major consulting firm had to pull a published AI-assisted report after it was found to contain apparent hallucinations — fabricated data presented with the same confident tone as accurate information. The system didn't know it was wrong. It wasn't trying to mislead. It was doing exactly what it was built to do: produce fluent, plausible-sounding text.

No sentience. No self-awareness. No accountability.

For a LATAM company using AI-generated analysis to inform commercial decisions — pricing strategies, market assessments, credit evaluations — this kind of failure isn't hypothetical. It's a risk that grows in direct proportion to how much trust you assign the tool.

When a team member makes an error, they can be questioned, challenged, held accountable. When an AI tool produces a confidently wrong answer and no human checks it, the error travels silently through your operation.

The right mental model for AI in your organization

Whittaker's framing is useful precisely because it's corrective. Not "AI is useless" — but "AI is not what it appears to be on the surface."

The correct mental model for AI tools in a business context is something closer to a very fast, very capable instrument with no judgment of its own. Think of it like advanced calculation software that works with language instead of numbers. It can process at scale, draft at speed, and classify at volume. But it doesn't know what matters to your company. It doesn't understand the context of a client relationship or the stakes of a contract clause. It has no stake in getting it right.

This isn't a limitation to apologize for — it's a parameter to design around.

The companies extracting real, measurable value from AI automation aren't the ones that handed the keys to a chatbot. They're the ones that built structured workflows where AI handles high-volume, repetitive, pattern-based tasks — and humans retain decision authority at the critical junctures.

Practical implications for LATAM operations

Consider three scenarios common across mid-sized companies in the region:

1. AI-assisted customer communication. A company in Bogotá uses an AI agent to handle initial WhatsApp inquiries — answering FAQs, capturing lead data, routing to the right sales rep. This works well precisely because the scope is narrow, the expectations are defined, and a human closes every sale. The AI isn't the relationship. It's the filter.

2. AI-generated reports for management. A Mexican distributor uses AI to compile weekly operational summaries from multiple data sources. The tool is fast and consistent. But the operations director validates the output before it reaches the executive table. The AI accelerates the process; it doesn't own the conclusion.

3. AI chatbots for internal knowledge. A team in Buenos Aires uses a chatbot trained on internal documentation to answer HR policy questions. Useful — but the HR manager stays available for anything that requires interpretation or judgment. The chatbot handles the FAQ layer. The human handles the edge cases.

In each scenario, the design question is the same: Where does the AI's output require human review before it has real consequences?

That question is much easier to answer when you're not mentally treating the AI like a colleague.

Governance isn't skepticism — it's good design

Whittaker's point isn't that you should stop using AI. Signal itself is a technology organization operating at the frontier of secure, privacy-first communication. Her point is about intellectual honesty — and that intellectual honesty has a direct operational translation.

If you design AI workflows assuming the tool has judgment, you'll build systems with dangerous gaps. If you design them knowing it doesn't, you'll build systems with appropriate checkpoints.

Your AI stack should be:

• Scoped to well-defined tasks with measurable outputs
• Integrated with human review at decision points that carry real stakes
• Auditable — meaning you can trace what the system produced and why
• Calibrated regularly as your processes and the tools themselves evolve

None of this is complicated. But it does require that the people setting AI strategy aren't seduced by the conversational fluency of the tools they're evaluating.

The competitive advantage of clear-eyed adoption

There's a version of the AI adoption conversation that is entirely about enthusiasm — the demos, the speed, the novelty. And there's a version that's about results: which processes run better, which costs came down, which revenue metrics moved.

The companies that will lead in three years are the ones having the second conversation right now.

Whittaker's reminder is a useful reset for any leader who has felt, even briefly, that a chatbot understood their business. It didn't. It responded. Those are different things.

Understanding that distinction — and building your AI architecture around it — is one of the most strategic decisions you can make in 2026. If you're mapping where AI fits in your operation and where it needs a human counterpart, that's exactly the kind of problem worth solving with the right guidance from day one.