How much do you trust AI agents?
With the advent of clawdbots, it's as if we've all lost our inhibitions and "put our lives completely in their hands."
I'm all for delegating work, but not giving them too much personal/sensitive stuff to handle.
I certainly wouldn't trust something to the extent of providing:
access to personal finances and operations (maybe just setting aside an amount I'm willing to lose)
sensitive health and biometric information (can be easily misused)
confidential communication with key people (secret is secret)
Are there any tasks you wouldn't give AI agents or data you wouldn't allow them to access? What would that be?
Re. finances – Yesterday I read this news: Sapiom raises $15M to help AI agents buy their own tech tools – so this may be a new era when funds will go rather to Agents than to founders.
Replies
The trust issue for me isn't about how much personal data an agent has, it's whether it's honest about how sure it is. Building FounderFlow taught me this the hard way: an AI that gives a confident-sounding answer when it's actually guessing is more dangerous than one that says it isn't sure yet. We ended up building explicit confidence grading (Verified vs Needs Review vs Monitor Only) into it for that exact reason, people trust a system more once it's allowed to admit what it doesn't know.
the health/biometric line is the one that stopped me - it's why HealthOS runs the voice checkin entirely on-device, nothing leaves the phone. curious if that's a hard no for you regardless of where processing happens, or does on-device change the calculus at all?
finances and health data are the easy no's for me, everyone agrees on those. the one that trips people up less obviously is calendar/email access, because that's where the "personal" and "professional" data actually mix - a client's home address in a meeting invite, a family emergency mentioned in a thread you forwarded to yourself. you don't think of it as sensitive because it's just logistics, but an agent reading your inbox to be useful ends up seeing all of it anyway. I've made peace with API-key-scoped stuff, but full inbox access is still a hard no for me even though it would make half my workflows easier.
I run into this a lot. I'm the founder of FounderFlow, and my honest answer is I don't hand AI agents control, I hand them visibility. It watches what's going on across my businesses and tells me what actually needs my attention today, but it doesn't touch money, doesn't message people for me, doesn't make decisions on its own. That's the line for me. Delegate the noticing, not the acting.
I’m comfortable letting AI handle productivity tasks, but I’d be cautious with anything involving money, private relationships, or sensitive personal data. The ideal future is probably not “AI does everything,” but “AI handles the work while humans keep ownership of important decisions.”
Great question. My trust in AI agents scales with how verifiable the task is. For low-stakes, checkable work (summarizing, drafting, research starting points) I lean on them daily. For anything sensitive or irreversible, I still want a human in the loop. What's interesting to me is the flip side: it's not just how much we trust agents, but how much they "trust" or correctly interpret the content they read. I've been digging into how AI agents actually parse websites, and they often miss or misread key info that looks obvious to a human. Makes me think reliability runs in both directions.
The two dial framing above is right, but for money there's a third thing that doesn't map cleanly to either one: reversibility isn't binary. A wrong email you can apologize for. A wrong transfer might get pulled back if you catch it same business day, or might just be gone once it clears overnight or crosses institutions. That's the actual reason I'd rather have an agent draft a transfer for me to approve than execute one directly, not distrust of the model, just distrust of how hard undoing a mistake really is once money has moved.
The health data point is the one worth sitting with. Financial risk is bounded and reversible in a way biometric data never is. Once something about your body or mind has been inferred and stored, you can't really take that back, even if you delete the raw inputs. I'd rather see products default to on-device processing for that category specifically, instead of folding it into a general least-privilege policy that treats all data the same.
Softorino 💻📲
My honest split after this year: I let AI agents do the work, but never the final word. Research, drafts, tracking, scheduling – delegated completely, and it genuinely gave me hours back. But nothing goes public under my name until I've read it, because the agent's mistake instantly becomes MY reputation, and trust has no undo button. The part that worries me isn't agents doing tasks badly – it's them doing the wrong task confidently. So the approval step stays human at our company, probably forever.