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
I think trust is the wrong frame. I trust a calculator too but I still check the output before sending the invoice.
the way I think about it: agents get access to anything where the downside of a mistake is fixable. drafting emails, sourcing candidates, summarizing docs, scheduling. all fine. the blast radius of a bad output is low and you can catch it before it goes anywhere.
what they don't get: anything where a single wrong action is irreversible. moving money, sending legal docs, deleting data, communicating on my behalf without review. not because the AI is bad but because there's no undo button.
the Sapiom thing is interesting but "AI agents buying their own tools" sounds like a solution looking for a problem. if my agent needs a tool, I'll buy it. I don't need my agent to have a credit card.
minimalist phone: creating folders
@curtis_swick don't look only at reversability but also the extent of harm it can cause :D
MacQuit
As a developer, I think about this a lot. My personal rule is: I trust AI agents with tasks that are reversible, but not with things that are hard to undo.
For example, I'm comfortable letting AI write code, generate drafts, or analyze data. If it gets something wrong, I can review and fix it. But I would not let it push code to production, send emails on my behalf, or make financial decisions without me reviewing first.
The key issue is that most AI agents today don't have a good "undo" mechanism. When a human assistant makes a mistake, you can usually catch it and correct course. With AI agents running autonomously, by the time you notice a problem, the damage might already be done.
I think the most practical approach is permission boundaries. Give agents access to do specific, well-scoped tasks, and keep a human in the loop for anything with real consequences. It's not about trusting or not trusting AI in general, it's about designing the right guardrails for each use case.
minimalist phone: creating folders
@lzhgus Has anytime happened that your AI agent did something wrong? What was the worst thing?
I think the trust boundary is less about the tool and more about the type of action.
For me it usually looks like:
- high trust: analysis, drafts, research
- medium: suggested actions with approval
- low: anything irreversible (money, messages, system changes)
Feels like most issues happen when those boundaries aren’t clear.
Once agents cross from “suggesting” to “acting” without friction, that’s where things get risky.
minimalist phone: creating folders
@bigcat_aroido I think we are oscillating on the same level of trust :)
I like to use Claude Code and it is a daily tool that I use, but I never trust it - I always tell it to verify all its changes and check for race conditions, edge cases, and bugs, infinitly until it approves that everything is good and correct
minimalist phone: creating folders
@adi_leviim no ai tool is saint :)
@busmark_w_nika Not at all, but they help us a lot, making development super fast - just need to review any change those tools do.
Don't trust but I have to. That's it. Feel like we need to make a lot of things on security and awareness
minimalist phone: creating folders
@german_merlo1 don't we have a choice? :D
@busmark_w_nika seems not Nika! That's the game and we're already playing
minimalist phone: creating folders
@german_merlo1 😭 That's scary.
lead gen. mostly. and automatic replies. can't fully trust with money tho... there's a news here that some Claude bot users bought an entire course just to serve it's master useful information regarding what he's looking for.
minimalist phone: creating folders
@kilopolki Damn, I would go crazy if it used my money like that. 😂
minimalist phone: creating folders
@kilopolki This? :D https://www.instagram.com/p/DUL0RCLFEvv/
I trust agents with execution, not judgment — scheduling, research, drafts are fine, but anything involving money movement, health data, or irreversible decisions still needs a human in the loop.
The trust question is really about boundaries and observability. I think about it in tiers:
Tier 1 (full trust): Research, drafting, data analysis, coding assistance - tasks where I can verify outputs before acting on them.
Tier 2 (supervised): Content publishing, email responses, social interactions - tasks where there's a review step or low blast radius if something goes wrong.
Tier 3 (manual only): Financial transactions, legal commitments, anything with compliance implications, direct customer communication without review.
The key is having clear handoffs between autonomous work and human decision points. If you can't articulate exactly what an agent is allowed to do and where it stops, that's a sign you need stronger guardrails.
Has anyone built explicit "stop and ask" checkpoints into their agent workflows? Curious what triggers you've found useful.
minimalist phone: creating folders
@taylorbrooksops But how you would traing the agent the way that it will not mess up the Tier 3? :)
I couldn’t agree more. AI agents are great for boosting productivity, but I’d never let them handle my finances, sensitive health data, or confidential conversations—trust needs clear boundaries.
minimalist phone: creating folders
@vivianzheng There is nobody to rely on :D
TinyCommand
AI agents are very strong at analysis-heavy work like forecasting, scenario modeling, and competitive analysis because they can process large datasets faster than humans. The boundary for me is not insight generation but autonomous execution. I am comfortable letting agents crunch data and propose decisions, but a human should own the final call when accountability or second-order effects are involved.
minimalist phone: creating folders
@priyanka_gosai1 work-related stuff to filtering, analysing – okay, but rather nothing else, right? :)
TinyCommand
@busmark_w_nika mostly work-related stuff. sometimes also personal stuff like expenses.
minimalist phone: creating folders
@priyanka_gosai1 aaaa, that would be too personal for me :D (no, I am not hiding anything, but such info can also be misused) :D