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
As someone with around 12 years experience in ML/AI apps, I will say I dont have full faith. The problem is AI agents lack accountability. I have seen many times people write bad code and introduce bugs (myself included ofc) but its always a human being behind that can be blamed but also re-trained, take responsibility, fix, improve. When AI agents do all the work then we end up with systems that are intransparent, inefficient, with lots of technical debt and with noone to actually know what has happended and how to fix. That causes serious security issues and I wouldnt feel (at least for the moment) comfortable giving over full control of entire processes to AI agents. Thats why, even though I am a big advocate of vibe coding, I always oversee commits, I check the unit tests, I resolve merge conflicts myself and generally play a big role in creating my new app (soon to launch but not yet :)).
minimalist phone: reduce your screentime
@georgios_sarantitis_ IMO, vibecoding has meaning only for those, who are willing to understand the code (so at least understanding coding/programming itself). and not blindly copy paste.
Honestly? I trust them more than I expected, but only because I treat them like a junior dev — great at execution, terrible at judgment. I use Claude Code daily to build my app and it's been a game changer for shipping faster. But I always review the code, always test, and never let it make architectural decisions alone. The moment you blindly trust the output is the moment you get a beautiful function that subtly breaks three other things. Trust the speed, verify the output. That's the balance I've found so far.
minimalist phone: reduce your screentime
@thenomadcode what are you working on, and how do you use AI for getting your project done? Does it show you the code lines you need to insert there, or do you just know what to insert there, but double-check with Claude?
I am trying to learn coding by using that, but it always shows me the solution. I need to remind him not to do that.
Loova Agents
It really depends on whether it can deliver what I need. So it’s not about trust, it’s more about whether the outcome matches my expectations.
minimalist phone: reduce your screentime
@constance_tong but with AI, you can have more attempts (within a short period of time), so maybe it is about "how many times I will try to get the right output" :)
Trust comes down to one thing: can you verify the output?
We build AI that analyzes zoning codes and building permits. High stakes stuff — get it wrong and someone builds something illegal. So we designed it around "trust but verify": the AI does the heavy lifting (parsing 10,000+ page municipal codes), but every output links back to the source data so a human can gut-check it in seconds.
I'd never trust an AI agent I can't audit. The best AI tools aren't black boxes — they're transparent systems that make humans faster, not replaceable.
WUPHF by Nex.ai
Trust for me is less about the category of task and more about reversibility. I will happily let an agent draft emails, reorganize files, or write code all day. But the moment an action is hard to undo, like sending money, deleting data, or posting publicly, I want a human checkpoint. The real unlock will be agents that understand their own confidence level and ask for approval only when the stakes are high. We are not there yet, but the trajectory is encouraging.
minimalist phone: reduce your screentime
@najmuzzaman I think that if once something is done, it is hard to make it "reverse". The thing is already happened :)
Konfide
Great topic @busmark_w_nika I am particularly focused in this topic lately.
I will not give access to: 1- Write love letter for my wife :) and 2- Anything that has no human oversight and it is mission critical such as financial data, taxes, work related confidential documents and health information.
minimalist phone: reduce your screentime
@felipe_daguila love letter should be personalised, but yeah, maybe AI would leak something that wouldn't be okay :D
Okan
Giving an AI agent write access to my Stripe account or production database is still terrifying to me, so I'm totally with you on keeping finances siloed. For my own workflows, I restrict agents to strictly "read-only" permissions for customer data, leaving any actual destructive actions or database writes to manual approval. That Sapiom news about agents getting their own funding to buy tools is wild, but until we completely solve the hallucination problem, I'd only ever trust them with a virtual card that has a very strict spending limit.
minimalist phone: reduce your screentime
@y_taka Well, currently my profits on Stripe are not so good, so at this point, I would be okay to give the AI agent access :D
Depends on the agent.
I trust the ones I built myself, because I know how they work and what they actually have access to. Third-party agents, much less.
But even with my own, I keep the scope tight. Only what's needed for the task, nothing more.
No finances. No life management. No super sensitive data. AI is a tool for me, it helps me get work done, it doesn't run my life.
minimalist phone: reduce your screentime
@luka83184 Can you control your agent? At what extent?
Depends entirely on the blast radius of being wrong.
I build voice agents for small businesses. Plumbers, roofers, dentists, and law firms, so I think about this every day. The trust ladder I've landed on, from "ship it tomorrow" to "not yet":
🟢 High trust - ship it now
- Intake. Name, number, problem, urgency. The agent is literally just a smart form with a voice. Worst case: human re-asks one question on callback. Cost of failure: ~zero.
- Triage. "Is this an emergency or can it wait until Monday?" Binary routing. Easy to verify.
- FAQ deflection. Hours, address, do-you-service-my-area. Static info, static answers.
🟡 Medium trust - needs guardrails
- Booking. Fine when the agent can only book inside a pre-approved calendar window. Disaster when it confidently invents an appointment slot.
- Lead qualification. Good for "is this person in our service area." Bad for "how much should we quote them."
🔴 Low trust - not yet, maybe never
- Anything with money attached. Quoting jobs, processing payments, issuing refunds. One hallucinated number = a real customer pissed off in the real world.
- De-escalating an angry customer. The models can do this in a demo. In production, the failure mode is too expensive.
The pattern: trust scales with how cheaply you can verify the output. Intake you can verify in 5 seconds. A quote you can't verify until the truck shows up.
Curious where everyone else draws the line, especially folks running agents in customer-facing roles. What's the highest-stakes task you've actually let an agent own end-to-end?
(Wrote up the after-hours intake pattern in more detail here if useful: https://pulpaistudio.com/after-hours-answering-service/)
minimalist phone: reduce your screentime
@adam_pichardo Thank you for your POV, I would frame it the way that it is really individual on use-case, but we all can angry that money-related things are really sensitive to handle.
In my opinion, AI is currently incapable of replacing humans. Its ultimate limit is performing routine tasks. Today, humans should be the ones doing the thinking, and AI will then assist.