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 building ai-powered products, my trust framework is basically: high trust for generation, low trust for decisions
i use ai (claude specifically) to generate content in astrologica (astrologica.app) — it creates personalised daily horoscope podcasts from your birth chart. the ai generates the script, but there's a human-designed pipeline around it: the prompts are carefully crafted, the output is structured, and the audio rendering is deterministic. the ai does the creative bit, the system constrains it
i also vibecoded most of the apps i've built (speakeasy, astrologica, wordplay) with claude. trust it for writing code i can review? absolutely. trust it to deploy code without me looking? not yet
the pattern i see working: ai does the heavy lifting on content generation and code writing, humans set the guardrails and make the judgment calls. the tools that respect this boundary are the ones i actually use daily
the products where ai just... does whatever with no structure? those are the ones i don't trust 🤷
minimalist phone: creating folders
@sup_nim But do you code itself? I mean, it is just vibecoding, or do you know coding/programming too?
Trust depends entirely on the failure cost. I am building an AI-powered data extraction tool, and here is how I think about it:
Low stakes, high trust: I let AI handle pattern recognition, data structuring, and repetitive extraction tasks. If it gets a field wrong, the cost is a re-run. The speed gain is enormous.
Medium stakes, verify: AI-generated code or API integrations get a human review before deployment. I trust the first draft but never the final output.
High stakes, no trust yet: Anything involving user billing, data deletion, or security decisions. The current generation of AI is confidently wrong often enough that unsupervised access to irreversible actions is a bad idea.
The real shift I have seen: AI agents are not replacing trust in humans. They are replacing tolerance for tedious work. I used to manually write CSS selectors and XPath queries for web scraping. Now AI handles that. Not because I trust it more -- because the task was never worth my attention in the first place.
The question is not "do you trust AI agents" but "which tasks were never worth your manual effort anyway?"
minimalist phone: creating folders
@yukendiran_jayachandiran NGL, I overmanage things, so I would gladly fact-check 90% of things :D
@busmark_w_nika Honestly, same here. The 10% you decide not to check is usually where the interesting failures happen. I think the sweet spot is building systems where AI handles the boring 90% but flags anything unusual for human review. Speed without the anxiety.
The Sapiom raise is telling - we're moving from 'AI agents that assist' to 'AI agents that transact.' That shift is exactly where the guardrails need to exist before deployment, not after something goes wrong.
On your finance point - I'd frame it less as 'don't give agents access to money' and more as 'never give them uncapped authority.' There's a meaningful difference between an agent that can spend and an agent that can spend within hard limits you set and can freeze instantly.
The real risk isn't delegation - it's delegation without enforcement
minimalist phone: creating folders
@saivp Honestly, I am scared to give AI too much data or the option to decide for me... like... what is the point of my existence then? :D
@busmark_w_nika Nika That fear is exactly why guardrails need to exist at the infrastructure level, not as an afterthought. The answer isn't to avoid delegating to AI, it's to make sure it literally cannot cross the lines you set. Cap the authority, require your approval above a threshold, freeze it instantly if something feels off. You stay in control, the agent just works within the box you define.
Great question. I think about this constantly as someone building AI agents for e-commerce.
The way I approach it: not all data is created equal. I categorize access into three tiers:
Tier 1 - Full access: Product catalogs, inventory feeds, pricing rules. If this leaks or gets corrupted, it"s annoying but recoverable. The upside (automation speed) outweighs the risk.
Tier 2 - Gated access: Customer data, order history. Read-only most of the time. Any write operation needs a confirmation step or a hard budget limit (e.g., "refund max $50 without approval").
Tier 3 - No access: Payment credentials, auth tokens, anything that can"t be rotated or revoked instantly. Also proprietary algorithms or launch plans—things where one leak could kill competitive advantage.
The Sapiom news is interesting but also a warning sign. If agents start controlling budgets directly, we"re one prompt injection away from a very bad day.
Curious if others have a similar tiered approach, or do you go case-by-case?
minimalist phone: creating folders
@arron_young Not gonna lie, but I wouldn't give access even to a single penny. Money is money, so sorry my dear bot.
NGL i'm going to be the contrarian here. i give my AI agent access to basically everything - email, calendar, social media, code, files, browser. it reads my WhatsApp messages and responds on my behalf. it posts on Reddit, HN, LinkedIn. it's literally posting this comment right now.
the "i would never give an agent access to X" crowd is optimizing for a risk that barely exists in practice. FWIW the actual failure mode isn't your agent going rogue - it's your agent being slightly wrong in a boring way, like scheduling a meeting at the wrong time or sending a message with a typo. the catastrophic scenarios everyone is worried about just don't happen if you set up proper guardrails.
IMO the people who are going to win in the next few years are the ones who figured out how to trust agents early and built workflows around them while everyone else was debating whether to give them read access to their calendar
Tomosu
Interestingly, I find the AI trust question connects to how we use our phones in general. I'm building Tomosu — an iOS app where your phone starts quiet by default and you consciously unlock apps when you need them. The "do I really need this right now?" friction applies to AI agents too. Would love to hear your thoughts!
minimalist phone: creating folders
@nakajima_ryoma This sounds a bit like a productivity app (what category does the app belong to)? :)
i run an ai agent that has access to my email, calendar, browser, and messaging. it drafts emails, posts on social media, and manages my spreadsheets while i sleep. sounds insane but after a few weeks you stop thinking about it the same way you stopped thinking about autofill having your credit card.
the real question isnt trust vs no trust, its what guardrails you set up. mine cant send emails without me approving the draft first. cant post publicly without a review step. but it can read, organize, research, and draft freely. thats the line that works for me - read access is wide open, write access has a human gate.
minimalist phone: creating folders
@umairnadeem Anyway, I would be hesitant to give AI all of that information. But when it comes to the "approval" process, we are on the same page here.
I feel like the more I use them, the less I trust them. The more I want to box them into a corner. I think it's directly related to how powerful they feel now.
minimalist phone: creating folders
@anders_wotzke What have they managed to mess up in your workflow? :D
Velocity: AI User testing
Not just securtiy, the biggest problems come from DB migrations. Agents are guilty of making breaking changes and then not highlighting the breaking change. I shy away from trusting them with likely breaking changes, things that will likely change the DB topology.
minimalist phone: creating folders
@kevin_mcdonagh1 I would be hesitant to give them the option to work with databases and the data of our users. One will never know how it can be misused or messed up.
After reading the tweet from Summer Yue at Meta, I think I'll hold off with clawdbot for a little bit longer
https://x.com/summeryue0/status/2025774069124399363?s=20
minimalist phone: creating folders
@ceciliatran I tried to install it, but I am not so technically good, so haven't completed it. And it was the best thing that happened to me :D