In the AI era, should products aim for complete automation or minimal access?
AI makes complete automation possible.
Need expense tracking?
Give the app access to:
- Bank accounts
- SMS
- Emails
- Transaction history
The AI automatically categorizes everything and generates insights.
From a product perspective, this sounds like the ideal user experience.
But I'm starting to question whether complete automation should always be the goal.
As builders, we often assume users want maximum convenience. Yet many users are becoming increasingly cautious about sharing sensitive data, especially when it comes to finances.
What if the role of AI isn't to access more data, but to require less of it?
For example, instead of reading bank transactions, an AI could understand:
"Paid ₹220 for petrol"
and handle categorization, insights, budgets, and analysis from there.
The user still stays in control of their data.
The experience is not fully automated, but it is dramatically simpler.
This question has come up repeatedly while building Vitmora, a conversational personal finance app, where we're constantly evaluating the tradeoff between automation, privacy, and user trust.
I'm curious how other founders think about this tradeoff.
As AI capabilities improve, do you believe the future belongs to:
- Complete automation with broad access to user data
or
- Intelligent assistance with minimal access and greater privacy
Which approach do you think users will trust more over the next 5 years?
I tend to lean toward minimal access by default, with deeper automation becoming an opt-in as users gain confidence in the product. At Technource, we've seen that trust often becomes a bigger adoption factor than the AI capabilities themselves. A product that demonstrates value before asking for broader access is far more likely to build long-term engagement. Complete automation is powerful, but only when users feel they're still in control. I think the products that balance intelligence with transparency will have the strongest advantage over the next few years.
The framing I would add: the question isn't only how much access the AI gets, it's whether the user can catch a wrong call before it costs them. In finance the scary failure is a confident miscategorization nobody reviews, so the design that earns trust shows its work and lets you reject a line before it sticks. A cheap reject step matters more than the access level.
That's a great point. I think "showing the work" is an underrated part of building trust with AI.
In finance, a transparent AI that explains why it categorized something a certain way—and gives users an easy way to correct it—can build far more confidence than a system that's fully automated but opaque.
For me, the goal isn't just reducing access; it's giving users control at every step. If AI can be both privacy-first and easily correctable, I think that's where long-term trust will come from.
Control at every step is the right frame, but the correction has to be cheap and right there, or people stop using it and just do it by hand. The version that stuck for me was making "reject this line" a one-tap step before anything commits, so a wrong call costs a tap, not a cleanup. Opaque-but-automated loses the moment the first wrong call slips through unseen.
We ran into this same question building AI workflows for clients. Full automation works great until it hits something the model wasnt trained for. Then you get a confident wrong answer that looks right.
For finance apps trust is the actual product. Plaid access is elegant on paper but asks for trust upfront. Letting someone describe expenses conversationally builds that trust over time.
Have you tested both approaches with users yet? Curious which one resonates more.
That's exactly the concern I've been thinking about while building Vitmora.
Complete automation is incredibly attractive when it works, but in finance a single incorrect assumption can damage trust much faster than it saves effort.
One thing we've noticed is that many people are surprisingly comfortable spending a few seconds logging something like "Paid ₹220 for petrol" if it means they don't have to connect bank accounts, share SMS access, or give an app visibility into their entire financial life.
The interesting part is that AI can still provide categorization, spending insights, budgets, and analysis from that simple input. So the question becomes: does AI always need more access, or can it create value with less?
My current hypothesis is that the next generation of finance products will sit somewhere in the middle — using AI to remove friction while still keeping users in control of their data.
We're still learning from users, but it's been fascinating to see how often privacy and trust come up in conversations around Vitmora.
Curious whether you've seen users become more privacy-conscious over the last couple of years, or if convenience still wins most of the time?
Depends on whose money it is. Personal finance, people lean toward privacy. They'll type expenses manually to avoid connecting a bank account. Business spending, convenience wins immediately because the time saved justifies the data shared.
The underexplored space is where you're building -- AI that works with less. Most finance apps go full automation or full manual. That middle is wide open.
That's a great insight. The personal vs business spending distinction makes a lot of sense.
I also agree that the middle ground feels surprisingly underexplored. AI getting smarter with less access is exactly the space I'm interested in exploring with Vitmora.
Building something similar in that middle space. The personal vs business distinction was the thing that cracked it for me too - most people want different tradeoffs depending on context, not a one-size compromise.
Curious how you're handling the boundary setting in Vitmora. Per-session choices or a more fixed permission model?