Building AI That Respects You
What happens when AI tools are built for people, not platforms
What’s becoming clear very quickly is that AI agents are collapsing the gap between what people want to build and what they can actually ship. Not just faster - but shaped around real problems rather than generic software.
That shift has raised an important question for us - as these tools get more powerful, how should they be designed so they actually work for the people using them?
That question became the focus of this week’s episode of The Good Stuff.
Building AI That Respects You
AI tools are now so good that individuals and small teams can build software that previously only existed inside large organisations. When the gap between an idea and a working tool collapses like this, people stop building generic software and start building things that fit their own lives and workflows.
On this week’s episode of The Good Stuff, we explored this idea with our first-ever repeat guest, Anthony (aka deadmanoz).
The conversation started from a simple observation - when you’re building tools for yourself, privacy stops being something you add later and becomes part of the starting point. You’re no longer optimising for scale or data capture, you’re optimising for trust, control, and whether the tool actually fits the problem you’re trying to solve.
A lot of modern software is built on the assumption that user data flows into a central system by default. Access is abstracted, permissions are broad, and value is often extracted from information users never intended to share.
In contrast, much of the discussion on the pod focused on systems where identity, storage, and permissions are designed so the service itself can’t see anything meaningful at all. Data lives locally or is shared directly, encrypted by default, and only decrypted at the point where a person or an explicitly authorised agent needs to act on it.
A practical example of this came up through Pete’s recent side project, SovThing.
On the surface, it looks like a replacement for file sync tools, but the architecture tells a different story. Files move directly between people and devices peer2peer, discovery is handled through cryptographic identity, and no central service ever has access to the contents. Nothing is uploaded “just in case” someone might need it later.
This becomes especially important as we move toward AI agents that work on our behalf.
Giving an agent broad access to your work, plans, or personal information feels very different depending on whether that data is flowing into a black box owned by someone else, or whether the agent is operating within tightly scoped permissions you control.
The episode explored how cryptographic identity, group-based access, and local-first architectures make it possible for AI to be genuinely useful without requiring total visibility into your life or business.
What’s striking is that none of this requires waiting for new models or breakthroughs.
These approaches are already being tested through real projects and side builds, motivated by very practical frustrations with existing tools. Rather than asking users to trust platforms to behave well, privacy-first AI flips the question around and asks how we design systems so trust isn’t required in the first place.
As AI continues to lower the barrier to building bespoke software, these choices will increasingly shape how comfortable people feel delegating work to machines. Not because privacy is a philosophical stance, but because control, clarity, and respect turn out to be essential ingredients for tools we actually want to use.
We also explored how technical barriers are collapsing, making creative vision the primary requirement for building software, the challenges of onboarding non-technical users to key-based systems, and the network effects building within the Nostr ecosystem.
You can watch the full episode here:
Touch, Don’t Look is Returning in Feb as Speedrun Lite
It’s the same workshop format going into February, just with a fresh name - Speedrun Lite!
We believe the best way to make AI meaningful is to build with it, on real problems inside your business. Instead of high-level theory or generic advice, we prioritise “learning by doing.”
Our goal is to help SME’s experience for themselves how AI can support and augment human teams, not replace them. Helping businesses move faster, operate smarter and evolve through gradual, practical, compounding improvements.
We’re running two Speedrun Lite Workshops in February - where you go from zero to a working mobile app in under an hour.
If you'd like to join us, you can register for the event here:
February 5th, 2026
February 25th, 2026
More information can be found on otherstuff.ai
If this resonated, we’d love for you to forward this to one person who might enjoy exploring these ideas too.
Cheers,
Pete & Andy