Designing trustworthy AI features
The model is sometimes wrong. How do you ship it so users can still trust, and verify, it?
The idea inside
Design for the verification gap: stream, cite sources, and let users correct it.
After this lesson
You can name UX patterns that build trust (streaming, citations, human-in-the-loop) given the verification gap.
Where it leads
These patterns surface the model's output, but how do we guarantee its shape?
Inside this lesson
That's the real lesson stage, paused. Claim your pass to operate it.
See how AI actually works, end to end.
This lesson is one stop on the full arc. Unlock all of it, and keep it for life.
What you get
- The 34-lesson main path, a finishable route from a word to agents
- Goal tracks for using AI at work and building AI features
- Boss labs that make you apply a whole act, not just recognize it
- Spaced recall that brings each idea back before you forget
- Course memory: every term defined, with links to where it first appears
- A shareable capability card when you finish the main path
- Lifetime access on every device, every future lesson included
Not videos to watch. You predict, operate the machine, then prove it. That is why it stays.
99 interactive lessons and challenges. No videos, no code.
Free launch pass: lifetime access, no card needed
New here? The first lessons are free to try. Start with lesson 0.1
What this lesson shows
Design for the verification gap: stream, cite sources, and let users correct it.
The question it opens with
The model is sometimes wrong. How do you ship it so users can still trust, and verify, it?
The walkthrough, in the lesson's own words
- This answer is take-it-on-faith. Flip a pattern on and watch trust climb.
- Trust isn't the model being right, it's the UX letting users verify.
- The meter's weights are invented to show each pattern's relative impact, not measured user trust.
- A wall of text, no sources, dropped all at once. No way to check it, would you act on it?
- Same answer, fully verifiable, and now a guess is something users can stand behind.
- Citations move trust the most: a claim you can open and check beats one you take on faith.
- Citations. Streaming feels alive, but verifiability earns trust. People don't fully trust AI output, and they're right not to: the danger isn't that the model errs, it's shipping the answer with no way to tell. The risk lives in the UX, not just the weights.
- Three safety layers people mix up, because they do different jobs. Evals measure quality, before you ship (offline) and on live traffic (online); they tell you whether it's good enough. Guardrails act at runtime: filters that block a bad input or output as it happens, like a prompt-injection or PII check. Permissions scope what the system can do at all: which tools, which data, which actions need a human. You want all three: evals to know, guardrails to catch, permissions to contain.
- Your legal-research assistant gives confident, well-written answers, but lawyers won't rely on it because they can't tell which ones are made up. What would you add?
- Design for the verification gap. Add inline citations so every claim links to a source the lawyer can open and check, surface the model's uncertainty instead of bluffing, and keep a human approving anything high-stakes. Trust comes from letting users verify, not from the model being right.
- Confidence: medium, verify the exact window for international cards.
Key takeaway
You can design an AI feature people trust: visible sources, easy verification, graceful wrongness.
What you can do after this lesson
You can name UX patterns that build trust (streaming, citations, human-in-the-loop) given the verification gap.
Check yourself: Which design makes an AI feature more trustworthy?
- Cite sources, let users verify, and require approval for high-stakes actions(correct)
- Hide where answers came from
- Answer instantly with no detail
- Never let users correct it
Trust comes from letting users check the work: show sources, make answers verifiable, and require a human OK before high-stakes actions.
Where it leads: These patterns surface the model's output, but how do we guarantee its shape?
This is the written summary. The lesson itself is interactive: you predict, drag and operate the mechanism above, and the reveal answers you.