Jonathan Scanzi

Super Idea, how to mix Art with Work. We need more tools like that. Just for fun :-) (And also to remind to everyone how it's hard to build a tool)

The '120+ rotating prompts' detail is interesting — do people actually exhaust those or does novelty wear off around week 3 regardless? The real problem with journaling apps isn't the blank page, it's that reflection feels like homework after a long day. Three guided questions might lower the barrier enough, but I'd be curious whether the AI pattern-matching is what retains users or just what...

Curious how you're handling document variety across different markets — because a borrower's financials in Nigeria look nothing like what you'd see in Indonesia, and that inconsistency is usually where these systems break. The fraud-checking angle is the real differentiator here though. Most lenders I've talked to aren't losing to bad credit models, they're losing to document manipulation they...

Curious what the actual retention looks like once someone sets up their cutefolio — because the hard part with link-in-bio tools isn't the setup, it's whether people come back to update it. The 'cute' positioning is a real differentiator though, most of these tools feel like they were designed by a SaaS form builder. Analytics by country is a nice touch for creators trying to understand where...

Curious how the swarm coordination actually works when agents hit conflicting conclusions mid-task — like if one agent's research contradicts another's during a 50-page strategy doc. That's where these multi-agent setups tend to fall apart in my experience. The visual canvas angle is smart though, auditability is genuinely the missing piece in most AI workflows right now. Most people don't...

"AI slop" is basically all presentation tools right now — they generate something that looks polished until you actually read it. The conversational refinement loop is interesting, because the problem isn't usually the first draft, it's getting from draft to something you'd actually present without wrestling with a rigid editor. Curious how it handles brand guidelines — that's usually where the...

Curious how you handle the matching — does the developer self-select based on price, or is there some kind of routing on your end? The gap between "AI built it" and "I can actually maintain it" is very real, and I've seen a lot of teams hit a wall three weeks after shipping something on Lovable. Fixed pricing is smart here because the ambiguity of hourly rates would kill trust with...

Curious how you handle the quality control loop when agents push content directly to a CMS — that's where I'd get nervous. Auto-publishing brand-aware articles sounds good until one goes out that's slightly off-tone and you're doing damage control. The citation analysis piece is the most interesting part to me, because understanding what content actually influences AI answers is still pretty...

The discovery problem is real — most people don't know what to ask, so they never push the tool past the basics. Showing 15 concrete use cases is probably more useful than any onboarding flow. The part I'm curious about is how you handle the gap between what the agent can do in demo mode versus what actually works in someone's specific stack — that's usually where expectations break. Does it...

The on-screen text problem is actually one of the most annoying parts of video localization — everything else gets handled but then you've got slides or lower thirds in the wrong language and it breaks the whole thing. Curious how it handles text that's embedded in complex backgrounds or motion graphics — that's usually where automated tools struggle. If the detection is solid, this fills a...

The Stripe-native + full code ownership angle is the part I'd actually lead with more — most agency teams I talk to are scared of AI tools that create lock-in or bury the payment logic in a black box. What I'm curious about is how it handles the store design side once the data model is sorted — because generating a working product catalog is one thing, but getting to something a client would...

Curious how you handle the cases where the AI-generated PR fixes the symptom but not the root cause — that seems like the failure mode that would erode trust the fastest. The monitoring + auto-fix loop is interesting but it requires a lot of trust in the system, and most eng teams I know are still pretty cautious about auto-merging anything. What does the human review step actually look like in...

The visual test editor is the thing I'd actually use — clicking a step to see a live snapshot is way more useful than debugging a wall of logs after the fact. My question is how it handles flaky tests in CI; 4–5x faster is great until the suite randomly fails on a PR and someone has to chase it down. At told.club we hear a lot about teams abandoning automated test suites not because they're...

Curious how Penny handles edge cases — like when a borrower’s docs are inconsistent or their situation doesn’t fit a standard risk profile. That’s usually where loan processing actually breaks down, not in the straightforward cases. The ‘hours to seconds’ promise is compelling but I’d want to know if the AI is making final calls or just surfacing recommendations for underwriters. The trust gap...

How opinionated is the agent definition layer — can you bring your own prompting patterns or does it nudge you toward a specific structure? The infrastructure bundling (streaming + sessions + billing in one deploy) is genuinely useful, that’s the part most teams waste weeks on. My concern is that when the abstraction breaks — and it always does at some point — you want devs to be able to debug...

The ‘just scroll’ interaction model is interesting — most design tools put you in prompt mode, which assumes you know what you want before you’ve seen anything. Variant flips that, which is actually closer to how creative decisions get made in practice. My concern is whether the outputs stay coherent when you’re scrolling through variants or if it just becomes aesthetic noise after a while. The...

Running video generation fully local is the right call — latency kills creative flow, and waiting on API calls while editing is brutal. The interesting tension here is whether users want their NLE and their gen AI in the same tool, or if that coupling creates friction when one part is great and the other isn’t. What’s the inference speed like on mid-range GPUs? That’s going to determine whether...

Curious how the SVG output handles complex prompts — that’s usually where these tools fall apart, with messy paths that are technically editable but practically useless. The doodle-to-explainer-video pipeline is actually a smart angle though; most teams I talk to are drowning in static assets and have no cheap way to animate them. The real question is whether the vectorization is clean enough...

The original macOS volume overlay is one of those UI patterns that nobody notices until it breaks their flow — the fact that it doesn't respect your wallpaper has always been jarring. Curious whether VolumeGlass handles the edge case where the glass effect readability breaks against very light backgrounds, because that's usually where ambient UI elements start to feel like noise. Does it adjust...

The silent failures framing is sharp — half-implemented features and unclaimed value are the real activation killers in most B2B products, not churn from explicit dissatisfaction. Curious how Vet surfaces these gaps: is it correlating usage data against the expected activation path, or more of a qualitative signal from user sessions? The distinction matters because one tells you what's broken...