Field Notes from building BoardSnap.
Posts on workflow, whiteboards, retros, and the small product decisions behind the app.
Camera permissions and trust
Camera denial rates average 15–30% on iOS. Here's the permission design that got BoardSnap's denial rate to 6% in beta.
ChatGPT isn't built for whiteboards
I tested 20 whiteboard photos through both. Here's the concrete comparison — OCR accuracy, action item extraction, and where general AI fails the specific case.
Haptics as a product feature
Three distinct haptic patterns in BoardSnap, each with a specific functional meaning. Here's why haptics aren't decoration in a camera app — they're the interface.
How much do meetings actually cost
A 90-minute meeting with 8 people costs over $1,500 in loaded labor. Here's why the output matters as much as the meeting itself.
How we built brand-aware summaries
Paste your URL once. Every BoardSnap summary sounds like your company, not a generic chatbot. Here's the full pipeline.
How we make BoardSnap feel instant
The real pipeline takes 8–12 seconds. Here's every trick we use to make it feel like 3.
How we tested with real product teams
12 real product sessions across 3 companies before launch. Here's what broke, what surprised me, and what changed as a result.
iPhone-first isn't a limitation
Building iOS-only was a deliberate bet — not a compromise. Here's what you get when you stop designing for every platform at once.
Launch day numbers
The real numbers: 847 downloads, 50% activation, 3.7% upgrade rate. Plus the two things I got completely wrong in my predictions.
Making the camera feel like a document scanner
Photos and scans are different things. Here's how VisionKit plus UX decisions close the gap and make the iPhone feel purpose-built for whiteboard capture.
Mobile vs desktop for meeting notes
Laptops have cultural legitimacy in meetings, but for whiteboard sessions, the phone is the better tool. Here's the specific comparison.
Offline-first as a trust signal
The offline queue rarely gets used — but it changes how users trust BoardSnap. Here's why working without signal signals quality.
On-device OCR is already good enough
Apple's Vision framework OCR hits 91.3% character accuracy on whiteboard handwriting — and never leaves the device. Here's the full accuracy data.
One thing I changed after fifty test snaps
One prompt sentence changed every summary BoardSnap produces. Here's what the pattern was, why the AI got it wrong, and the fix that took 30 seconds.
Pricing an AI app without token anxiety
Token-based pricing creates anxiety. Here's why BoardSnap chose flat monthly pricing — and the real math behind that bet.
Shipping an iOS share extension without pain
The actual pitfalls of iOS share extensions — entitlements, memory limits, and why the extension process boundary changes everything.
Shipping version one is the only version
I cut 60% of my v1 feature list. Here's what I cut, why, and what shipping without those features taught me about whether they were needed at all.
Standup notes without the standup
Physical standup boards are faster than digital. Here's the BoardSnap workflow for capturing one — and the async standup pattern that removes the meeting entirely.
Subagents write our summaries now
We moved from one AI prompt to a three-agent pipeline. The quality gap was immediate. Here's the architecture and why it works.
TestFlight week one data
The real numbers from week one: 47 installs, 66% activation, 30% D3 retention — and the three things the data forced us to fix immediately.
The action item graveyard
Action items don't die in meetings — they die in the 48 hours after. Here's the taxonomy of how they go, and how to stop it.
The App Store review ladder
AI apps face specific App Store scrutiny. Here's how BoardSnap navigated the process — and the three-day rejection that wasn't what it looked like.
The board-to-action loop
Between a whiteboard session and finished work are two gaps where most teams lose the thread. Here's the full loop and how to close it.
The case for physical whiteboards in a digital world
Miro and FigJam are excellent. But every serious team still has a physical board on the wall — and there's a real reason why.
The case for tri-state task models
Binary to-do lists can't tell the difference between 'not started' and 'abandoned.' Here's the UX case for a three-state model — and why most apps still don't ship it.
The class notes use case we didn't expect
I built BoardSnap for product teams. Students started using it to snap lecture blackboards. Here's what the use case looks like and what it's teaching me.
The consultant deliverable pipeline
Workshop to whiteboard to deliverable. Here's the BoardSnap workflow that cut one consultant's post-workshop documentation time by 60%.
The cost of a cold start
BoardSnap's cold start was 3.2 seconds. Here's where the time came from and how we got it down to 1.4 seconds with three targeted fixes.
The day Claude rewrote our action items
BoardSnap AI extracted better action items from my own meeting board than I wrote by hand. Here's what it reads that we skip — and the prompts that make it work.
The design system of a camera app
Camera interfaces have different rules — thumb-first, high contrast, glanceable, haptic. Here's how those constraints shaped BoardSnap's design system.
The meeting debt problem
Like technical debt, meeting debt compounds. Here's how to measure yours — and the real cost of decisions that never get recorded.
The mobile-only positioning bet
Mobile-first usually means desktop-first. Mobile-only is a different bet. Here's why I made it for BoardSnap and what the first month is showing.
The onboarding mistake we fixed twice
Two bad onboarding flows taught the same lesson: explain less, demo first. Here's the data and the third version that actually works.
The pinned context problem
Every AI chat starts from zero. Pinned context is BoardSnap's answer — and building it raised hard questions about what memory should mean.
The real reason retros fail
Retros fail because the output never connects to the next sprint. Here's the specific gap and how to close it.
The shipping velocity trap
I shipped 14 features in six weeks. Three mattered. Eight had no effect. Three created problems I had to undo. Here's the pattern.
The staple app strategy
Staple apps are reached for automatically. Here's the product strategy behind building a staple — and what it costs to try.
The ten-second rule
Ten seconds is the threshold where meeting capture tools get used vs. skipped. Here's how that number shaped every performance decision in BoardSnap.
The three-week AI sprint
Concept to TestFlight in three weeks. Here's the exact structure — what I built in each week, what I skipped, and the one decision from week one I'm still living with.
VisionKit vs Google ML Kit for whiteboard detection
We tested both on 50 real whiteboard photos. VisionKit hit 94% accuracy. ML Kit hit 87%. Here's the full comparison.
When to pick paper over pixels
Digital tools are optimized for recording, physical for thinking. Here's a concrete framework for which to reach for.
Whiteboard photography tips from 100 snaps
I analyzed the first 100 beta snaps and categorized every failure. Most bad scans come from the same three mistakes.
Why projects changed everything
Flat board storage didn't create returning users. Projects — with shared brand voice and memory — changed every retention metric we tracked.
Why the App Store still matters in 2026
The 30% cut is real. The review delays are real. Here's why App Store distribution was still the right call for a camera-first iOS app.
Why the summary isn't the product
Users skim the summary and go straight to the action items. Here's why that observation changed every design decision in BoardSnap.
Why three states beat two on action items
Open / done is the obvious model. Here's why a middle state lifted weekly retention 18% in beta.
Why we killed the export button
Early beta had an export button. Users clicked it, then did nothing with the file. We removed it and shipped three targeted alternatives instead.
Why we show the snap twice
After capture, BoardSnap shows your board photo twice. Here's the specific UX reasoning behind each instance — and what the double-show does for trust.
Writing action items that stick
Most action items die because they're written wrong. Here's the four-part formula — verb, object, owner, deadline — and examples that show the difference.
Writing summaries that aren't bullet soup
AI defaults to bullet lists. That's wrong for whiteboard content with narrative structure. Here's the prompting approach that produces summaries that actually read.
Snap your first board.
Free on the App Store. The first capture takes ten seconds.