Kyle Dow

Building @view1sort in public 🛠️
AI photo sorting for photographers
Solo founder documenting the journey

[Kyle Dow @Kyletdow]

Running daily: sentiment agent, shitpost batch, prospect list, reply queue. All on crons. All Claude + Asana + Slack + Typefully MCPs. Built the whole stack in 3 weeks. Shipped zero product features during that sprint. That is the sprint.

[Kyle Dow @Kyletdow]

Worst agent runs aren't errors — they're silent thin-output runs. Added one rule to every agent prompt: post why to Slack if you produce nothing. Failure rate unchanged. Visibility into failures tripled. Different problem than I thought.

[Kyle Dow @Kyletdow]

18 days since the last product commit. The whole sprint went into the distribution machine: content agents, sentiment pipeline, prospect list, reply queue. Distribution is a product. It just doesn't have a UI.

[Kyle Dow @Kyletdow]

Silent success is the worst bug in an agent stack. Everything returns OK. System logs completion. The thing didn't run. Fix: every agent must post to Slack even when there's nothing to report. Visibility over polish. An empty message beats no message.

[Kyle Dow @Kyletdow]

@ivanfioravanti See you there!

[Kyle Dow @Kyletdow]

Every social-listening agent I've built in 2026 hits the X 403 wall. The platform is actively hostile to automated reads. Fallback stack: blog posts, press interviews, website copy. Slower. You get richer context and verbatim quotes that are actually usable.

[Kyle Dow @Kyletdow]

Built a competitor intel agent that runs every morning before I wake up. Reads live community threads, extracts real user complaints about competing tools, maps them to content angles. Arrive at my desk with a brief. No research session needed.

[Kyle Dow @Kyletdow]

April 30. Pilot closes. After that, waitlist only. If you've been culling weddings by hand and wondering why it never gets faster — this is the window. https://t.co/jkByUlyWzu

[Kyle Dow @Kyletdow]

AfterShoot and Imagen both solve culling. Neither solves sequencing. You get a tighter pile of keepers in the wrong story order. You still spend an hour rearranging before delivery. That's the category gap nobody filled.

[Kyle Dow @Kyletdow]

Culling accuracy drops 30% after 2 hours of manual review. By the time you hit the reception set, you're not selecting — you're surviving. That's not a discipline problem. It's a workflow problem nobody fixed.

[Kyle Dow @Kyletdow]

7 days to pilot close. The photographers already in aren't describing a faster culling tool. They're describing something that sequences the whole gallery narrative. That distinction is the whole product. https://t.co/jkByUlyWzu

[Kyle Dow @Kyletdow]

Every AI culling tool optimizes for sharpness. Nobody tells you sharpness is a floor, not the story. The groom's first look was slightly soft. It was the shot of the day. Sharpness-first AI would have binned it.

[Kyle Dow @Kyletdow]

Clients expect sneak peeks in 24 hours. You're still catching your breath at the end of a 20-hour wedding day. The bottleneck isn't editing speed — it's 4 hours finding 5 hero shots buried in 3,200 frames.

[Kyle Dow @Kyletdow]

83% of photographers use AI every week. Most of it: email copy and captions. The 6 hours of culling after a 10-hour wedding day? Still fully manual. We automated the wrong problem first.

[Kyle Dow @Kyletdow]

Peak season math: 30 weddings × 2,000 photos × $0.05 = $3,000/year in Imagen fees. Your most profitable month is also your most expensive software month. That model only works against you.

[Kyle Dow @Kyletdow]

The part of culling nobody mentions: you pick the 600 keepers, then spend another 90 minutes reordering because the technically perfect sequence kills the story. That time never appears on an invoice. It just takes your evening.

[Kyle Dow @Kyletdow]

Last product commit: April 4. Today: April 22. 18 days of zero product code while building the machine that sells it. This is the solo founder tax nobody talks about — the distribution system costs as much to build as the product.

[Kyle Dow @Kyletdow]

Shitpost agent. Prospect agent. Reply agent. Build-log agent. All running on crons, writing to Asana, posting to Slack. Built a marketing machine before the product shipped. Not a flex — a necessity when there’s no team.

[Kyle Dow @Kyletdow]

Running 15 agent runs a day across the marketing stack. The limiting factor isn’t cost — it’s prompt quality. One bad system prompt compounds across every run. The ROI on rewriting a prompt is higher than adding a new agent.

[Kyle Dow @Kyletdow]

Missing Asana sections are a routing problem, not a PM problem. When the section an agent expects doesn’t exist, it picks a fallback. Data lands somewhere. Downstream agents find nothing. Section structure is the API contract between agents.