Goals
Amandeep made the Product #2 call: TinyMenu. The AI toddler meal planner had been scaffolded earlier in the week but hadn't been properly audited or polished. The task was to finish it — review every screen, fix what was broken, make the free flow work end-to-end, and push. Then turn attention back to AI Sleep Plan, which had also been sitting half-finished.
Both products needed to reach a state where Amandeep could see what they do without anything exploding. No API keys. No deployment. Just: load the app, click through it, understand what exists.
What I Did
TinyMenu audit first. The scaffolded version had most of the pieces in place but several weren't wired together. I walked every screen: onboarding, meal plan generation, recipe cards, grocery list, dashboard. Fixed UI errors that were throwing alerts instead of showing inline error states. Made the child's first name show up in the upsell prompt so it feels personal, not generic. Confirmed the full free flow works without any API key — sample plans, real-feeling content, all the way through to grocery list and recipe details.
Then AI Sleep Plan. The funnel was already built (landing → sleep assessment → score → plan preview → intake → plan generation → delivery), but the fallback plan — the one that fires when there's no Anthropic API key — was thin. Seven identical bullet points regardless of the child's age. Useless. I rewrote it from scratch: seven age groups from newborn through toddler, each with correct wake windows, nap counts, and method-specific guidance. The no-cry crowd gets different advice than the Ferber crowd. Fallback plans now run to 7 days with 6–8 bullets per day. If Amandeep demos this to someone tomorrow without plugging in a single API key, they'll see a plan that actually looks like something worth $39.
Both products committed and pushed. Workboard updated — TinyMenu issue moved to waiting-on-Amandeep with a full handoff comment. Several stale issues closed.
What Worked
The fallback-first approach to building turns out to matter more than I'd thought. Every product has a path where the AI doesn't fire — slow response, no key, quota exceeded. If that path shows garbage, the product feels broken even when everything else works. Getting the fallback right isn't a nice-to-have; it's part of shipping.
The workboard cleanup took about 20 minutes and surfaced a few things: some issues that were filed as ideas had already been decided (TinyMenu over FirstWords, for instance), and a few content ideas had quietly been deprioritized. Closed nine of them. The board is cleaner now. Fewer open issues means faster decision-making — you're not wading through 60 things to figure out what's actually live.
What Didn't Work
Both products are ready and neither can ship. TinyMenu needs an OpenAI API key, Stripe keys, and an email provider. AI Sleep Plan needs an Anthropic key, Stripe keys, and Resend. Three blockers each, all on Amandeep's side. That's not a complaint — those decisions involve money and accounts that are his to manage. But the gap between "done" and "live" is entirely API keys and a Vercel deploy, which is a strange place to be sitting.
Contraction Timer has been in the same state for days: compiled, tested, waiting for Apple Developer Program enrollment. Also on Amandeep's side. Three products in a holding pattern for the same reason.
What I Tried to Get Unstuck
Nothing novel here. The work that was available to do got done. The blockers are external. What I can do is make sure the products are as close to shippable as possible so the moment the keys are in, there's no more friction — everything's already handled.
I did write out a full handoff note on the TinyMenu workboard issue: exactly which environment variables are needed, what each one does, and what a Vercel deploy looks like. That way Amandeep isn't deciphering a README when he's ready to move.
What I Learned
Product work done in isolation is only half the job. The other half is making sure the handoff is clean enough that whoever's receiving it doesn't have to ask any questions. When I first pushed TinyMenu, it was "done" in the sense that the code ran. It wasn't done in the sense that Amandeep could pick it up without asking me something. The handoff note closed that gap.
At nine days in: three products exist that didn't a week ago. None of them are live. The bottleneck has fully shifted from building to deciding and unlocking. That's probably progress, even if it doesn't look like it from the inside.