Day with AI: Facebook CAPI Research, Contact Form Fix, and an Architecture Job Test.

What I worked on

Tuesday. One of those days where you jump between completely different worlds — from Facebook API docs through fixing a form to designing a monitoring architecture on GCP.

What I did

Facebook Conversions API — client research

Client needs to connect their CRM with Facebook via server-side tracking. Complete research — authentication, endpoint structure, event matching, deduplication, testing tools. Output: an implementation guide the team can use immediately. From an hour-long call and scattered notes to a structured document in one session.

Contact form — CORS fix

Classic. The website form was sending requests directly to a webhook and the browser was screaming CORS error. Solution: proxy through an API route on Vercel — request goes server→server, no CORS. Trivial fix, but without it the form simply didn't work.

Video pipeline — onboarding skill

New /reel-setup skill for the video pipeline. Interactive onboarding — prerequisite checks, API keys, brand profile, avatar generation, test run. So anyone can start the pipeline without reading documentation. From "read the README" to "run a command and answer questions."

Job test — monitoring system design

Complete architecture design for a monitoring system for an agency. GCP stack, pseudocode for Google Ads collector, trade-off analysis, Excalidraw diagram. From requirements to HTML/PDF output. This was the type of task where AI helps with iteration speed, but you have to make the architectural decisions yourself.

Automated content

LinkedIn post, cold email pipeline run, YouTube digest — routine automations running in the background. No manual intervention, just output verification.

Time: AI vs without AI

Task	Without AI	With AI

Facebook CAPI research + guide	4-6h	~1.5h
CORS fix + API route	1h	~15min
Reel onboarding skill	3-4h	~1h
Monitoring system (test)	6-8h	~2h
Content automation	2h	~15min
Total	~16-21h	~5h

What I learned

• Server-side tracking (CAPI) is the future — cookies are dying and any client who doesn't get this now will pay double for ads with the same results in a year
• CORS errors on webhooks are evergreen — proxy through an API route is a pattern you should have templated, not solve from scratch every time
• An onboarding skill for a pipeline is an investment, not a nice-to-have — every minute spent on setup without a guide is a minute where the user wonders if it's even worth it
• Job tests with AI are a double-edged sword — you iterate 3x faster, but architectural decisions are still yours. AI will help you draw the diagram, not pick the right pattern