Unofficial. A job application for Pika’s Product Manager role, built from public pages. Not affiliated with Pika.

How I built this.

I ran this with Claude as my lead agent. AI agents did the research and built the pages. Here is the workflow, what was checked, and where the AI was wrong or thin.

The workflow

MeSet the briefPitch the role with a plan and a working prototype. Read Pika the way one of its creators would, before pitching anything.
AIRead everything publicTwo research agents: one on Pika, one on its rivals. Every page in Pika’s two public sitemaps, 112 in all, saved and searched.
AICheck each findingEvery claim about Pika was checked against the saved page text. Thirteen suspected disagreements became ten.
AIBuildThe plan, the prototype and a private findings report.
MeRead it and stand behind itI read every page of this site before sending it.

The lead agent was Claude. It ran the research agents, did the checking and wrote the pages. One working day, November 22, 2026.

The rules the agents worked under

  • Every claim carries its source.A page address and where on the page. The page text is saved, so any claim can be checked again.
  • “Not found” is never rewritten as “missing”.The agents had to say where they looked. Only a statement by Pika itself counts as confirmed absent.
  • Read only.No account, no login, no form, no testing of any system. So nothing behind Pika’s login was seen.
  • Separate agents for Pika and for rivals.So what one vendor says about another never stood in for that vendor’s own page.
  • The starting idea was allowed to lose.The lead agent proposed music-led video because I make music videos with generative tools. The research agents were told to say so if Pika already did it well, or if creators were asking for something else.

Where the AI was wrong or thin

  • The starting picture of Pika was out of date.The brief assumed a video tool. The pages show a platform with its own music, scoring, speech and effects models. The plan was rewritten around what already exists.
  • Three suspected disagreements did not survive checking.Each had an innocent explanation further down the same page. They were dropped.
  • No request from a Pika creator was found.The community is on Discord, which cannot be read without joining. The plan says so and starts with listening.
  • The rival list had to change.One rival named in the brief has been discontinued. The agent found three specialists the brief did not name, and they turned out to matter most.
  • The rival rows are thinner than the Pika row.They were spot-checked, not checked line by line. One vendor’s site would not load as text, so its row rests on a summarised copy.
  • The rivals agent read about twice as many pages as it was asked to.It reported that itself. Several vendors had moved their sites.
  • Third-party roundups were set aside.Several say Pika has no beat sync, but vendors who sell one wrote them. Nothing here rests on them.

Where my judgment came in

Agents can read a hundred pages in a few minutes. Five calls were mine.

  • Pitch with a plan and a working prototype.A résumé says I can do the work. A plan and a page you can click show it.
  • Read the company as its creator would, before pitching anything.A plan that opens with “add beat sync” is a guess. One that opens with what a musician finds on Pika today is a finding.
  • Report what turned up, privately.If I find holes, I bring them up. They went into a report for Pika, not onto this site.
  • Say what I cannot see.The plan has a section on what would change it, because I have no view behind the login or into the usage numbers.
  • Stand behind the output.I read every page of this site before sending it. If something here is wrong, that is mine, not the agent’s.