If I started tomorrow...

Not a framework per vertical or stage. A method I run faster every year, because a good part of it is now done by agents I build myself.

  1. + before day 1

    Use the product like a stranger

    I sign up, go through onboarding, hit the paywall and read the public site like a new user would. A couple of Claude Code routines crawl the site, pull the SEO picture and map the funnel from the outside.

    I show up on day one with opinions I can back with data, not with a list of questions.

    + a list of first frictions on day 1, not in month 1

  2. ~ week 1

    Get into the data and the code

    Access to the analytics, the repo and the backlog. If the funnel isn't instrumented, that's the first fix, because without reliable events every experiment is an opinion.

    I read the code behind the funnel too. Knowing what's cheap to change and what isn't changes what I prioritise.

    + the funnel map with its three biggest leaks

    How I set up event tracking for experiments
  3. + day 30

    Put agents to read the system

    I set up the first agents: Claude Code skills that turn support tickets, product data and session recordings into prioritised hypotheses. They're triggered by webhooks from Jira or ClickUp, so they run inside the team's workflow, not on my laptop.

    And the first experiment is already live. Small, but real.

    + one experiment in production and a ranked hypothesis queue

  4. ~ day 60

    Build the experiment engine

    Instead of a PRD per idea, I prototype the variants myself and ship the clearest UI changes with engineering review. Agents take the repetitive part: drafting variants, checking the tracking, writing the readout.

    The metrics that matter here are how many experiments the team runs a week and how long it takes to go from hypothesis to live test.

    + more experiments a week, less time from hypothesis to live test

    Product components I shipped with my own code
  5. + day 90

    Turn what worked into a system

    The loops that moved numbers become roadmap commitments, not loose ideas. The agents and routines get documented and handed over, so they keep running without me.

    Every decision is written up in the same format as my case studies (context, hypothesis, decision, result, what I'd do differently). The habit I leave behind: hypothesis → prototype → decision → result, not meeting → meeting → meeting.

    + a growth loop on the roadmap and agents the team owns

    A growth loop at scale: 200K+ monthly organic visits