Act 1 · 6 min

What I Saw Changing

The paradigm shift, told as personal observation.

I bet this would be transformational

Software output exploding — 90M PRs · 1.4B commits · 20M new repos / month on GitHub.

Record acceleration: 90M PRs merged, 1.4B commits, 20M new repos/month on GitHub, 2023-2026
GitHub Octoverse 2026 · trailing-12-month metrics Source: github.blog · An update on GitHub availability

1.1 · The Overwhelm

Three years of alphabet soup. Couldn't keep up part-time.

Sunholo
Multivac
The laboratory

One AI platform on Google Cloud — capture, storage, modelling, activation — where every new model, protocol, and agent framework gets plugged in and tested as it lands. Same data discipline · richer raw material

1.2 · The Rosetta Stone

AI is still finding its abstractions. The mature ones already exist — we built them in digital analytics.

Stage
Digital analytics
 
AI / Agents
Capture
events, dataLayer, gtag // what happened?
prompts, system message, tool defs // what was asked?
Store
warehouse, BigQuery, GA4 export // the lake
vector DB, RAG, knowledge graph // pulled into context
Model
attribution, funnels, segments // inferring intent
the LLM itself // reading intent
Activate
audiences, personalization, bidding // act on it
UX, harness, tool calls // agent acts
Same four stages from Learning Google Analytics · O'Reilly 2022 · new raw material, same discipline

1.3 · Same Discipline, Richer Data

Capture the moment, add context, send it somewhere. Look what's new.

GA4 event gtag.js 
gtag('event', 'view_pricing', {
  page_path: '/pricing',
  user_id: 'u_8421',
  plan_tier: 'team',
});
page user dimension
Langfuse trace @langfuse/sdk 
langfuse.trace({ id: 't_8421', userId: 'u_8421' })
  .generation({
    name: 'pricing_qa',
    model: 'claude-opus-4-6',
    input: 'is this worth it for a small team?',
    output: 'For a 5-person team...',
    usage: { in: 412, out: 186, cost: 0.011 },
    latencyMs: 1820,
  });
prompt response model tokens cost latency
Session→Trace Pageview→Generation Event→Tool Call / Span User ID→Thread ID Custom dim→Metadata
Coming up tonight · Julius Fedorovicius
The Two Pillars of Modern Tagging: Security and Enrichment
"Julius will show you how to harden your tagging tonight. I'm here to ask whether you're capturing the right surface in the first place."

1.4 · The dashboards look familiar too

Same boxes, same charts, same drill-downs — radically richer signal inside.

GA4 · Reports snapshot analytics.google.com
GA4 reports snapshot dashboard
users sessions pageviews events conversions
Langfuse · LLM analytics cloud.langfuse.com
Langfuse LLM analytics dashboard
traces generations tool calls + tokens + cost + latency + model
1 / 1
←→ navigate