CODE WITH CLAUDE LONDON 2026 · DEEP KNOWLEDGE SUMMARY

Claude
Complete Video Knowledge Base

Every talk, workshop, keynote & case study from Anthropic's official Claude channel — with detailed summaries, speaker bios, key topics, and actionable takeaways for each video.

36 videos ~18 hrs content 35+ speakers 15+ companies

36

Total Videos

~18h

Total Runtime

5

Categories

35+

Speakers

May '26

Published

Content Breakdown by Category

Table of Contents

Deep Video Summaries

Each card includes speaker info, detailed summary, key topics, and actionable takeaways. Click to expand.

Key Numbers from the Conference

96%

Spotify engineers using AI-assisted coding

300%

PR throughput increase at Anthropic

600M+

Monthly sessions on Lovable

750+

Developers using Claude at Man Group

100+

PRs/day merged by agents at Delivery Hero

80x

Growth in Claude Code demand

6x

Completion rate increase with Dreaming (Harvey)

$80M

Wix acquisition of Base44

Key Themes Radar — Coverage Across All Videos

Top Takeaways Across All Videos

Start Simple, Add Complexity With Evidence

The consistent theme across keynotes and workshops: single-agent + good prompting beats over-engineered multi-agent systems until evidence shows otherwise. Omni's "blabbotomy" (merging split agents back into one) proves this.

Memory + Dreaming = Self-Improving Agents

Persistent memory + dreaming (batch consolidation) turns throwaway sessions into evolving systems. Harvey saw 6x completion rate improvement. Warp's Buzz agent updates its own skill files via daily PR.

Evals Are the Primary Development Methodology

Rubric-driven, replayable eval systems give quality/cost/latency signals per change. Private eval sets of 20-50 tasks outperform public benchmarks. Cost-per-success beats raw token cost as the key metric.

Enterprise Adoption Is Real & Measurable

Spotify (96% AI adoption, 60% PR increase), Man Group (AI trading signals in production), Delivery Hero (100+ PRs/day merged by agents), Base44 ($80M acquisition with Claude Code as primary enabler).

Verification > Autonomy

Stop hooks, Turing-incomplete DSLs, and governed skill frameworks prove that constraints accelerate agents. Trust comes from architectural design, not just model alignment. Spotify: "firmer guardrails are accelerators."

Context & Caching Are the Hidden Multipliers

Aim for 80-90% cache hit rates. JSON-to-Markdown conversion saves 66% tokens. Prompt caching can cut costs 80%+. The advisor strategy (cheap model for routine, expensive for hard) optimizes cost vs. intelligence.