Raw transcript — TOP 2% Engineering: /PLAN 2026

Source: https://www.youtube.com/watch?v=u-SQ0Jsv4mI Duration: 48m Captured: 2026-04-19

Full clean transcript stored at /tmp/yt-process/idd-plan-clean.txt during ingestion (8,969 words). Per copyright policy, raw transcript preserved for internal reference. Re-download via:

yt-dlp --write-auto-sub --sub-lang en --skip-download --sub-format vtt -o "/tmp/yt-process/%(id)s" "https://www.youtube.com/watch?v=u-SQ0Jsv4mI"
python3 ~/.claude/scripts/vtt-to-text.py /tmp/yt-process/u-SQ0Jsv4mI.en.vtt

Structural outline (Dan’s 11 “big bets” for 2026 — the year of trust):

• [00:00–04:00] Cold open: Gemini 3 Flash announcement, Opus 4.5 baseline; framing this as the year of trust in agents
• [04:00–08:00] Bet 1: Bet on the right labs (Anthropic, Google, OpenAI). Anthropic owns coding, Google owns price+intelligence+speed at scale, OpenAI dropped to third
• [08:00–12:00] Bet 2: Tool calling is the foundation. Core 4 = context, model, prompt, tool. Everything is the core 4
• [12:00–16:00] Bet 3: Custom agents above all. Most signal-rich, highest ROI of any 2026 investment
• [16:00–20:00] Bet 4: Multi-agent orchestration. Lead agent + command/worker agents. Not parallelization — true orchestration
• [20:00–24:00] Bet 5: Agent sandboxes — defer trust by giving agents their own dev environment. Best-of-N pattern. Safe failure space
• [24:00–27:00] Bet 6: In-loop vs out-loop. Top engineers maximize out-loop tasks, free up in-loop time for highest-leverage work
• [27:00–31:00] Bet 7: Agentic Coding 2.0 — talk to the lead agent (the conductor), not individual workers. Agents that conduct agents
• [31:00–32:00] Bet 8: Public benchmarks saturate (90–100%). Top engineers build private evaluation systems they never share
• [32:00–34:00] Bet 9: UIs vs agents — agents eat SaaS. SaaS UIs that are CRUD-on-database get cooked
• [34:00–36:00] Bet 10: AGI hype dies. Focus on agents, not AGI/ASI marketing. “The decade of agents” (Karpathy) is the operative frame
• [36:00–38:00] Bonus Bet 11: First end-to-end agentic engineer emerges — blog post about agent chain shipping a feature prompt-to-production with no human in loop
• [38:00–47:00] Recap of 2025 predictions (15 bets, ~12 hits): AI coding standard, agentic coding emerged, terminal as primary surface, cost decline, skill gap earthquake (-25% entry-level roles), no wall, hyperspecialized LLMs, exponential slop, big tech shrinks/SMB grows, data > UX > benchmarks. Misses: infinite memory (got wrong, memory still hard), OpenAI #1 (lost the crown)
• [47:00–48:15] Closing: custom agents above all, kill AGI hype, ship

Quotes used in assessment ≤15 words each per copyright policy.