Issue #3 — AI Agent Debrief — Week 3
Sent May 1, 2026 to AgentTree Army subscribers.
Subscribe to get the next one →
AgentTree Army — AI Agent Debrief — Week 3 May 01, 2026
Section titled “AgentTree Army — AI Agent Debrief — Week 3 May 01, 2026”🔨 BUILD OF THE WEEK: AI Dev 2026 Sponsor Replicability Audit
I’m at Andrew Ng’s AI Dev 2026 in SF this week. The BYU MBA alumni AI group keeps asking the same question: what’s actually defensible vs. what’s a thin LLM wrapper? So I scored every sponsor on the floor — 56 of them, 55 with full scores — on a 0-100 “can a solo BYU MBA clone this in a weekend with Claude Code” axis. Higher = easier to replicate = the moat is something other than engineering. Built it as a filterable Next.js site so the alumni group can sort by tier, search, see the replication notes. Live at aidev-2026-sponsors.vercel.app.
🛠 TOOL SPOTLIGHT: Claude Code as the audit tool
What made this same-day buildable: Claude Code didn’t just write the site — it scraped the sponsor list off the conference page, ran a per-sponsor research pass, scored each on the rubric, generated the replication note, AND scaffolded the Next.js + Tailwind UI in one session. One git repo, one Vercel project, no glue services. The whole “market research → notes → site → deploy” chain collapsed into one workflow.
Verdict: Ship it. “Is this defensible in 2026?” is now a question you can answer in a day.
📡 INTEL BRIEF: Three things from the floor
Boring is the moat. Average sponsor scored 34/100 on cloneability. Only 6 of 55 (11%) scored 70+. The low-scorers — the actually-hard ones — all had something dull underneath: a real data network, a hardware integration, a regulatory wedge, an enterprise contract base. If your defensibility story is “we built a multi-agent system,” you’re in the easy-to-clone bucket.
The agentic-AI startups all looked the same. Same orchestration, same tool-use vocabulary, same “compound system” pitch. The differentiator wasn’t engineering — it was who picked up the phone and called real customers in 2024 before everyone else.
The three most-said phrases on stage — “agentic,” “reasoning model,” “production-ready” — beat anything else by a margin. (I built a buzzword bingo board to track it. Link below.)
🏆 COMMUNITY HIGHLIGHT: AI Buzzword Bingo + the synthetic customer thing
Two side things from this week. (1) AI Buzzword Bingo — 5×5 board, 24 of the most-overused phrases at AI Dev 2026, free space in the middle, leaderboard. Bring it to your next agent demo. (2) Also stood up a synthetic customer agent fleet — 100-persona panel that drives a real browser session. Still in private testing; more in a future issue.
Standing weekly vibe-coding session continues — Friday 1-4 PM MT. Discord or BYU Provo in person. Reply if you’re showing up.
============================================================ AgentTree Army · Provo, UT 84601 Unsubscribe: https://agenttree.army/newsletter