Contrarian-Depth Essay — produce one Dwarkesh-Patel-style 2,500-5,000 word piece grounded in Product.ai proprietary data

Why we want this done

Generic AI content failed empirically — Digital Trends -97% Google traffic loss, affiliate sites -71% post-March 2026 update, tech publishers -50-60%. Substrate-anchored content compounded — proprietary data + named-expert quotes + working demonstrations produce 4.4-23x conversion uplift on AI-cited content. Top 10 domains capture 46% of ChatGPT citations within a topic; the leverage play is one anchor artifact dominating the niche. The Phase 3 Content briefing names this as the deferred-multi-modal alternative: 4 contrarian-depth essays per year is the substitution for pipeline buildout. Product.ai sits on substantial proprietary substrate — ARC verification benchmarks, Axiomatic Intelligence physics, agent-commerce telemetry, Wisdom synthesis across 11K+ axioms. None of it has been turned into a public artifact yet. The candidate produces the first one — proves they can operate at Era 3-4 (substrate + recommendation observability), demonstrates voice and editorial judgment under brand scrutiny, and ships an artifact that becomes a domain-authority compound for Product.ai going forward.

Scope

1. Day 1-3 — survey Product.ai proprietary substrate (axiom corpus, ARC runs, Wisdom meta-patterns, telemetry the candidate negotiates access to)
2. Day 4-5 — pick the thesis. The thesis must be contrarian (a position not currently represented in mainstream AI commerce discourse), proprietary (anchored in data only Product.ai has), and falsifiable (the reader can disagree on substance, not just preference)
3. Day 6-9 — draft. 2,500-5,000 words. Primary-source data. Named-expert quotes if the candidate can secure them. Working demonstrations or screen-captures where applicable
4. Day 10 — voice edit pass. Manual. Non-skippable. The piece must read as a specific human author, not as an AI-detectable mode draft
5. Day 11-12 — distribution prep — LinkedIn intro draft (in Michael's voice with /michael-voice if Michael is co-publisher), Twitter/X thread, partner-syndication outreach drafts, internal team review
6. Day 13-14 — publish. Earned distribution. Track inbound for at least 7 days post-publication

What success looks like

• One essay published, 2,500-5,000 words
• The thesis is genuinely contrarian — not "things are changing" but "X is structurally wrong about Y because Z, and here is the proprietary data"
• The piece contains at least three pieces of proprietary substrate (data points, axioms, demonstrations) that no competitor can replicate
• The voice edit pass is visible in the prose — AI-mode patterns systematically removed
• Earned distribution lands: Twitter/X traction (5K+ impressions or 50+ thoughtful replies as a baseline; the candidate argues for what "good" looks like), at least one syndication or unprompted reshare from a named source, inbound from at least one new commerce or AI-engineering interlocutor
• Voice Brain calibration is documented — what voice choices were made, why, what was rejected

References

Constraints

• Claude Code or Co-work as primary substrate
• Manual voice edit pass is non-skippable; AI-detectable surface patterns must be removed by hand
• Proprietary substrate must be authentically proprietary — no fabricated data, no recycled public benchmarks dressed up as new
• The piece must be opinionated and falsifiable — readers should be able to disagree on substance, not just dislike
• 2,500-5,000 words is the cap; expansion is a fail
• IP separation: brand surface and proprietary substrate access negotiated explicitly with engineering ownership; no aios-methods access required
• Voice Brain calibration is the load-bearing voice control mechanism — read it, apply it, document the calibration choices