METHODOLOGY

RAZEM

Partnership, not servitude. A protocol for humans and AI to produce work neither could alone.

THE PROBLEM

Most AI adoption is backwards

Companies teach employees to write better prompts. They should be teaching them to build better partnerships.

Research from Harvard Business School (2025) identifies three modes of human-AI collaboration on bounded tasks: Centaurs who strategically divide tasks, Cyborgs who fuse workflows, and Self-Automators who delegate everything. Centaurs achieve the highest accuracy. Self-Automators perform worst. Note: this research studies single-task performance, not sustained partnerships — RAZEM extends these findings to ongoing collaboration, which the original research does not address.

The difference isn't the AI. It's how the human structures the relationship.

Source: Dell'Acqua et al. (2025), "Cyborgs, Centaurs, and Self-Automators." HBS Working Paper 26-036. Study of 244 BCG consultants.

Key research findings
  • AI users outperform non-users by 43% on tasks within AI's capability frontier
  • AI users perform 19% worse on tasks outside AI's frontier (over-reliance risk)
  • Centaur mode (human directs, AI executes subtasks) achieves highest accuracy
  • AI collaboration may reduce intrinsic motivation (Nature, 2025)

How RAZEM works

1. Context Architecture

Define shared memory, goals, constraints, and decision protocols before any task begins. The partnership needs a charter, just like a human team does.

2. Role Clarity

Specify what the human brings (judgment, values, domain expertise) and what the AI contributes (pattern recognition, research speed, execution). Overlap is deliberate, not accidental.

3. Continuity Protocol

Structured handoffs between sessions preserve context across conversations. Decision logs, memory systems, and checkpoint documents maintain the relationship over time.

4. Verification Gates

Built-in bias detection, fact-checking personas, and red-team protocols prevent the partnership from reinforcing blind spots. The AI challenges the human. The human challenges the AI.

EVIDENCE BASE

What the research says

FindingSource
Three collaboration modes: Cyborg, Centaur, Self-AutomatorDell'Acqua et al. (2025), HBS Working Paper 26-036
Co-intelligence framework: AI as co-worker, not toolMollick (2024), Co-Intelligence, Wharton/Penguin
Trust calibration in sustained human-AI collaborationCHAI-T Framework, ScienceDirect (2025)
Memory power asymmetry in human-AI relationshipsDorri & Zwick (2025), arXiv:2512.06616
AI collaboration undermines intrinsic motivationNature Scientific Reports (2025)
What RAZEM adds

Existing research is strong on single-session task performance. No framework addresses sustained cross-session partnership: how to maintain context, build trust, manage delegation, and prevent over-reliance over months of collaboration.

RAZEM fills this gap with operational protocols tested through 18 months of daily use.

WHO IT'S FOR
Teams adopting AI workflows

Move beyond "prompt training" to structured partnership protocols that improve output quality and prevent over-reliance.

Leaders managing AI transformation

Framework for deciding which collaboration mode (centaur vs. cyborg) fits which task, and how to train teams accordingly.

FALSIFIABLE PREDICTION

How we'll know if it works

Teams that adopt the full RAZEM protocol for AI-assisted consulting work will report at least 20% higher self-assessed output quality compared to their pre-RAZEM baseline, measured via post-engagement survey within 90 days of adoption.

Falsification: If fewer than 50% of adopting teams report a 20%+ quality improvement within 90 days, the hypothesis is falsified and RAZEM's positioning must be revised. This prediction was published February 2026 before any controlled measurement.

CURRENT STATUS

RAZEM is operational and available as a module within AI consulting engagements.

Working methodology. Used in daily professional practice since mid-2025. Not yet validated through controlled client outcome studies. February 2026.

Discuss for Your Team