The Agentic Harness War

2024-05-13 · Source: The Business Engineer · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation, Robotics & Autonomous Systems · Depth: Expert, extended

Summary

An analysis of recent usage data from OpenAI's "The Shift to Agentic AI: Evidence from Codex" and Anthropic's Economic Index reveals a critical shift: the "agentic harness," not the underlying model, now governs AI autonomy and captures value. Both labs' measurement systems broke, indicating a phase change from conversational AI to delegated workflows. Adoption is at an "early majority" for organizations, with 17.3% of users generating 63.3% of agentic output, while individuals are at "stage zero" (under 1% users, 16.5% output). OpenAI's internal usage, at 98% users and 99.8% output, previews near-saturation. This transition is organizationally constrained, not technologically. OpenAI pursues a "consolidation" strategy with one universal harness (Codex), while Anthropic favors "proliferation" with embedded agents (Claude Code, Cowork). Value is migrating from models to the harness, with compute as an unconditional winner and human judgment shifting to oversight. A hidden risk is that consolidation relies on model diversity, which cheap training methods may undermine, leading to a "barbell" outcome where both strategies win in different domains.

Key takeaway

For AI Product Managers evaluating agentic AI adoption, you must prioritize the "harness" layer over raw model capabilities, as it dictates autonomy and value capture. Measure adoption by output, not user seats, to accurately gauge progress, recognizing that organizational friction is the primary bottleneck. Strategically, decide whether to consolidate workflows into a universal harness or proliferate embedded agents, understanding that the former carries a hidden dependency on model diversity.

Key insights

AI's core value and autonomy are shifting from models to the "agentic harness," driving a transition from conversational to delegated workflows.

Principles

• Value migrates model → harness → skill marketplace.
• Read adoption stage by output, not user seats.
• Organizational friction, not tech, limits agentic adoption.

In practice

• Codify workflows into reusable SKILL.md infrastructure.
• Manage multiple agents concurrently for intensive work.
• Focus human effort on delegation and oversight.

Topics

• Agentic AI
• AI Adoption Metrics
• AI Value Migration
• AI Harnesses
• Workflow Automation
• Model Diversity

Best for: Investor, CTO, VP of Engineering/Data, Director of AI/ML, AI Product Manager, AI Scientist

Related on AIssential

• The Week the AI Story Shifted — AI's narrative is shifting from job displacement fears to a focus on infrastructure, enterprise integration, and new job creation.
• The following analysis details the twenty-five primary frictions governing the success of AI scaling, followed by an examination of the systemic consequences of this institutional impasse. — Institutional lag, not technical capability, is the primary barrier to widespread AI adoption and value realization.
• Rethinking organizational design in the age of agentic AI — Organizations must undergo "agentic business transformation" by redesigning tech, workforce, and metrics to fully integrate AI agents.
• My Life in the Harness — The edge in AI work shifts from model operation to human framing and system orchestration.
• AI Eats the World? A Reality Check with Benedict Evans — AI's rapid coding adoption highlights market disequilibrium and unresolved questions about long-term value capture.
• The Harnessing Players Map of AI — Enterprise AI value hinges on control infrastructure, not just model capability.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Business Engineer.