Authority shifts toward harness authorship and outcomes

Claim

Organizational authority increasingly derives from defining effective harnesses and delivering outcomes, not positional hierarchy or execution volume.

Rationale

Harnesses encode bets, constraints, and governance. Those who define them shape reality.

Supporting Signals

• Reputation-based authorship
• Outcome-driven legitimacy

Challenges

• Risk of informal power concentration
• Need for arbitration mechanisms

Evidence

• Spec-driven development: Unpacking one of 2025's key new AI-assisted engineering practices — Thoughtworks analysis of spec-driven development. Web page — requires manual capture (403 on automated fetch).
• Understanding Spec-Driven-Development: Kiro, spec-kit, and Tessl — Martin Fowler's exploration of spec-driven tools. Web page — requires manual capture (403 on automated fetch).
• AI writes code faster. Your job is still to prove it works. — Chrome DevRel lead blog post. Web page — requires manual capture (403 on automated fetch).
• Handbook for New Employees — Primary source document. PDF — requires manual download.
• Valve Corporation: Strategy tipping points and thresholds — Academic study. Journal article — requires manual capture.
• Spec-Driven Development: The Waterfall Strikes Back — Critique of SDD. Web page — requires manual capture (403 on automated fetch).
• Putting Spec Kit Through Its Paces… Radical Idea or Reinvented Waterfall? — Critical evaluation of spec-kit. Web page — requires manual capture (403 on automated fetch).
• Spec-Driven Development: From Code to Contract in the Age of AI — Academic paper formalizing spec-as-source workflows. Connects to Design by Contract. Positions vocabularies, invariants, and constraints as the 'real' system with code as a compiled by-product. First Tier A source for claim 2. LLM coding agents benefit immensely from SDD because structured specs reduce ambiguity.
• Spec-driven development | Technology Radar — Thoughtworks Technology Radar entry for SDD as an emerging technique. Different from the blog post — the Radar carries broader adoption signal. Teams maintain specifications, constitutions, and schemas as primary artifacts.
• These human capabilities complement AI's shortcomings — MIT EPOCH framework. Tasks involving judgment, ethics, creativity, and stakeholder influence are growing relative to purely technical or routine analytical tasks. Authority increasingly linked to ability to frame problems and govern automation. Suggests new mid-skill job categories, not only polarization.
• Building agents with the Claude Agent SDK — Agent SDK guidance emphasizing verification loops, sandboxed testing, and HITL checkpoints. Coding-agent examples rely on automated tests and deterministic checks as main verification mechanism, with human review for broader system goals.
• Mapping Global Dynamics of Benchmark Creation and Saturation in Artificial Intelligence — Nature Communications. Empirical analysis of 3,765 AI benchmarks. Benchmarks don't just measure progress — they steer it. Direct analog to harness authorship: whoever defines the evaluation criteria shapes what 'winning' looks like.
• 'Improving ratings': audit in the British University system — Classic essay behind Goodhart/Strathern's Law: 'When a measure becomes a target, it ceases to be a good measure.' Governance backbone for claims about evaluation harnesses and authority. Auditing produces its own logic and reshapes what it measures.
• Meta-Harness: End-to-End Optimization of Model Harnesses — Validates harness authorship as high-leverage (optimized >> hand-designed), but also shows harness design itself is automatable via agentic proposer. Authority may recurse: from harness authorship to meta-harness specification.
• AI Spillover is Different — Analysis of 460M Revelio records across 16k firms (Cobb-Douglas) finds AI spillovers accrue to flat and lean firms, not hierarchical ones. Macro, indirect support: advantage migrates away from rigid hierarchy, though the link to authority-sourcing is indirect.
• Acceleration Whiplash: AI's Effect on Software Delivery — Telemetry across 22k developers: median review time 5X, incidents per PR 3X, +66.2% epics alongside +861% churn. Execution volume rises while verification becomes the binding constraint (31.3% more PRs merging unreviewed), evidencing cost migration to review and authority concentrating where verification is defined.
• 2026 Agentic Coding Trends — Reports developers use AI in roughly 60% of work but fully delegate only 0-20% of tasks, and describes the shift from implementer to orchestrator. Acts as a brake signal (full delegation is self-capping) and a role-shift signal. Vendor forecast with no disclosed method; corroborated via secondary as the primary PDF was unreadable.
• The Megamanager Era — Using Gallup/Gartner data, reports spans widening to 12 reports (Meta 50:1) with the middle gutted as an efficiency move, framed explicitly as NOT upward power consolidation. Challenging: positional hierarchy persists or intensifies rather than dissolving into harness authority.
• AI Orchestrator Roles and Restructuring Signals — Cites SEC filings (Block 8-K $450-500M restructuring; Meta) and the emergence of AI orchestrator roles where metrics shift to business outcomes. Contextual support that legitimacy is migrating toward outcome ownership.
• Hierarchy Isn't Dying, It's Accelerating — Opinion piece (citing Deloitte/Gartner) arguing AI intensifies rather than flattens hierarchy. Steel-manned counter to the authority-shift claim; non-decisive (Tier D).
• 2026 Global AI Jobs Barometer — Analysis of over 1B job ads across 27 countries. Shows a two-track market: a 62% AI-skills wage premium, professionalised judgement-tier roles growing roughly 2x faster, and entry-level roles most exposed to AI now 7x more likely to require traditionally senior-level skills (seniorised entry roles +35% since 2019 vs -10%). A compensation/skills proxy, not a direct authority measure.

Evaluation Criteria

• Who defines constraints vs who executes
• How conflicts are resolved