Goal

1. Background and Motivation #

The THINK BIGGER essay (July 2026) identifies the actual opportunity created by current AI capability: not the accelerated production of trivial software, but the ability to wield enormous complexity with excellence.

Frontier AI systems are superhuman in generation, recall, and breadth of knowledge. They remain subhuman in long-horizon autonomy, physical-world grounding, sustained original research, judgment, and taste. As of 2026, humans retain the steering function; AI is the amplifier, not the driver.

The binding constraint on ambitious ventures has therefore changed. It is no longer the size of the engineering organization, but the founder's ability to structure work so that AI leverage compounds: clean decomposition, explicit Contracts, verifiable checkpoints, and disciplined human authority over what matters.

This framework exists to teach and operationalize exactly that structuring.

2. Premises #

The framework rests on four premises, stated precisely:

• P1: Leverage exists. A single individual, or a team of two or three, utilizing frontier AI efficiently and effectively, can now conceive, design, build, and operate Systems of a complexity that until recently required organizations of tens to hundreds of specialists.
• P2: Leverage is conditional. It is realized only when work is decomposed, specified, and verified in forms that AI agents can execute reliably. Unstructured prompting does not scale to complexity. Structure is the amplifier; without it, capability dissipates into drift, rework, and incoherence.
• P3: Structure is domain-independent. The principles governing this structuring are intended to support complex Systems across domains, including SaaS platforms, humanoid robots, and floating cities. Domain-specific practice layers on top of a shared core and adapts it where necessary.
• P4: Human authority is preserved. The human owns purpose, Invariants, taste, risk acceptance, and the definition of done. AI operates with bounded autonomy inside explicit Contracts and verification Gates. Nothing in the framework may transfer these authorities to agents.

3. Objective #

The Framework for Keeping Complex Work Coherent Across Time enables an individual or small team to:

1. convert an ambitious vision into a stable, agent-executable project structure;
2. decompose that structure into Components with explicit Contracts and testable Acceptance Criteria;
3. direct AI agents through research, design, implementation, and verification loops at a level of autonomy bounded by explicit risk, verification, and escalation controls;
4. maintain coherence, quality, and human authority over long time horizons and across personnel and tooling changes.

The structure in items 1 and 2, and a working agent execution loop under item 3, must be reachable within hours to a few days of first contact with the framework. All four must hold regardless of whether the target artifact is a software System, a robotic platform, or a physical megastructure.

4. Design Requirements #

Every requirement below constrains the content of the guidelines.

DR1: Domain-generality Core principles and the Normative Structural Specification are intended to support complex engineered Systems across domains. Software-specific assumptions must not define the shared core.

DR2: Concrete instantiation via domain playbooks The domain-general core is grounded through dedicated domain playbooks covering at least two worked tracks: software (SaaS: low capex, fast iteration, digital verification) and heavy industry / high capex / hardware (floating city: capex staging, physical verification, certification, long lead times). The playbooks demonstrate how the shared structural principles map to different verification regimes, risk profiles, and capital structures, keeping the normative standards lean and strictly domain-agnostic.

DR3: Fast, unambiguous adoption A team using the written framework alone must be able to set up its Project with a functioning, practical agent execution workflow without further external instruction, such that it can efficiently start work on its complex, long-horizon Project.

DR4: Methodological depth Techniques for working with AI agents (spec-driven delegation, context engineering, verification loops, autoresearch loops, parallel agent orchestration, failure-mode recovery, and related methods) are specified at the level of applicable procedure, not merely named.

DR5: Human-in-the-loop by construction Every mechanism states where human judgment is authoritative and where agents operate with bounded autonomy.

DR6: Verification-first No unit of work is considered done without checkable Acceptance Criteria. Verification Artifacts are first-class outputs of the process.

DR7: Self-consistency The framework applies to itself what it prescribes: Stability Class layering (enduring principles vs. volatile tooling), Component structure, and lintable completeness via Normative Checks.

5. Human-AI Operating Model #

The framework codifies a division of labor. This division is normative, not descriptive:

• Human-owned, non-delegable: problem selection, purpose, taste, Invariants and safety boundaries, Acceptance Criteria, risk acceptance, resource and capital allocation, escalation decisions, and any commitment with external consequences (regulatory, financial, contractual, etc.).
• AI-executed, bounded: research and surveying, generation of options and drafts, drafting of Specifications and Designs, implementation within contracted scope, generation of tests and Verification Artifacts, iteration within defined loops, status reporting, etc.
• Shared (AI drafts, human decides): decomposition, interface design, trade-off analysis, prioritization, etc.

The framework's job is to make this division operational: bounded agent autonomy, explicit Contracts, deterministic checkpoints, and explicit escalation rules.

6. Framework Contents #

The framework consists of:

1. Governing Principles: Component-Oriented Decomposition (COD); Layered System Architecture (LSA); Integrated Modularity First (IMF); Specifications as Contracts (SaC); Verification-First Engineering (VFE).
2. Normative Structural Specification: the structural requirements that apply the Governing Principles to complex engineering Projects.
3. Domain playbooks, currently in development:

• Software track: fast iteration, digital verification, deployment, and reliability engineering.
• Heavy industry / hardware track: physical verification, certification and regulatory interfaces, supply chain and procurement, and safety case management.