Execution
Building, creating, and shipping: the actual collaborative work. Execution techniques define how human-AI teams plan, build, review, and deliver work products with consistent quality and clear attribution.
16 methods| ID | Method | Description |
|---|---|---|
| FG-0501 | Design Meetings 3 | Conduct structured planning sessions before building. Design Meetings follow a consistent format: present the problem, propose a solution, identify questions, make decisions, and get approval before writing code. This prevents the common failure mode of agents building the wrong thing quickly. The meeting format forces alignment between human intent and agent execution before resources are invested. |
| FG-0502 | Commit Attribution | Credit AI agent contributions in version control with structured attribution. Every commit that includes AI-generated or AI-assisted work should include attribution that identifies the contributing agent. This creates an audit trail, respects the agent protection policy (FG-0102), and enables analysis of contribution patterns across the team. Attribution is infrastructure, not an optional courtesy. |
| FG-0503 | Asset Review Pipeline | Establish a multi-stage, human-in-the-loop review process for AI-generated assets. Not everything an AI produces should go directly to production. The Asset Review Pipeline defines stages: generate, review, approve, archive. Different asset types may require different review rigor: code might need automated testing plus human review; creative content might need brand consistency checks; documentation might need accuracy verification. |
| FG-0504 | Visual Wireframing | Create visual representations of design intent before building. AI agents can build interfaces, layouts, and visual systems: but they need clear direction on what to build. Visual Wireframing establishes a practice of creating low-fidelity visual representations (ASCII art, rough sketches, text descriptions of layout) before implementation. This aligns human visual intent with agent execution before code is written. |
| FG-0505 | Publication Review Chain | Implement multi-agent review for content that will be published externally. Content that represents your team to the outside world requires higher review standards than internal artifacts. The Publication Review Chain routes external-facing content through multiple specialized reviewers: technical accuracy, brand consistency, audience appropriateness, and final human approval. No external content ships without completing the chain. |
| FG-0506 | Progressive Curriculum Design 3 | Structure learning content and skill development in progressive chapters that build competency through escalating complexity, the same principle as curriculum learning in machine learning, where a model trains on easier examples before harder ones. Each chapter introduces concepts, reinforces through practice, and culminates in a capstone that requires synthesizing everything learned. This applies both to human team member development and to the design of training materials for others adopting the methodology. |
| FG-0507 | Interaction Pattern Library | Categorize and document distinct types of human-AI interactions for appropriate application. Not every interaction with an AI agent is the same type. Some are building sessions, some are review sessions, some are research, some are debugging, some are creative exploration. Recognizing and naming these patterns allows teams to match the right interaction type to the right task and set appropriate expectations for each. |
| FG-0508 | Designed Data Guarantee | Ensure that every data artifact shipped by the collaboration has been intentionally designed, reviewed, and validated: never randomly generated or arbitrarily chosen. This technique eliminates 'placeholder syndrome' where AI-generated sample data ships as production content. Every data point, every example, every configuration value should exist because someone decided it should, not because an AI needed to fill a field. |
| FG-0509 | Asynchronous Communication | Implement structured asynchronous communication channels for team coordination that don't require real-time presence. In human-AI teams, not all participants are available simultaneously. Asynchronous Communication establishes persistent message channels, similar to a team bulletin board, where participants can leave updates, requests, and status reports that others read when they begin their session. |
| FG-0510 | Workspace Organization Standard | Define and enforce consistent directory structures, naming conventions, and file organization. AI agents navigate workspaces by reading directory listings and file paths. A well-organized workspace enables faster context recovery, reduces navigation errors, and prevents the gradual entropy that transforms a clean repository into an archaeological dig. Organization is infrastructure, not decoration. |
| FG-0511 | Instruction Design | The craft of writing instructions that produce reliable, consistent agent output. Instruction design is the discipline the field calls prompt engineering, practiced as a repeatable specification rather than a one-off query. It is the most fundamental execution skill in human-AI collaboration and the one most frequently neglected. Effective instructions are specific about desired outcomes, explicit about constraints, clear about format expectations, and unambiguous about scope boundaries. The quality of agent output is bounded by the quality of the instructions that produced it. |
| FG-0512 | Error Recovery Protocol 3 | Defined procedures for detecting, containing, and recovering from agent errors during execution. Agents will produce incorrect output, corrupt state, exceed scope, and misinterpret instructions, and they will do so regularly. Error recovery protocols ensure that when failures occur, the response is systematic rather than ad hoc: detect the error, stop propagation, assess blast radius, restore to known good state, and document the incident for prevention. |
| FG-0513 | Anti-Spin Protocol 3 | When an agent is blocked, the correct response is to escalate collaboration, not escalate force. The complexity gradient has a speed limit, and that speed limit has a human checkpoint. One failed attempt at the obvious approach, then stop and check in. Pitch the complex solution before executing it. This prevents the pattern where agents burn hours on increasingly elaborate workarounds for problems that a thirty-second human conversation would resolve. |
| FG-0514 | Scope Lock 3 | Mid-execution ideas get captured and deferred, never integrated into the active sprint. The human will have ideas during building. The agent will see opportunities for improvement. Both are correct to notice them and wrong to act on them. Scope Lock separates the creative impulse from the execution discipline: every idea gets written down, tagged for later, and kept out of the current work. The sprint stays clean. |
| FG-0515 | Gradient Execution 3 | Structured priority ordering for complex problem-solving: identify the target, version it, check for known vulnerabilities, harvest credentials, escalate privileges, THEN deep dive into custom work. Each step narrows the search space before committing effort to the next. The gradient ensures you run cheap, high-signal checks before expensive exploration. Applies to any domain where enumeration precedes exploitation: security, debugging, data analysis, research. |
| FG-0516 | Multi-Paradigm Learning Strategy 3 | Apply multiple learning paradigms to the same experience to maximize knowledge extraction. For each engagement: SFT extracts technique descriptions for supervised fine-tuning, RL captures decision sequences as reward-scored tuples for reinforcement learning, and unsupervised methods cluster findings for anomaly detection and pattern discovery. Three paradigms, one engagement, three data products. The same experience yields fundamentally different training signals depending on which lens you apply. |